EnvironmentalCompensationEvents

How much an environmental condition (temperature, humidity, pressure) moves an inline measurement: correlation and sensitivity per window, a compensated signal referred to a nominal condition, and the windows where the environment left the valid range.

ts_shape.events.quality.environmental_compensation · Quality guide · API reference

Example

import numpy as np
import pandas as pd
from ts_shape.events.quality.environmental_compensation import EnvironmentalCompensationEvents

t = pd.date_range("2024-01-01", periods=2880, freq="1min")
hall = 22 + 4 * np.sin(np.arange(2880) / 2880 * 4 * np.pi)
width = 50 + 0.01 * (hall - 20) + np.random.default_rng(0).normal(0, 0.002, 2880)
df = pd.concat([pd.DataFrame({"systime": t, "uuid": "line1:hall_temp", "value_double": hall}),
                pd.DataFrame({"systime": t, "uuid": "line1:width", "value_double": width})])

env = EnvironmentalCompensationEvents(df, "line1:width", env_uuid="line1:hall_temp", env_value=20.0)
print(env.sensitivity(window="8h")[["start", "coefficient", "r_squared"]])
corrected = env.compensated_signal()
hot = env.detect_env_exceedance(valid_range=(15.0, 25.0), window="1h")

Methods

Method	Returns
detect_correlation(window='4h')	per window: start, end, duration_seconds, correlation, p_value, significant
sensitivity(window='8h')	per window: start, end, duration_seconds, coefficient, intercept, r_squared, unit_change_per_degree
compensated_signal(coefficient=None)	per sample: systime, value_double (measured), compensated_value, correction_applied
detect_env_exceedance(valid_range, window='1h')	windows outside the range: start, end, duration_seconds, env_mean, env_min, env_max, exceedance_type (high/low/both), severity

Notes

• env_uuid is required; the constructor raises ValueError without it. env_value is the nominal environment for compensated_signal.
• Both signals are read from the same value_column. Each measurement sample is paired with the nearest environment sample within 5 minutes; unpaired samples are dropped.
• compensated_value = measured − coefficient × (env − env_nominal), with env_nominal = env_value or else the environment mean. Without coefficient, one linear regression over all paired data estimates it.
• significant means Pearson p < 0.05. Windows with fewer than 5 paired samples are skipped. unit_change_per_degree equals coefficient (measurement units per environment unit).
• severity scales with how far the worst value lies outside valid_range, relative to the range width: low up to 5 %, medium up to 20 %, high up to 50 %, critical beyond.