Multivariate regression tree for pattern-based forecasting time series with multiple seasonal cycles

Abstract

Multivariate regression tree methodology is used for forecasting time series with multiple seasonal cycles. Unlike typical regression trees, which generate only one output, multivariate approach generates many outputs in the same time, which represent the forecasts for subsequent time-points. In the proposed approach a time series is represented by patterns of seasonal cycles, which simplifies the forecasting problem and allows the forecasting model to capture multiple seasonal cycles, trend and nonstationarity. In application example the proposed model is applied to forecasting electrical load of power system. Its performance is compared with some alternative models such as CART, ARIMA and exponential smoothing. Application examples confirm good properties of the model and its high accuracy.