Research extensions#
These public modules group the methods and connectors introduced by the 2026 research expansion. The modeling guides explain scientific assumptions, measurement boundaries, optional dependencies and validation maturity.
Spacecraft physics and power systems#
Public research physics API: thermal, radiation, spectral, circuits and EPS.
These models are scientific screening tools with explicit approximations. They are not flight-qualified designs or exact reproductions of the cited papers.
- class orbitalpv.physics.BatterySpec(capacity_wh, initial_soc=0.8, minimum_soc=0.1, maximum_soc=0.95, charge_efficiency=0.95, discharge_efficiency=0.95, max_charge_w=inf, max_discharge_w=inf)[source]
Usable model parameters; limits are bus-side W, capacity is nominal Wh.
- class orbitalpv.physics.ComputeTask(energy_wh, release_s, deadline_s, maximum_power_w)[source]
Deferrable compute energy with release/deadline aligned to interval edges.
- class orbitalpv.physics.PVSubstring(photocurrent_a, open_circuit_voltage_v, diode_voltage_v=0.6, series_resistance_ohm=0.0, bypass_voltage_v=0.5, fault='healthy')[source]
Equivalent generating substring and its bypass branch.
diode_voltage_vis n*N_s*k*T/q. Photocurrent is not exact short-circuit current when a nonzero series resistance is supplied. Anopenfault opens the generating branch; a present bypass diode can still carry current. Setbypass_voltage_v=Nonefor an unbypassed open circuit. Ashortfaults the entire substring to zero voltage.
- class orbitalpv.physics.RadiationExposure(transmitted_fluence_cm2, displacement_damage_mev_g, equivalent_fluence_cm2)[source]
Integrated particle exposure for one species and material response.
- class orbitalpv.physics.RadiationResponse(coefficient, reference_fluence_cm2, minimum_retention=0.0)[source]
Fit parameters for R = 1 - C log10(1 + equivalent_fluence / Phi0).
Parameters must come from a named technology/particle/reference-energy calibration; no universal spacecraft or perovskite default is supplied.
- class orbitalpv.physics.TandemCurrent(subcell_current_a_m2, series_current_a_m2, current_mismatch_fraction)[source]
Current-matching result, not a power conversion efficiency estimate.
- class orbitalpv.physics.ThermalSurface(area_m2=1.0, radiating_area_m2=2.0, heat_capacity_j_k=10000.0, solar_absorptivity=0.9, ir_emissivity=0.85, conductance_w_k=0.0, bus_temperature_k=293.15, sink_temperature_k=3.0)[source]
Effective isothermal surface; areas in m² and heat capacity in J/K.
radiating_area_m2includes every emitting face, whilearea_m2is the receiving/projected area associated with the supplied irradiance. Conductance couples the surface to a fixed-temperature spacecraft node.
- orbitalpv.physics.array_iv_curve(strings, voltage_v=None, *, points=501)[source]
Compose series substrings and parallel strings on a common voltage grid.
Nonnegative voltage/current represent the generating quadrant. Bypass conduction is a constant voltage-drop approximation. Current is obtained by monotone bisection of each string’s voltage-at-current relation. MPP attributes describe the sampled grid, so refine it for an accurate MPP.
- orbitalpv.physics.battery_cycle_damage(soc, *, cycles_to_failure_at_full_depth, depth_exponent=1.5)[source]
Return cumulative Miner damage from a supplied cycle-depth calibration.
One full-depth cycle contributes
1 / cycles_to_failure_at_full_depth. The result is a damage index, not a remaining-capacity percentage or a predicted failure date. Calendar aging, temperature and rate effects are not modeled. The calibration is intentionally required from the caller.
- orbitalpv.physics.dc_cable_delivery(*, source_voltage_v, load_w, length_m, conductor_area_mm2, resistivity_ohm_m, density_kg_m3, minimum_voltage_fraction=0.9)[source]
Return the high-voltage DC equilibrium and conductor-only mass.
Length is one-way; two identical conductors form the return loop. Resistivity is supplied at operating temperature. The load is power at the receiving terminals, before any downstream converter losses. An ideal stiff source and purely resistive cable are assumed. The high-voltage quadratic root is not proof of dynamic converter stability. No equilibrium is represented by NaN voltage/current/loss, never by zero loss. The double-root maximum-transfer point is reported but fails the strict branch-margin criterion.
Minimum receiving voltage is a fraction of source voltage, in (0.5, 1]. Feasibility here excludes ampacity, insulation, discharge, thermal balance, regulation, connectors, dust, mechanical qualification and source capacity.
- orbitalpv.physics.dispatch_power(time_s, generation_w, demand_w, *, battery, maximum_served_power_w=None)[source]
Dispatch generation and storage across piecewise-constant time intervals.
time_scontains N+1 interval boundaries; power arrays have N entries. Generation first supplies load, then charges storage; stored energy supplies deficits. Surplus is curtailed and unmet load is explicitly reported.maximum_served_power_wcan enforce an externally calculated thermal cap.- Parameters:
time_s (ArrayLike)
generation_w (ArrayLike)
demand_w (ArrayLike)
battery (BatterySpec)
maximum_served_power_w (ArrayLike | None)
- Return type:
- orbitalpv.physics.periodic_load_limit(sunlit_power_w, sunlight_h, eclipse_h, *, battery)[source]
Maximum constant bus load sustainable over repeated identical cycles.
Generation is constant in sunlight and zero in eclipse. Storage is lossless except for the specified charge/discharge efficiencies; the full SOC window is available. Initial SOC is deliberately ignored: this is a periodic-state feasibility bound, not a startup guarantee. No payload/base-load allocation, reserve margin, thermal constraint or degradation is inferred.
Returns the minimum of orbit energy, usable storage and converter bounds. All returned limits are bus-side watts. Infinite converter limits are allowed by BatterySpec. A zero eclipse bypasses storage entirely.
- orbitalpv.physics.pointing_power_envelope(normal_bus_w, load_w, *, reserve_fraction=0.2, offpoint_deg=30.0)[source]
Screen sunlit, fixed-array pointing against a load plus reserve.
normal_bus_wis a nonempty scalar or 1-D sequence of available bus power at normal incidence, in W, after distance and electrical/retention losses. Required power isload_w * (1 + reserve_fraction). The maximum angle between the array normal and Sun direction is acos(required/normal), in degrees. An infeasible normal-incidence state has a NaN angle, not zero.headroom_wis generation atoffpoint_degminus required power.This instantaneous cosine law assumes uniform, unshadowed illumination and constant electrical efficiency. It excludes eclipses, battery support, temperature response, gimbal limits and bus voltage constraints. A reserve fraction is an engineering assumption, not a probabilistic confidence bound.
