Medium and Long-term Load Forecasting Method Considering Multi-time Scale Data

Accurately load forecasting is critical to power system. Massive smart grid data with complex structures is collected from different electrical installations. The factors affecting the load are complicated. Most of them are distributed on different time scales. It is necessary to consider the dependence between multi-scale data and make full use of data from different time scales in order to make accurate, stable and reliable medium and long-term predictions. This paper proposed a stacked long-term and short-term memory (LSTM) network model, which integrated three different time scales data of monthly historical load, annual regional economic and daily climate information. In the model, in order to avoid the algorithm is overfitting, all the parameters excepted the offset parameters which in the circulatory neural network were regularized. An example of Guangdong load forecasting proved that compared to the method using the support vector machine (SVM) and deep neural networks (DNN) method, the accuracy of load forecasting is effectively improved. © 2020 Chin. Soc. for Elec. Eng.