Decomposed Profile

Goal

Display daily electricity consumption profiles with the trend component removed, isolating the intra-day pattern from long-term drift.

Data Basis

Electricity data (ele_meter.csv) bundled with the pyedautils package: the mean electric power of a household in W per 15-minute interval (the 15-minute consumption in Wh × 4), for 2015.

Data Preparation

import pandas as pd
from importlib import resources

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

	timestamp	value
0	2015-01-01 00:00:00	40.0
1	2015-01-01 00:15:00	32.0
2	2015-01-01 00:30:00	32.0
3	2015-01-01 00:45:00	34.0
4	2015-01-01 01:00:00	34.0

Visualization

from pyedautils.plots import plot_daily_profiles

fig = plot_daily_profiles(
    data=df,
    method="decomposed",
    loc_time_zone="Europe/Zurich",
    title="Electricity - Decomposed Weekly Pattern",
    ylab="Power deviation [W]",
)
fig.show()

Customization

The trend is a centered rolling mean over k rows. With 15-minute data the default k=672 is one week, so the weekly pattern stays in the profile and slower changes (seasons, holidays) are removed. A window shorter than a day would also remove the daily pattern.

fig = plot_daily_profiles(df, method="decomposed", loc_time_zone="Europe/Zurich", k=4 * 672)   # 4-week trend