- orbitalpv.physics.radiation_equivalent_fluence(energy_mev, differential_fluence_cm2_mev, niel_mev_cm2_g, *, reference_niel_mev_cm2_g, shielding_transmission=1.0)[source]
Integrate differential particle fluence against a tabulated NIEL curve.
Energy is MeV, differential fluence is particles/(cm² MeV), and NIEL is MeV cm²/g. The resulting displacement damage dose is MeV/g. Equivalent fluence divides that dose by the explicitly supplied reference NIEL.
shielding_transmissionis a dimensionless, energy-dependent transfer factor obtained externally. Simple attenuation omits energy redistribution, secondaries and angular transport; use a transported spectrum instead when those matter. Do not sum proton and electron equivalent fluences unless their material/reference normalization has been made compatible.- Parameters:
energy_mev (ArrayLike)
differential_fluence_cm2_mev (ArrayLike)
niel_mev_cm2_g (ArrayLike)
reference_niel_mev_cm2_g (float)
shielding_transmission (ArrayLike)
- Return type:
RadiationExposure
- orbitalpv.physics.radiator_feasibility(heat_load_w, *, area_m2, radiator_temperature_k, emissivity=0.85, absorbed_environment_w_m2=0.0, sink_temperature_k=3.0)[source]
Evaluate a steady radiative heat-rejection bound for one effective area.
absorbed_environment_w_m2already includes solar/albedo/IR absorption and view factors. No convection exists in this vacuum model. Required area is infinite if the environment prevents any positive net heat rejection. This does not guarantee chip-to-radiator heat transport or safe junction temperature, nor does it include radiator mass or puncture tolerance.
- orbitalpv.physics.rainflow_cycles(soc)[source]
Count closed/full and residual/half SOC cycles with a stack algorithm.
Consecutive identical samples are removed before extracting reversals. Columns are depth_of_discharge, mean_soc and count (0.5 or 1.0). Residual half cycles are retained; a constant trace therefore returns no cycles.
- Parameters:
soc (ArrayLike)
- Return type:
- orbitalpv.physics.retained_power_fraction(equivalent_fluence_cm2, response)[source]
Return a bounded empirical remaining factor (zero exposure gives one).
This monotone model does not describe annealing or perovskite self-healing. Extrapolation beyond the calibration fluence range requires validation.
- Parameters:
equivalent_fluence_cm2 (ArrayLike)
response (RadiationResponse)
- Return type:
- orbitalpv.physics.schedule_compute_workload(time_s, generation_w, base_load_w, *, battery, tasks, compute_power_limit_w, thermal_power_limit_w=None)[source]
Schedule tasks by earliest deadline, preferring current solar surplus.
Each interval serves mandatory near-deadline work and otherwise uses solar surplus. Storage supports mandatory work subject to SOC/power limits. The deadline lookahead includes known global and interval thermal power caps. Per-task future capacity is an upper bound because tasks share resources; the heuristic still need not find every feasible multi-task schedule. The thermal cap applies to total served bus load; battery/converter heat and radiator transients need a separate thermal node. This deterministic heuristic does not guarantee the optimal feasible schedule.
Returns dispatch telemetry and a per-task completion/unserved-energy table.
- orbitalpv.physics.simulate_panel_thermal(time_s, direct_solar_w_m2, *, surface=None, albedo_w_m2=0.0, earth_ir_w_m2=0.0, internal_heat_w=0.0, pv_efficiency=0.25, efficiency_temperature_coefficient_per_k=-0.002, reference_temperature_k=298.15, initial_temperature_k=293.15, max_step_s=30.0)[source]
Integrate a thermal/PV energy balance with auditable interval energies.
For interval
(time_s[i-1], time_s[i]], forcing at indexiis piecewise constant. Efficiency is evaluated at each substep’s starting temperature and clipped to [0, absorptivity]; radiative and conductive heat rejection are solved implicitly at its ending temperature. The temperature coefficient is a relative coefficient, e.g. -0.002 means -0.2 %/K.Albedo uses the same scalar PV efficiency as direct sunlight; a spectral model is needed where spectral differences matter. PV electricity is removed from this surface’s thermal balance and is not destroyed: downstream payload/electronics heat belongs in the corresponding spacecraft node.
- Parameters:
time_s (ArrayLike)
direct_solar_w_m2 (ArrayLike)
surface (ThermalSurface | None)
albedo_w_m2 (ArrayLike)
earth_ir_w_m2 (ArrayLike)
internal_heat_w (ArrayLike)
pv_efficiency (ArrayLike)
efficiency_temperature_coefficient_per_k (float)
reference_temperature_k (float)
initial_temperature_k (float)
max_step_s (float)
- Return type:
- orbitalpv.physics.spectral_photocurrent_density(wavelength_nm, spectral_irradiance_w_m2_nm, eqe)[source]
Integrate EQE times photon flux, returning photocurrent density in A/m².
Wavelength is strictly increasing in nm and spectral irradiance is W/m²/nm. The final axis of irradiance/EQE must be wavelength; leading dimensions may be broadcast, allowing several spectra or devices. No interpolation or extrapolation is hidden inside this calculation. EQE must be in [0, 1].
- Parameters:
wavelength_nm (ArrayLike)
spectral_irradiance_w_m2_nm (ArrayLike)
eqe (ArrayLike)
- Return type:
- orbitalpv.physics.tandem_current_density(wavelength_nm, spectral_irradiance_w_m2_nm, subcell_eqe)[source]
Calculate the limiting current of series-connected, equal-area subcells.
Each EQE row is the external response of that subcell in the finished stack, referenced to the same incident spectrum. Their wavelength-wise sum cannot exceed one: using standalone EQEs without optical filtering would double count photons. Mismatch is a current-utilization metric, not a measured energy loss. Radiative coupling and carrier multiplication are outside this screening model.
- Parameters:
wavelength_nm (ArrayLike)
spectral_irradiance_w_m2_nm (ArrayLike)
subcell_eqe (ArrayLike)
- Return type:
TandemCurrent
Orbital compute and deployment#
Screen solar-powered orbital compute against power, heat and launch limits.
This composes OrbitalPV’s circular-shadow, periodic-load and battery-dispatch models. Inputs are design assumptions, not a flight-qualified spacecraft model. The radiator has a prescribed temperature and total emitting area, not a CAD footprint. A fixed heat allowance reserves capacity for EPS/storage losses.
