Time Series Decomposition

STL (Seasonal and Trend decomposition using Loess) splits a time series into three components: trend, seasonal, and remainder.

This example shows two use cases:

1. Long term — monthly energy data over ~10 years (period=12)

2. Short term — 15-minute power data over 5 days (period=96)

All plots are interactive — hover to see exact values.

import plotly.io as pio
pio.renderers.default = "notebook_connected"

Long Term Decomposition

Monthly heating energy consumption of a flat over 10 years. An energy optimization in mid-2017 is clearly visible in the trend and the magnitude of the seasonal pattern.

import pandas as pd
from importlib import resources

data_path = resources.files("pyedautils") / "data" / "decomposition_long.csv"
df_long = pd.read_csv(data_path, sep=";")
df_long.head()

	timestamp	value
0	2010-02-01	600.0
1	2010-03-01	550.0
2	2010-04-01	528.0
3	2010-05-01	259.0
4	2010-06-01	27.0

from pyedautils.plots import plot_decomposition

fig = plot_decomposition(
    data=df_long,
    period=12,
    s_window=7,
    title="Time Series Decomposition over 10 Years",
    ylab="Energy Consumption [kWh/month]",
)
fig.show()

Short Term Decomposition

15-minute electricity meter data over 5 days. The daily cycle (period=96, i.e. 96 intervals of 15 min per day) is clearly decomposed into a weekday pattern (trend) and the repeating daily profile (seasonal).

data_path = resources.files("pyedautils") / "data" / "decomposition_short.csv"
df_short = pd.read_csv(data_path, sep=";")
df_short.head()

	timestamp	value
0	2015-01-26 00:00:00	86.0
1	2015-01-26 00:15:00	100.0
2	2015-01-26 00:30:00	60.0
3	2015-01-26 00:45:00	82.0
4	2015-01-26 01:00:00	90.0

fig = plot_decomposition(
    data=df_short,
    period=96,
    s_window=193,
    title="Time Series Decomposition over 5 Days",
    ylab="Power [kW]",
)
fig.show()