Peak-Valley Load Forecasting Based On Neural Network And Fuzzy Comprehensive Reasoning

Daily peak load and valley load are the highest and the lowest electric power consumption in a day,respectively.Studying the forecasting of daily peak and valley electric power load are of great significance for the structural optimization of electric power system in load shifting,risk reduction as well as efficiency improvement.In this paper,we study the prediction of daily peak and valley load of power system,and give a prediction method based on SOM neural network clustering and fuzzy comprehensive reasoning.The effectiveness of the method is verified by an example analysis of daily peak and valley load forecasting in DOM's power supply area.Firstly,through the character analysis of daily peak load and valley load,it is found that daily peak load and valley load have obvious seasonal characteristics and annual quasi-periodical trend.Then,daily maximum temperature,daily average temperature,daily minimum temperature,dew point,daily peak(valley)load occurrence time,and daily peak(valley)load etc.,are selected as forecasting input variables by statistics analysis.In addition,the SOM neural network is used to clustering the historical power load of the past three years taking the DB index as evaluation indicator.Next,for the daily peak(valley)load forecasting,the fuzzy similarity degree is used to measure the similarity between the input vectors of the day to be predicted and the input vectors of the similar historical days.After,the normalized information gain is used to set the weight of input variables,and then the fuzzy comprehensive similarity between the historical day input and the daily input to be predicted is calculated.Moreover,the similar historical day is selected by the similarity threshold,and the peak(valley)load of each similar historical day is integrated by the fuzzy inference method to determine the fuzzy prediction value of the daily peak(valley)load.Finally,the crisp value of daily peak(valley)load of the day to be predicted is obtained by using the gravity defuzzification method.In order to judge the validity of the forecasting model based on SOM neural network clustering and comprehensive fuzzy inference,the average relative percentage error and the posterior difference ratio were selected as the evaluation indicators,and the prediction results were evaluated and analyzed.The results show that the new model can achieve better prediction results.Compared with the classical Mamdani fuzzy inference method,the average absolute percentage error of the daily peak and valley load forecast based on the model were reduced by 6.67%and 7.93%,respectively.