- class orbitalpv.orbital_compute.OrbitalComputeSpec(altitude_m, beta_angle_deg, pv_area_m2, efficiency_am0, packing_fraction, pv_temperature_c, temperature_coefficient_per_k, array_derate, eps_efficiency, retained_power_fraction, irradiance_w_m2, platform_load_w, average_compute_w, peak_compute_w, radiator_emitting_area_m2, radiator_temperature_k, radiator_emissivity, absorbed_environment_w_m2, thermal_allowance_w, battery)[source]#
Explicit SI inputs; AM0 efficiency must be independently characterized.
pv_area_m2is the deployed array footprint. Packing excludes inactive area; derate covers mismatch/wiring, and EPS efficiency is separate. Retention is an imposed sensitivity, not a prediction of radiation damage.thermal_allowance_wmust cover converter/storage heat on this radiator; PV blanket absorption is assumed to radiate through its separate surface.- Parameters:
altitude_m (float)
beta_angle_deg (float)
pv_area_m2 (float)
efficiency_am0 (float)
packing_fraction (float)
pv_temperature_c (float)
temperature_coefficient_per_k (float)
array_derate (float)
eps_efficiency (float)
retained_power_fraction (float)
irradiance_w_m2 (float)
platform_load_w (float)
average_compute_w (float)
peak_compute_w (float)
radiator_emitting_area_m2 (float)
radiator_temperature_k (float)
radiator_emissivity (float)
absorbed_environment_w_m2 (float)
thermal_allowance_w (float)
battery (BatterySpec)
- orbitalpv.orbital_compute.screen_orbital_compute(spec)[source]#
Calculate repeatable continuous compute and necessary sizing conditions.
The electrical bound includes orbit energy, SOC window and battery power limits. Heat rejection assumes all served electrical load becomes heat. Neither a nameplate compute rating nor an initially full battery creates sustainable energy. Area/storage requirements alone are not sufficient: charge/discharge ratings and thermal capacity must also pass.
- Parameters:
spec (OrbitalComputeSpec)
- Return type:
- orbitalpv.orbital_compute.simulate_orbital_compute(spec, *, cycles=40, intervals_per_orbit=120, peak_fraction=0.2, dispatch_policy='greedy')[source]#
Dispatch orbit-repeating bursts with exactly the requested mean load.
Each cycle starts in sunlight, followed by eclipse. The peak occupies the last
peak_fractionof each orbit (a deliberately adverse alignment). Boundaries include shadow entry and burst start exactly, avoiding timestep aliasing. Platform demand has priority within each interval, with no lookahead reserve for future platform demand; overload can cause outages. A fixed radiator cap excludes thermal inertia, and the full trace reports SOC drift and unserved energy. Initial SOC is set to the reserve floor to avoid gifting stored energy.continuous_capadmits at most the repeatable continuous compute bound at every interval. This conservative admission policy protects platform demand under this fixed repeating geometry, provided the platform alone is feasible. It can reject useful sunlit bursts; it is not an optimal scheduler or protection against uncertain generation.greedypreserves the original unconstrained demand. Requested, admitted and rejected compute are reported separately; rejected work is not automatically rescheduled. Dispatch conservation columns refer to admitted electrical demand.
- orbitalpv.orbital_compute.deployment_screen(*, manufacturing_gw_per_year, factory_utilization, manufacturing_yield, space_allocation_fraction, array_area_m2, packing_fraction, efficiency_stc, satellite_mass_kg, payload_to_target_orbit_kg, launches_per_year, continuous_compute_kw, compute_utilization=1.0)[source]#
Annual deployment ceiling on an explicitly assumed STC factory basis.
Annual GW nameplate production is NOT delivered orbital GW or compute GW. Array watts here use 1000 W/m² STC and a separately supplied efficiency. Full spacecraft mass includes PV, radiator, storage and structure. Launch throughput is mass-only: fairing volume, deployment reliability, networking, approvals, satellite factory throughput and replacements are excluded.
- Parameters:
manufacturing_gw_per_year (float)
factory_utilization (float)
manufacturing_yield (float)
space_allocation_fraction (float)
array_area_m2 (float)
packing_fraction (float)
efficiency_stc (float)
satellite_mass_kg (float)
payload_to_target_orbit_kg (float)
launches_per_year (int)
continuous_compute_kw (float)
compute_utilization (float)
- Return type:
Flight telemetry and uncertainty#
Measured flight-energy budgets, causal forecast baselines and uncertainty.
- class orbitalpv.flight.BayesianLinearCalibration(model, coefficient_covariance, observation_noise_std, prior_std)[source]
Normal likelihood + zero-mean normal coefficient prior, known noise.
- Parameters:
- predict(features, *, timestamps=None, coverage=0.95, include_observation_noise=True)[source]
Return posterior mean and Gaussian credible/predictive intervals.
Coverage is a model-based probability, not verified frequentist coverage. Future model discrepancy and parameter drift are not included.
- class orbitalpv.flight.BlockConformalVI(quantile, scales, coverage, block_size, calibration_blocks, calibration_samples, discarded_tail_samples, trained_through, calibrated_through, horizons)[source]
Maximum normalized residual over channels, horizons and each time block.
- Parameters:
- predict(predicted, *, origin_times)[source]
Apply frozen intervals after the last calibration target timestamp.
- Return type:
VIPredictionIntervals
- class orbitalpv.flight.LaggedVIData(features, targets, origin_values, origin_times, target_end_times, feature_names, lags, horizons, cadence_s, source_windows)[source]
Features available at each origin and future measured V/I targets.
Array axes are sample, horizon, then (voltage, current). Horizons denote source-sample steps; actual origin/end timestamps are retained. No gaps are filled and any window containing a large cadence departure is excluded.
- class orbitalpv.flight.LinearCalibration(coefficients, feature_mean, feature_scale, trained_from, trained_through, purpose, evidence, samples)[source]
A ridge/least-squares fit with training-only feature normalization.
- Parameters:
- predict(features, *, timestamps=None, held_out=True)[source]
Predict; supplied held-out timestamps must follow all fit labels.
Omit timestamps only for algebraic/in-sample model inspection. A caller doing an evaluation should always supply them; no function can discover hidden future information in caller-constructed feature columns.
- class orbitalpv.flight.LinearVIForecaster(model, horizons, feature_names)[source]
Multi-output ridge baseline trained only on preceding V/I labels.
- class orbitalpv.flight.MeasuredEnergyBudget(intervals, summary)[source]
Energy integrated over common valid observation intervals only.
- class orbitalpv.flight.SatelliteCOTSConnector(*, timeout_s=30.0, user_agent='space-pv-analytics/0.2')[source]
Download the authors’ original files, pinned by commit and SHA-256.
- class orbitalpv.flight.VIPredictionIntervals(prediction, lower, upper, power_prediction_w, power_lower_w, power_upper_w, coverage)[source]
Joint V/I rectangles and sign-safe extrema of their products.
