The Method of Short Term Load Forecasting for Micro Grid Using Limit Learning

Abstract

Considering the cost constraint and the uncertainty of power consumption, a short-term load forecasting method for microgrid based on kernel function extreme learning machine is proposed. The use of kernel extreme learning machine and heuristic genetic algorithm and time of training samples, the offline optimization of the parameters of prediction model and online load forecasting including periodic update of model parameters; through to ensure the timeliness of algorithm of optimal parameters, while reducing the computational complexity of online prediction system and historical data storage. Through short-term load forecasting for different capacity and type of user side microgrids, the accuracy of prediction results, the effect of parameter cycle update, the impact of prediction results on economic operation and the computational efficiency of prediction methods are analysed.