Power Load Forecasting Based On User Clustering And Adaptive Combinatorial Algorithm

With the transformation and upgrading of China’s economy and the deepening of electricity market-oriented reforms,power load forecasting has become an important issue in the power system.In this thesis,we firstly summarized the research status of power load forecasting,introduced the basic theory and related technologies of power load forecasting in detail.Data preprocessing was carried out on the 183-day hourly power consumption of 1293 power users,including 0 values,negative values,and abnormal values in more than 5.67 million pieces of data,and data normalization and calendar attribute marking were completed.Through the analysis of the power load data line chart,we found that there are differences in the power consumption characteristics among users.According to the characteristics of daily electricity consumption,K-means cluster analysis was carried out on the power users,which were divided into four types: "W" type,"M" type,peak fluctuation type and valley fluctuation type.Secondly,the clustered power loads were forecasted using three prediction algorithms,ARIMA,MLP,and LSTM.The daily average prediction errors are 3.74%,3.73%,and 3.90%,respectively.VMD method was used to overcome the limitations of the ARIMA linear model,the VMD-ARIMA prediction algorithm was constructed,and its daily average prediction error is reduced by 0.39% compared with the single ARIMA algorithm.Finally,based on analyzing the respective advantages of the three single algorithms VMD-ARIMA,MLP,and LSTM,according to the prediction results of the single algorithm in the past,the genetic algorithm was used to assign weights to the single algorithm,so that the weights are dynamically updated in time,and the adaptive combinatorial prediction algorithm was built.The actual prediction results of the power load data of1293 power users show that,after the power users were clustered,the prediction errors of a single algorithm are reduced by 0.05%,0.09% and0.36% respectively;The daily average MAPE of the built adaptive combinatorial forecasting algorithm is 3.62%,which are 0.29%,0.28% and0.62% lower than the single algorithm respectively,and the daily maximum error and error variance are lower than those of the single algorithm.The error of the prediction algorithm studied in this thesis meets the requirements of practical application in the market,and was applied to the corresponding power load forecasting market.