- Parameters:
- orbitalpv.flight.calibrate_block_conformal_vi(observed, predicted, *, origin_times, target_end_times, trained_through, scales, coverage=0.9, block_size=20)[source]
Calibrate joint multi-horizon V/I envelopes on a later time partition.
scalesis positive (V, A) or (horizon, 2), selected on training data or independently, never tuned on this calibration/evaluation block. Blocks are nonoverlapping consecutive forecast records. The incomplete trailing block is discarded from scores but its labels remain inside the embargo boundary. This row-block method is not a proof of temporal independence.
- orbitalpv.flight.chronological_forecast_split(data, *, training_stop, calibration_stop)[source]
Split by clock time and purge targets that cross either boundary.
Training/calibration labels end strictly before their stop. Subsequent origins start at or after the stop; lagged historical observations remain allowed because they were already observable at each forecast origin.
- Parameters:
data (LaggedVIData)
- Return type:
tuple[LaggedVIData, LaggedVIData, LaggedVIData]
- orbitalpv.flight.conformal_order_statistic(scores, *, coverage)[source]
Finite-sample split-conformal quantile; returns infinity if insufficient.
The order is ceil((n+1)*coverage), without clipping it to n. Clipping would silently advertise a confidence level unsupported by the calibration size.
- orbitalpv.flight.fit_bayesian_linear_calibration(features, target, *, timestamps, purpose, observation_noise_std, prior_std=10.0, evidence='')[source]
Compute an exact Gaussian coefficient posterior with explicit noise/prior.
Prior standard deviation applies to all coefficients, including intercept, in standardized feature coordinates and the target’s units. Choose noise on training data or independent calibration evidence, never evaluation data.
- orbitalpv.flight.fit_linear_calibration(features, target, *, timestamps, purpose, evidence='', ridge=0.0)[source]
Fit a deterministic reference or forecast, declaring its intended role.
timestampsidentify target/label times, including the last forecast horizon. The intercept is unpenalized. Ridge is in standardized-feature coordinates; this is a statistical baseline, not an inferred physical law.
- orbitalpv.flight.fit_linear_vi_forecaster(data, *, ridge=1.0)[source]
Fit a direct multi-horizon, multivariate-lag regression baseline.
- Parameters:
data (LaggedVIData)
ridge (float)
- Return type:
LinearVIForecaster
- orbitalpv.flight.load_bupt1_telemetry(path, *, source='common', source_timezone=None, time_min=None, time_max=None, verify_source=True, chunksize=100000)[source]
Stream a local pinned archive, filtering inclusive/exclusive source times.
Bounds must match the timestamp basis: naive for an unspecified source timezone, or timezone-aware if
source_timezoneis supplied. The full archive is large; supply a time window. No ZIP member is extracted to disk. Setverify_source=Falseonly for independently provided files, which are then labelled unverified rather than claimed as the pinned archive.
- orbitalpv.flight.make_lagged_vi_data(frame, *, lags=(0, 1, 2), horizons=(1, 2, 4), feature_columns=('voltage_v', 'current_a'), cadence_s, cadence_tolerance_s=0.5)[source]
Build causal lag features and paired targets, rejecting discontinuities.
Caller-supplied feature columns must themselves be available at the origin. Telemetry from multiple panels must be separated before calling this method.
- orbitalpv.flight.measured_energy_budget(frame, *, solar_column='solar_bus_power_w', load_column='bus_power_w', battery_column=None, battery_discharge_sign=1, max_gap_s)[source]
Reconcile solar + battery discharge - load using common valid intervals.
A residual is unexplained measurement/boundary mismatch, not an attributed physical loss. Supply battery sign explicitly when source data use charge- positive current. Missing battery telemetry is not assumed to be zero unless the caller chooses
battery_column=None(a solar-versus-load comparison). Multiplexed group rows require a separately justified alignment first.
- orbitalpv.flight.normalize_bupt1_telemetry(raw, *, source='common', source_timezone=None)[source]
Normalize units without interpolating or demultiplexing measurements.
The common archive supplies separate MPPT1/2 and battery1/2 columns. The short-term source cycles through hardware groups; its group-specific power must not be mistaken for simultaneous whole-spacecraft PV power. Duplicate source clocks and signed sensor offsets remain visible. Zero temperature is retained because zero is not documented as a universal missing sentinel.
- orbitalpv.flight.persistence_vi_forecast(data)[source]
Repeat the last observed operating voltage/current at each horizon.
- Parameters:
data (LaggedVIData)
- Return type:
- orbitalpv.flight.seasonal_lag_vi_forecast(data, *, period_steps)[source]
Use a historical seasonal/orbit-lag baseline with explicitly chosen lag.
A sample period is an empirical assumption, not a propagated orbit. Every needed lag (period minus horizon) must be present in
data. No future values are accessed when a horizon exceeds the chosen period: it raises.
Environmental observations and attitude#
Environmental data, provenance-preserving imports and Earth-view geometry.
- class orbitalpv.environment_data.CeresConnector(*, timeout_s=30.0, user_agent='space-pv-analytics/0.2')[source]
Discover exact CERES collections through NASA CMR and download selected files.
Supply the collection short name explicitly: different editions and time aggregations are scientifically different products. Earthdata credentials may be needed by the file host; catalog discovery is public.
- collections(*, short_name_pattern='CER_SYN1deg*', max_results=100)[source]
Discover current CMR short names and separate edition/version IDs.
NASA’s cloud migration changed catalog identifiers. A product title containing
_Edition4Bneed not have that suffix in its CMR short name.
- search(*, short_name, version, start, end, bounding_box=None, page_size=100, max_pages=20)[source]
Return granule metadata and direct-data URLs, failing on truncation.
- class orbitalpv.environment_data.NaifKernelConnector(*, timeout_s=30.0, user_agent='space-pv-analytics/0.2')[source]
Download explicitly selected NASA NAIF SPICE kernels with provenance receipt.
- class orbitalpv.environment_data.PoesMetopConnector(*, timeout_s=30.0, user_agent='space-pv-analytics/0.2')[source]
Download documented NCEI daily SEM-2 Level-1b files.
Instrument availability varies by date. Missing provider files raise the original HTTP error; no fabricated observations or nearby-day substitution.
- class orbitalpv.environment_data.ProjectedEarthFlux(irradiance_w_m2: 'float', valid_view_fraction: 'float', visible_cells: 'int', partial: 'bool', geometric_view_factor: 'float', model: 'str' = 'Lambertian spherical Earth')[source]
- class orbitalpv.environment_data.SpaceTrackConnector(*, timeout_s=30.0, user_agent='space-pv-analytics/0.2')[source]
Authenticated Space-Track GP/GP-history connector.
