KETEP 세계에너지시장정보

1.A power load combination prediction method based on a neural network model is characterized by comprising the following steps:S1 collects a plurality of power consumption index features of each user to form an original data set. The data in the original data set is processed to generate an initial data set which is consistent with the characteristic attribute and/or the data format, and the characteristic attribute comprises one or more of a time attribute, an air temperature attribute, a weather attribute, a wind power attribute and a holiday attribute. The data processing includes a mean filling of the missing value and/or the abnormal value with a normal value. S2 standardized processing on the index data in the initial data set, the dimensional difference between different index data is eliminated, and a standard data set is acquired as a sample point. S3 optimizes the clustering model, the standard data set is input to the clustering model to determine the approximate optimal cluster number and optimize the initial center point, and the characteristic features of the plurality of standard data sets are extracted. S4 inputs the original features of the data set and the characterization characteristic into the neural network, outputs a power load value corresponding to the output power load value, and trains and acquires an initial neural network prediction model. S5 performs parameter optimization on the initial neural network prediction model, obtains the optimal power load combination prediction neural network, combines the neural network with the cluster model, and is used for carrying out combination prediction on the electric load. Wherein, step S3 includes. S31 plots the center number and cost function curve of the cluster model, and determines the optimal clustering center number of the current cluster model according to the curve change. S32 selects the initial clustering center point according to the optimal clustering center number randomly, calculates the probability that the sample points in each cluster set are selected as the next cluster center point, and updates the center point of the cluster set until the cluster center point of the current cluster model is obtained. S33 extracts characterizing features of a plurality of data sets using the best clustering center point of the current cluster model.