Use Vector AutoRegression (VAR) model to forecast the power consumption
- Stationarity Check and Differencing: The project involves a step where the stationarity of the dataset is confirmed, possibly using the Dickey-Fuller test. If the data were not stationary, a differencing approach would be applied to achieve stationarity, which is crucial for time series forecasting.
- Forecasting Problems: The notebook outlines three forecasting scenarios:
- Short-term forecasting (predicting 1 week ahead based on historical data).
- Long-term forecasting (predicting 2 months ahead based on historical data).
- Multi-step forecasting for 1 month, where each prediction is used as input for subsequent predictions.
- Prediction and Model Fitting: It includes steps for creating predictions using the VAR model. For instance, one step involves creating an array
prediction1
that contains predictions for 7 values based on a test dataset. - Multi-step Forecast Implementation: The project describes implementing a multi-step forecast function,
ms_forecast
, which predicts multiple future observations in a sequence, using each prediction as input for the next.