Credentials should be supplied by environment/secrets management, never committed.
- latest_gp_cached(norad_cat_id, *, cache_path)[source]
Reuse current OMM for two hours; authenticate first on a cache miss.
Shares the same persistent snapshot/provenance contract as CelesTrak. This conservative interval also satisfies Space-Track’s hourly GP limit.
- Parameters:
- Return type:
- gp_history(norad_cat_id, *, start, end, cache_dir, max_records=100000)[source]
Fetch and cache a bounded, complete EPOCH interval in chronological order.
Space-Track requests that GP history be downloaded once and reused from disk. A mandatory cache prevents repeated calls for an identical interval.
startandendare inclusive, timezone-aware element epochs, not publication dates. Authentication is required for a cache miss. Queries over one year should use provider bulk/year files instead.
- class orbitalpv.environment_data.SpenvisBlock(comments: 'tuple[str, ...]', metadata: 'dict[str, dict[str, Any]]', annotations: 'tuple[str, ...]', variables: 'tuple[SpenvisVariable, ...]', data: 'pd.DataFrame', footer: 'str', following_blocks: 'int')[source]
- class orbitalpv.environment_data.SpenvisVariable(name: 'str', units: 'str', dimension: 'int', title: 'str', metadata_names: 'tuple[str, ...]' = ())[source]
- orbitalpv.environment_data.earth_flux_to_panel(spacecraft_position_m, panel_normal, latitude_deg, longitude_deg, outgoing_flux_w_m2, cell_area_m2, *, earth_radius_m=6371008.8, allow_partial=False)[source]
Integrate visible Earth patches onto one side of a plane receiver.
outgoing_flux_w_m2 / piis the assumed patch radiance. Both source emission angle and panel incidence angle are included. Missing flux in the supplied visible patches produces NaN unlessallow_partial=True, in which case the returned sum covers valid cells only, without extrapolation.valid_view_fractiondescribes finite values among the supplied patches; it does not prove that the full visible Earth disk was supplied. Patches must be nonoverlapping. Center-point quadrature requires a sufficiently fine grid; a geometric view factor above 1.01 raises instead of being normalized.
- orbitalpv.environment_data.earth_grid_cell_areas(latitude_edges_deg, longitude_edges_deg, *, earth_radius_m=6371008.8)[source]
Exact spherical grid-cell areas from increasing latitude/longitude edges.
- orbitalpv.environment_data.frame_rotation_matrices(from_frame, to_frame, ets)[source]
Return rotations mapping column vectors from one SPICE frame to another.
CK attitude frames require the relevant CK, FK, SCLK and time kernels, with coverage at every requested epoch. An unavailable transformation raises the original SPICE error. These are position rotations, not 6D state transforms.
Source: https://naif.jpl.nasa.gov/pub/naif/toolkit_docs/C/cspice/pxform_c.html
- orbitalpv.environment_data.inspect_environment_netcdf(path)[source]
Inspect provider variable names/dimensions/units before selecting channels.
- orbitalpv.environment_data.parse_spenvis(text)[source]
Parse scalar/vector variables, metadata, annotations and multiple blocks.
Vector elements become
NAME[0],NAME[1], etc. Their energy/coordinate association remains inmetadataand must be selected explicitly. Units are retained verbatim in each variable and indata.attrs['units']. Truncated records, inconsistent column/header counts, duplicate names, and provider error footers raiseValueError. FORTRAN D exponents are supported.
- orbitalpv.environment_data.read_ceres_toa(path, *, shortwave_variable, longwave_variable, time_variable='time', latitude_variable='lat', longitude_variable='lon', selectors=None)[source]
Read an explicit CF-netCDF CERES subset into a long-form TOA flux table.
selectorsare xarrayiselindices applied before loading. Raw HDF4 editions are not converted implicitly. Use provider-exported netCDF files. Invalid negative flux is marked missing, never clipped to valid zero flux.
- orbitalpv.environment_data.read_poes_metop(path, *, channels, time_variable='time', quality_variables=())[source]
Read chosen SEM-2 channels with their exact units and quality columns.
Select detector-specific variables after inspecting the file. This function does not reinterpret counts as flux, integral channels as differential flux, or local flux as accumulated solar-cell damage. Supports CF time and NOAA’s native daily format: independent same-length 1D variables and milliseconds since 1970 in the documented
time.long_name(units='time'). The latter is recognized only with NOAA POES/MetOp provider metadata, not guessed for arbitrary numeric coordinates. Provider negative sentinel values are retained.
- orbitalpv.environment_data.read_spenvis(path, *, max_bytes=134217728)[source]
Read a local export and attach its checksum to each returned data frame.
- orbitalpv.environment_data.transform_vectors(vectors, from_frame, to_frame, ets)[source]
Rotate a fixed vector or a time-aligned (n, 3) array using SPICE attitude.
- orbitalpv.environment_data.utc_to_ephemeris_seconds(timestamps)[source]
Convert explicit UTC strings to SPICE ET using the loaded leap-second kernel.
SPICE errors for missing kernels or invalid dates propagate; no approximate UTC/TDB conversion is substituted. A loaded LSK is normally required.
Ground PV research and inventories#
Reproducible terrestrial PV research with explicit reference and model provenance.
The ANDPI spectral formula and histogram decision rule follow Zeng et al.
(2025), DOI 10.1016/j.jag.2025.105005, and the authors’ MIT-licensed
Fusion_PV_Mapping implementation at commit
ba2f0cd340c57c6780ed63770a9a89b855fde54b. This module does not silently
substitute a spectral threshold for the paper’s prior-inventory fusion pipeline.
Other algorithms are identified as baselines or caller-supplied model adapters.
ANDPI source license (retained here for installed distributions):
MIT License. Copyright (c) 2025 Xiaopeng Zeng.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- class orbitalpv.ground_research.CopernicusDataSpaceConnector(*, client=None, timeout_s=60.0)[source]
Current CDSE STAC discovery and authenticated OData product retrieval.
Catalog queries are public. Product download requires an explicit access token obtained using the user’s CDSE account; it is never written to disk. S3 URLs in STAC are retained verbatim, not falsely treated as public HTTPS.
- Parameters:
client (httpx.Client | None)
timeout_s (float)
- class orbitalpv.ground_research.EumetsatConnector(*, consumer_key=None, consumer_secret=None, datastore=None)[source]
Thin, working EUMDAC SDK adapter with explicit account credentials.
EUMDAC is imported only when this connector is created without an injected SDK store. Authentication, license checks and token refresh remain with the official SDK. It is not an anonymous-access or license-bypass wrapper. Source: https://gitlab.eumetsat.int/eumetlab/data-services/eumdac.
