Similar Daily Analysis And Short-term Load Forecasting Under The Influence Of Environmental Factors

With the construction of digital power grid and the reform of electricity marketization,the power demand is increasing,which puts forward higher requirements for the balance between power supply and demand,and accurate power load forecasting is important.Short-term power load forecasting makes a great contribution to the safe operation of the power grid and provide significant references for power grid dispatching.Our study takes the short-term power load as the research object.In view of the characteristics of short-term power load and the influence of many uncertain factors,the PCA method was used to reduce the dimension of the main influencing factors after normalization and K-means clustering method was applied in the feature analysis of similar days.The similar daily load data with similar power characteristics at forecast day was screened out.The similar daily load data and main influence factors were regarded as new data sets of Attention-ResNet-LSTM model and was conducted short-term load forecasting.The specific research contents are as follows:(1)Firstly,the characteristics of power load were analyzed and obtained,then the influence of various influencing factors on load was analyzed,the important influencing factors related with load forecasting including temperature indicator,weather factor,date type,seasonal factor,holiday factor,relative humidity and wind scale were screened out,which prepared for feature analysis of similar days of subsequent multiple influencing factors.In addition,the basic principle,the cause of the error,evaluation index and prediction process of power load were expounded.(2)The similar day characteristic analysis model of multiple influencing factors was established.Considering the problems of many influencing factors and massive load data,the principal component analysis(PCA)was used to reduce the dimension of the main influence factors after normalization,the principal component with a cumulative contribution of 80%was used to represent the original data,and the new data was considered as the input variable of K-means clustering to conduct the analysis of similar day characteristics,the optimal clustering results was attained by contour coefficient method to select similar daily load data with similar electrical characteristics at prediction days,the similar daily load data was the sample data of similar days for the next short-term load forecasting model.(3)The short-term load forecasting model was established.considering the temporal characteristics of power load data and the nonlinear relationship between power load data and influencing factors.In our study,Attention-ResNet-LSTM was selected to predict short-term load,and the similar daily load data,temperature index,weather factor,date type,seasonal factor,holiday factor,relative humidity and wind scale were the input variables of the model.Compared with the RNN,LSTM,Attention-LSTM model based on the similar day and RNN,LSTM,Attention-LSTM,Attention-ResNet-LSTM without the similar day,Attention-ResNet-LSTM has more prediction accuracy.