Short-term Load Forecasting Based On Elman Neural Network

Short Term Lord Forecasting (STLF) is one of the most important contents of dispatching power system. It is the premise and basic of safe, economic and reliable operation of power system. The forecasting precision will greatly affect the economy t and stability of power system. Therefore, STLF has always been one of key directions of research at home and abroad.Research shows that power load is influenced by many factors, while meteorological factors affect the power load most significantly. In this thesis, the load characteristic of Hangzhou area is analyzed, and abnormal data are pre-processing, using the similarity method to investigate the influence of meteorological factors on the load.In recent years, ANN as an intelligent way has been widely used in STLF. However, in load forecasting applications, mainly using BP neural network which is a static network and easy to fall into local minimum, the prediction accuracy is not satisfactory. In this paper, Elman is used in STLF model as a dynamic recurrent neural network. The result shows that the effect base on Elman is significantly better.Since the regional power load is significantly affected by weather factors, a method considering weather factors is proposed. This method uses comprehensive weather factors, namely the human body amenity indicator and THI (temperature and humidity index), as inputs, which overcomes the disadvantages such as too many inputs and long forecasting time when weather factors are direct inputs. Besides, improvements on learrning algorithm, excitation function and the structures of network have been made. The improved model considers the grid’s dynamical performance, decreases the number of inputs and enhances the adaptability of the load forecasting model. This paper has verified the method and model using the data of Hangzhou. The results show that the method and model can significantly increase the precision of prediction.Considering the influence of photovoltaic grid connected to STLF, this paper analysis the physical model and load forecasting model of photovoltaic system. Besides, a computational method to count clipping capacity is proposed. Using the historical data of Hangzhou area to calculate, the result shows that photovoltaic power generation system has the ignored capacity value.