- class orbitalpv.ground_research.GeometryCapacityModel(feature_names, mean, scale, coefficients, intercept, residual_quantile, coverage, capacity_basis, area_basis, feature_min, feature_max, training_ids, calibration_ids, provenance)[source]
Log-linear area/geometry regression with separate conformal calibration.
Calibration uses absolute log residuals and the finite-sample order statistic ceil((n+1)*coverage). Coverage requires exchangeability; a held-out country’s observed coverage is reported rather than guaranteed. AC/DC/reported and polygon-area semantics must match the calibration.
- Parameters:
- predict(frame, *, area_basis, output_basis=None, dc_ac_ratio=None, allow_extrapolation=False)[source]
Predict median-like MW and calibrated intervals; refuse unseen feature ranges by default.
Converting AC/DC requires an explicit DC/AC ratio and is labelled an assumption. An unspecified
reportedbasis cannot be converted. The ratio’s own uncertainty is not represented by these intervals.
- class orbitalpv.ground_research.GoesABIConnector(*, client=None, timeout_s=60.0)[source]
Anonymous NOAA GOES-R ABI original-netCDF listing and downloads.
Select the satellite explicitly for the observation date. Listing a satellite does not assume that it was the operational GOES-East then. Source: https://registry.opendata.aws/noaa-goes/.
- Parameters:
client (httpx.Client | None)
timeout_s (float)
- class orbitalpv.ground_research.InventoryAsset(provider: 'str', version: 'str', url: 'str', license: 'str', citation: 'str', observation_period: 'str')[source]
- class orbitalpv.ground_research.PNOAConnector(*, client=None, timeout_s=60.0)[source]
IGN PNOA original COG downloads and explicitly rendered WMS images.
Select original COG URLs from the CNIG catalog. WMS is a display product: the requested width/height is not a claim of native spatial resolution. See https://pnoa.ign.es/web/portal/pnoa-imagen/visualizadores-y-servicios-web.
- Parameters:
client (httpx.Client | None)
timeout_s (float)
- download_original(url, destination, *, acquisition_date, attribution, max_bytes=1000000000, overwrite=False, expected_sha256=None)[source]
Download a caller-selected IGN/CNIG original, without resampling.
- download_map(destination, *, bbox, layer, width=1024, height=1024, historical=False, acquisition_date='mixed_or_unknown; inspect original metadata', overwrite=False)[source]
Save a WMS PNG using explicit longitude/latitude WMS 1.1.1 axes.
- class orbitalpv.ground_research.PVInventoryConnector(*, client=None, timeout_s=60.0)[source]
Download pinned GRW, GM-SEUS, and ANDPI inventory releases.
The multi-GB GM-SEUS archive is downloaded only on explicit request. A spatial subset is not independent validation of a model trained on it.
- Parameters:
client (httpx.Client | None)
timeout_s (float)
- class orbitalpv.ground_research.SegmentationAdapter(name, predictor, threshold=0.5, training=(), metadata=<factory>)[source]
Uniform callable adapter for an explicitly supplied segmentation model.
The predictor receives ONLY
PVFeaturesand returns an aligned score grid orSegmentationResult. For PANEL/TransPV the caller loads their authorized local checkpoint in its original runtime and passes a callable. This class neither recreates their architecture nor downloads gated weights. Record training regions and pretraining overlap when known.- Parameters:
- classmethod from_callable(name, predictor, *, weights_path=None, threshold=0.5, training=(), license_name, implementation='caller-supplied runtime; architecture not implemented by orbitalpv', pretraining_overlap='unknown')[source]
Register an external implementation and checksum local weights.
No pickle/checkpoint is opened by orbitalpv; the caller controls the runtime. Supplying a path is provenance, not automatic weight loading.
- predict(features)[source]
Run inference without passing labels and validate the output grid.
- Parameters:
features (PVFeatures)
- Return type:
SegmentationResult
- class orbitalpv.ground_research.SegmentationResult(scores, mask, metadata)[source]
Scores and mask on the original feature grid, with invalid pixels masked.
scoresmay be a spectral index or an uncalibrated classifier score; its meaning is recorded inmetadataand never assumed probabilistic.- Parameters:
scores (MaskedArray)
mask (MaskedArray)
- orbitalpv.ground_research.adaptive_andpi(bands, *, candidate_mask=None, mode='new', retain_ratio=None)[source]
Apply the published ANDPI histogram valley rule to candidate pixels.
mode='new'uses positive peaks above 0.02; fewer than two peaks give the upstream threshold of 1.0.mode='old'uses the last two peaks, falling back to the candidate mean. The source uses 200 density bins over [-0.5, 0.5], height 2, distance 5 and prominence 0.5. Withretain_ratioan insufficient retained fraction restores the input candidate mask, as in the authors’ prior-inventory retention step.Omitting a candidate mask applies the rule to the whole usable scene. That is an explicitly identified standalone adaptation, not the full paper’s historical-inventory fusion or a supervised global detector.
- orbitalpv.ground_research.andpi_index(bands)[source]
Return the paper’s NDPI spectral index used by adaptive ANDPI.
Formula:
(B11 - max(B2, B3, B4, B8, B12)) / (B11 + 1e-10). Named inputs are calibrated reflectance, on the same grid: blue, green, red, nir, swir16 and swir22. Sentinel-2 scale AND offset must already have been applied. Negative reflectance is retained, as in the source rule; nonfinite and near-zero denominators are masked for numerical safety. This spectral score does not resolve module outlines from 10/20 m pixels.- Parameters:
- Return type:
- orbitalpv.ground_research.benchmark_segmentation(adapters, samples)[source]
Evaluate geographically held-out scenes on explicitly known labels.
Rejects training-site reuse and geographic overlap BEFORE inference. Unknown labels and invalid pixels are excluded from both area measures. Completely labelled false-positive connected objects are counted; objects touching unknown support are counted as unevaluable, not false positives. Metrics inherit the supplied reference’s quality (e.g. OSM weak labels). Foundation-model pretraining overlap cannot be inferred from checkpoints.
- orbitalpv.ground_research.evaluate_geometry_capacity(model, reference)[source]
Score untouched sites with empirical coverage, error and interval width.
- orbitalpv.ground_research.fit_geometry_capacity_model(training, calibration, *, feature_names=('footprint_area_m2',), capacity_basis, area_basis, source, coverage=0.9, ridge=1e-06, group_column=None)[source]
Fit on training sites and calibrate on disjoint sites/groups only.
Tables contain stable
site_id, positivecapacity_mwand the selected predictors. Area columns ending_m2are log transformed; other geometry predictors stay linear. Never select features or tune thresholds on the calibration or final test set.group_columnalso rejects reused countries/regions across fitting and calibration.
