Study On Modeling Of Multi-factors Load Forecasting Based On Combinations Of Neutral Networks

Short-term power load forecasting is the precondition and basic work to ensure the safe and stable operation of the power grid.How to improve the short-term load forecasting accuracy with neutral network technology has always been a heat topic in this field.With the rapid development of power grid construction and the improvement of residents' living standards,China's power energy consumption and meteorological sensitive load have increased significantly.At the meantime,load characteristics have changed significantly,which leads to decreasing prediction accuracy of traditional short-term load forecasting technology because of complex and diverse influencing factors.Therefore,it becomes harder to forecast loads and these problems have been major difficulties for the work of power load forecasting.Artificial neural networks have strong self-learning and generalization capabilities,and have been widely used in the field of power load forecasting.In recent years,with the rise and development of deep learning,more complex deep neural networks have been applied.Based on the latest research results of artificial neural network,this paper adopts a variety of algorithm models combined with actual load data and relevant meteorological factors to conduct short-term electric load forecasting research,and eventually proposes an optimal machine learning model w:ith considerations of characteristics and advantages of different models.This load forecasting model further improves the accuracy of load forecasting.The main research contents of th is paper include:(1)Introducing the load characteristic indexes which are commonly used in electric load characteristics and analyzing the correlations between the characteristics changes and development rules of electric load.Meanwhile the relationship between electric load and influencing factors is deeply explored.The main factors will be selected by correlation(2)Considering the strong periodicity of load data and other factors,a load forecasting model based on Grey Elman neural network is established.This model combines the advantage of high precision of Grey Theory in condition of a small amount of inputs and the adaptability and controllable error of Elman neural network.The practical model proves that this method can effectively improve the accuracy of load forecasting.(3)The deep learning model can gradually discover the rules from the original load data by adding multiple hidden layers which achieves increasingly accurate result of load forecasting.This paper establishes three deep learning models using multi-factor meteorology data:long-term and short-term memory network(LSTM)and its variant structure-gated cyclic neural network(GRU),and stack self-encoding neural network(SAE).The validity of the three models was verified based on the actual load data and the prediction results are also compared with the outcome given by the model based on the Grey Elman neural network.(4)In order to achieve better load forecasting,this paper proposes a combination use of three kinds of deep learning model by giving a weight to each part.The paper compares three different types of weight allocation,which include equal weights,variance combination and error square sum minimum combination.Eventually the Fourier series residual correction method will be adopted to further improve the load prediction accuracy for the most the superior combination of models.