A Morphological Filter-based Local Prediction Method with Multi-variable Inputs for Short-Term Load Forecast

This paper presents a prediction model for short-term electric load forecast based on Local Prediction (LP) with a dual-SE weighted morphological filter derived from Mathematical Morphology (MFLP). The historical load data with frequent fluctuations is processed by a morphological filter to obtain a relatively smooth load curve and meanwhile reserve the characteristics of the load. After filtering out the volatility, the obtained time series is embedded into a high-dimension phase space by the LP. Moreover, weather conditions such as local temperature and humidity can also be involved in the proposed MFLP, by embedding them as an individual temperature series and a weather series, respectively, to form a forecast sample. The nearest neighbours who have high similarity to the forecast sample are selected to construct the training set and then predicted by Support Vector Regression (SVR). In order to evaluate the performance of the proposed model, simulation studies have been carried out, respectively, on data collected by AEMO and Elia, in comparison with the SVR, Back Propagation Neural Network (BPNN) and persistence (Per.) models. The results demonstrate that the accuracy and stability of the proposed model are much better than the traditional models.