- orbitalpv.ground_research.load_pv_inventory(path, *, bbox=None, layer=None, archive_member=None, max_features=100000)[source]
Load/subset GeoJSON or optional GeoPackage/Shapefile using original CRS.
Non-GeoJSON uses geopandas/pyogrio lazily. For a ZIP select its vector member explicitly; no files are extracted into the working directory. Raw source fields (including Source/license/measurement provenance) are not renamed. An available download sidecar is checksum-verified before reading.
- orbitalpv.ground_research.match_pv_changes(before, after, *, before_time, after_time, complete_coverage=False, min_overlap=0.1, relative_area_tolerance=0.05)[source]
Match two dated inventories, grouping polygon splits and merges.
Links require intersection/minimum(area) >=
min_overlap. Connected bipartite groups avoid treating a split as construction. Areas use an equal-area projection. Added/removed interpretations require an explicit assertion that BOTH snapshots completely survey the same footprint. Otherwise report newly observed/not redetected. Detection date brackets never assert commissioning or an exact construction date.
- orbitalpv.ground_research.summarize_pv_geometry(collection)[source]
Measure polygon footprint geometry in each feature’s local UTM CRS.
Reports minimum-rectangle long-axis azimuth modulo 180 degrees. This is a row/footprint axis, NOT panel facing azimuth or tilt. A farm-boundary polygon does not become module surface area. Input attributes, original geometry source and the metric CRS are preserved in each output record.
- orbitalpv.ground_research.supervised_segmentation_adapter(samples, *, method='logistic', random_state=0, max_samples_per_class=5000, threshold=0.5)[source]
Fit balanced logistic regression or the existing spectral random forest.
Class sampling, scaling and fitting see training scenes only. Logistic regression is a useful inexpensive supervised alternative, not a neural paper reproduction. The RF uses orbitalpv’s persisted numeric-tree model. Both report uncalibrated scores and retain geographic training provenance.
Satellite irradiance forecast baselines#
Transparent irradiance baselines for calibrated cloud/clear-sky-index grids.
These are operationally useful comparison methods, not reproductions of a foundation weather model. Input irradiance or cloud transmission must already be physically retrieved; raw ABI/SEVIRI radiance is not irradiance. Forecasts record retrieval provenance, motion assumptions and missing inflow boundaries.
- class orbitalpv.ground_forecasting.CloudMotion(row_pixels_per_minute, column_pixels_per_minute, minutes_between, method, quality)[source]
Global or dense apparent motion in pixels/minute on the supplied grid.
- class orbitalpv.ground_forecasting.IrradianceForecast(ghi_wm2, lead_minutes, motion, metadata)[source]
GHI forecast (lead, rows, columns), motion and auditable assumptions.
- orbitalpv.ground_forecasting.advect_cloud_field(field, motion, *, lead_minutes)[source]
Backtrace a scalar field under frozen motion without wrapping image edges.
Missing incoming boundaries and missing source pixels remain masked. Dense flow is treated as constant over the forecast horizon; no learned cloud creation/dissipation is implied. Zero lead returns the usable field.
- Parameters:
- Return type:
- orbitalpv.ground_forecasting.blend_irradiance_forecasts(nowcast, weather_forecast, *, lead_minutes, forecast_origin, weather_issue_time, weather_kind='forecast', transition_minutes=120)[source]
Blend with exponential nowcast weight exp(-lead/transition_minutes).
Use a weather forecast issued no later than the forecast origin. A reanalysis is rejected as operational forecast input to prevent hindsight leakage. Inputs are already on the same time/grid/unit support. Missing one source uses the other; neither source available remains masked. Transition time is a caller choice, not fitted on the evaluation period.
- orbitalpv.ground_forecasting.clear_sky_persistence(observed_ghi, latest_clear_sky_ghi, future_clear_sky_ghi, *, minimum_clear_sky_wm2=20.0, maximum_clear_sky_index=1.5)[source]
Persist the latest daytime clear-sky index into future clear-sky GHI.
Inputs broadcast, e.g. a spatial grid and
(lead, row, column)future values. Future night (clear-sky GHI == 0) is zero. A nighttime/latest low-sun observation does not define a future daytime index and is masked. The maximum index allows specified cloud enhancement and is explicit. Values are W/m², not accumulated radiation energy.
- orbitalpv.ground_forecasting.estimate_cloud_motion(previous, current, *, minutes_between, method='lucas_kanade', max_shift_pixels=30, local_window=9, iterations=3)[source]
Estimate translation, optionally refined by local Lucas-Kanade optical flow.
Phase correlation estimates integer global advection. The default then solves local gradient least squares after warping, with Gaussian spatial averaging and at most two pixels correction per iteration. Texture-poor pixels retain the global estimate. This transparent classical baseline assumes brightness conservation; cloud growth and parallax violate it. A rectangular projected grid and unchanged retrieval scale are required.
- orbitalpv.ground_forecasting.evaluate_irradiance_forecast(predicted, observed, *, persistence=None)[source]
MAE/RMSE/bias on shared valid support; optional RMSE skill vs persistence.
Day/night and lead stratification should be applied by the caller before scoring. When a persistence reference is supplied, all metrics use the exact same three-way support to avoid favourable missing-pixel selection.
- orbitalpv.ground_forecasting.forecast_irradiance(previous_clear_sky_index, current_clear_sky_index, future_clear_sky_ghi, *, minutes_between, lead_minutes, source, license_name, method='lucas_kanade', max_shift_pixels=30, maximum_clear_sky_index=1.5)[source]
Advect retrieved clear-sky index and multiply by future clear-sky GHI.
future_clear_sky_ghiis W/m², with shape (leads, rows, columns). Prior/current inputs are dimensionless retrieved GHI/clear-sky GHI on the same grid. Acquisition spacing, source and license are explicit. Raw satellite radiances and cloud mask class codes are not valid inputs.
Space-based AI screening#
Screening methods inspired by arXiv:2511.19468v2, not a flight design.
SI units unless indicated. Optical far-field and near-field geometry are separate. Radiation dose extrapolations are conditional on the tested device, workload, shielding and spectrum; they are not generic space qualification.
- class orbitalpv.space_ai.OpticalLinkSpec(transmit_power_w=5.0, aperture_diameter_m=0.1, wavelength_m=1.55e-06, aperture_efficiency=0.8, loss_db=3.0, photons_per_bit=195.6, channels=24, channel_rate_gbps=400.0, receiver_threshold_dbm_per_channel=-20.0)[source]#
One-direction total power across DWDM channels in one aperture.
channel_rate_gbpsis explicitly per direction. It must not be inferred by doubling a source’s ambiguous ‘bidirectional’ figure. Aperture efficiency applies separately at transmitter and receiver; loss_db is nonnegative.
