SignalComparisonEvents¤

Compares an actual signal with a reference signal (setpoint vs actual, sensor A vs sensor B): divergence intervals, deviation statistics per window, the tracking-error trend and per-window correlation.

ts_shape.events.engineering.signal_comparison · Engineering guide · API reference

Example¤

import pandas as pd
import ts_shape
from ts_shape.events.engineering.signal_comparison import SignalComparisonEvents

df = pd.concat([
    ts_shape.make_timeseries(["pressure_sp"], n_points=1440, freq="1min", baseline=8.0, noise=0.0),
    ts_shape.make_timeseries(["pressure_pv"], n_points=1440, freq="1min", baseline=8.0, noise=0.2, drift=3.0),
], ignore_index=True)

cmp = SignalComparisonEvents(df, reference_uuid="pressure_sp")
result = cmp.detect_divergence("pressure_pv", tolerance=2.0, min_duration="10min")
print(result[["start", "end", "max_deviation", "direction"]])
stats = cmp.deviation_statistics("pressure_pv", window="1h")
trend = cmp.tracking_error_trend("pressure_pv", window="6h")

Methods¤

Method	Returns
`detect_divergence(actual_uuid, tolerance, min_duration='1min')`	intervals with \|actual − reference\| > `tolerance`: `max_deviation`, `mean_deviation`, `direction`
`deviation_statistics(actual_uuid, window='1h')`	per window: `mae`, `max_error`, `rmse`, `bias`
`tracking_error_trend(actual_uuid, window='1D')`	per window: `mae`, `trend_slope`, `trend_direction`
`correlation_windows(actual_uuid, window='1h')`	per window: `correlation` (Pearson), `sample_count`

Notes¤

Deviation is actual − reference, so a positive bias means the actual reads high.
Two signals can drift apart slowly while staying inside tolerance; a falling correlation or rising mae trend shows it.
correlation_windows skips windows with fewer than two aligned samples.