energy_analysis.timestral_case#

Reproducible 24-hour Timestral demonstration on an attributed EIA capture.

Input: hourly CISO system generation, not individual plant SCADA. Method: joint solar/wind prediction using history and known calendar covariates only. Analysis: one chronological holdout against persistence and daily persistence. This is a small software demonstration, not a validated forecasting benchmark.

class energy_analysis.timestral_case.TimestralCase(profiles, context, actual, known_covariates, source_sha256)[source]#

Bases: object

Audited observations and disjoint model context/verification arrays.

Parameters:
  • profiles (DataFrame)

  • context (FloatArray)

  • actual (FloatArray)

  • known_covariates (FloatArray)

  • source_sha256 (str)

profiles: DataFrame#
context: FloatArray#
actual: FloatArray#
known_covariates: FloatArray#
source_sha256: str#
property context_sha256: str#

Stable little-endian float64 digest of the target context.

property covariates_sha256: str#

Digest of calendar inputs including the known forecast horizon.

energy_analysis.timestral_case.load_timestral_case(data_directory)[source]#

Verify the existing EIA source checksum and enforce consecutive UTC hours.

Parameters:

data_directory (str | Path)

Return type:

TimestralCase

energy_analysis.timestral_case.forecast_timestral_case(case, connector)[source]#

Run real model inference on history only; verification data never enter it.

Parameters:
Return type:

TimestralForecast

energy_analysis.timestral_case.load_timestral_snapshot(data_directory, case)[source]#

Read saved real inference, checking the output and all model-input digests.

Parameters:
Return type:

TimestralForecast

energy_analysis.timestral_case.evaluate_timestral_case(case, forecast)[source]#

MAE/RMSE/WAPE plus empirical 80% interval coverage, independently per fuel.

Parameters:
Return type:

DataFrame