- orbitalpv.space_ai.optical_link_budget(distance_m, spec=None)[source]#
Eq. 1 Friis power, photon ceiling and equal-channel receiver margin.
Reject distances below 2 D²/lambda, a conservative far-field boundary.
screened_rate_gbpsis a ceiling, not measured payload throughput. The full-rate receiver threshold is an additional feasibility condition; a negative margin does not establish operation at a reduced modulation rate. Pointing/beam quality must be included in the supplied loss allowance.- Parameters:
spec (OpticalLinkSpec | None)
- Return type:
- orbitalpv.space_ai.confocal_spatial_geometry(*, aperture_diameter_m=0.1, wavelength_m=1.55e-06, beams_per_axis=1, wander_fraction=0.1)[source]#
Paper’s ideal L=pi*a²/lambda geometry; NOT a spatial link budget.
An n-by-n layout uses diameter D/n per sub-aperture. Equal area packing, ideal beam quality and negligible truncation are assumed. Power sharing, crosstalk, terminal electronics and tracking still need separate validation.
- class orbitalpv.space_ai.FormationSpec(altitude_m=650000.0, cluster_radius_m=1000.0, grid_side=9)[source]#
Bounded planar HCW square lattice, independent of paper Fig. 2 data.
x is radial and y along-track in the rotating LVLH frame. Every satellite shares the circular reference mean motion. This omits J2, drag, control, navigation errors, eclipses, occlusion and collision avoidance operations.
- orbitalpv.space_ai.hcw_formation(time_s, spec=None)[source]#
Return (time, satellite, 6) LVLH [x,y,z,vx,vy,vz] in m and m/s.
x=A*cos(nt)+B*sin(nt), y=-2*A*sin(nt)+2*B*cos(nt) is the bounded solution of x’’-2*n*y’-3*n²*x=0, y’’+2*n*x’=0. The parameter lattice is scaled so no member ever exceeds cluster_radius_m in this model.
- Parameters:
spec (FormationSpec | None)
- Return type:
- orbitalpv.space_ai.formation_metrics(time_s, spec=None)[source]#
Sampled all-pairs minimum, cluster radius and neighbor link distances.
Computes one timestep at a time to bound temporary memory. Sampled minima do not certify continuous collision clearance; spacecraft size is omitted.
- Parameters:
spec (FormationSpec | None)
- Return type:
- orbitalpv.space_ai.radiation_event_screen(*, annual_dose_rad_si, dose_per_event_rad_si, mission_years, device_count=1, inferences_per_second_per_device=1.0)[source]#
Conditional constant-rate Poisson extrapolation of a beam-test event.
Events are not necessarily lost inferences. No mitigation, downtime or cross-device correlation is modeled. Input dose is supplied, not derived from orbit, shielding thickness or a solar-weather index.
- orbitalpv.space_ai.launch_learning_price(*, initial_price_usd_kg, initial_cumulative_mass_t, cumulative_mass_t, learning_rate=0.2)[source]#
Wright curve C=C0*(M/M0)**log2(1-learning_rate), without a date forecast.
- orbitalpv.space_ai.launched_power_cost(*, mass_kg, power_kw, lifetime_years, launch_price_usd_kg)[source]#
Undiscounted launch-only USD per nameplate kW-year; not LCOE or TCO.
- orbitalpv.space_ai.launch_energy_cost_parity(*, mass_kg, continuous_compute_kw, lifetime_years, electricity_usd_kwh, terrestrial_pue, utilization=1.0)[source]#
Launch price equating launch-only cost with terrestrial electricity spend.
Equal utilized compute electrical power is assumed, not equal useful work. Excludes spacecraft/chips, operations, replacement, financing and network.
- orbitalpv.space_ai.load_space_ai_evidence(path=None)[source]#
Load the packaged, manually curated v2 evidence snapshot or an override.
Paper numbers retain locators and evidence status. This is not PDF parsing or a live hardware catalogue. The source PDF is not redistributed.
Reproducible tables linking paper evidence to orbital compute screening.
- orbitalpv.space_ai_analysis.space_ai_trade_study(evidence=None)[source]#
Return offline sensitivity tables; no forecast or qualification verdict.
Paper inputs, independently calculated values and analyst scenarios retain distinct fields. Read the evidence snapshot for page locators and caveats. No external service is contacted and no output files are written here.
Panel electrical acceptance#
Public panel electrical-retention acceptance and test-planning API.
- class orbitalpv.acceptance.PowerRetentionAssessment(observations, retention, lower_retention, minimum_retention, confidence, repeat_mean_log_sigma, differential_calibration_log_sigma, combined_log_sigma, point_estimate_accepted, accepted)[source]#
Dimensionless retention and natural-log standard uncertainties.
acceptedconcerns electrical retention only, not flight qualification.lower_retentionis a one-sided bound conditional on the supplied known error model, not a posterior probability that this panel conforms.
- orbitalpv.acceptance.plan_power_retention_test(expected_retention, observations, *, repeat_log_sigma, before_calibration_log_sigma, after_calibration_log_sigma, calibration_correlation=0.0, minimum_retention=0.95, confidence=0.95)[source]#
Compute the bound at an assumed point estimate, without inventing samples.
This is a sensitivity calculation, NOT acceptance probability or assurance that the planned experiment will pass. Error assumptions match
power_retention_acceptance(); all sigmas are natural-log units.
- orbitalpv.acceptance.power_retention_acceptance(before_power_w, after_power_w, *, repeat_log_sigma, before_calibration_log_sigma, after_calibration_log_sigma, calibration_correlation=0.0, minimum_retention=0.95, confidence=0.95)[source]#
Assess paired Pmax measurements at matched AM0 spectrum and temperature.
Arrays are positive finite 1-D watts, same length, in matching pair order for ONE panel. Retention is exp(mean(log(after)-log(before))). The estimand is a common multiplicative retention, not a ratio of arithmetic means.
Each paired log ratio has independent Gaussian repeat error with KNOWN standard deviation
repeat_log_sigma. It must come from a separately validated error budget, not a small-sample plug-in standard deviation. Before/after campaign calibration errors are jointly Gaussian, centered, constant across repeats and independent of repeat error. Their supplied standard uncertainties and correlation determine differential uncertainty; these systematic terms do not shrink with the number of measurements.The lower bound is exp(mean_log_ratio - z_confidence * combined_sigma). A failed bound means insufficient evidence, not proven panel failure. Duplicated rows, drift, unmatched illumination/temperature, different panels or uncorrected fixed bias violate the model; this function cannot detect them.