Monthly Load Forecasting Based on an ARIMA-BP Model: A Case Study on Shaoxing City

Load forecasting is a basic issue in grid planning and operation arrangements, and the forecasting accuracy varies from system to system in terms of the local load demand characteristics and different external influencing factors. This paper contributes a monthly load forecasting method based on an autoregressive integrated moving average (ARIMA) model and the classical back-propagation neural network (BPNN). The original sequential monthly energy demand records are firstly decomposed into regular static component, seasonal component and irregular nonlinear component by multiplicative trend decomposition. After that, an ARIMA model is optimally selected and trained to emulate the trend component in terms of its suitable feature of describing linear time series. On the other hand, a three-layer BPNN is employed to govern the connection between the city-specific external meteorological factors with the nonlinear irregular components. the proposed ARIMA-BP model is validated against a representative deep-leaning method–-EMD-LSTM model, through their applications to the whole society monthly demand prediction of Shaoxing city. The application results show that the proposed ARIMA-BP model is obviously superior to the EMD-LSTM model in terms of prediction accuracy.