Dependent multi-series forecasting
Three hourly sensors of a compressor (flow, temperature and pressure) influence each other: a jump in pressure is followed by a peak in flow and then by a rise in temperature. The operating mode of the compressor (eco or full) is an exogenous variable scheduled in advance. ForecasterDirectMultiVariate forecasts one series, flow (the level), for the next 3 hours. Each training row uses the last 2 values of every series and the mode at each target hour as features, and the next 3 values of flow as targets. Following the direct strategy, one model is trained for each step with the lags, the mode of its step and its own target. To forecast, the last values of all the series and the future mode form a single row and each model predicts its own step.
Dependent multi-series forecasting
Three sensors of a compressor (flow, temperature, pressure) and its operating mode, scheduled in advance.
1
Series
2
Training table
3
One model per step
4
Predict
flow
62
73
temp
41
43
pressure
4.3
5.1
skforecast.org
0.3 s