In order to obtain higher prediction accuracy,a distributed short-term load forecasting based on κ-Medoids clustering and deep learning is proposed.Based on the energy consumption distribution of distribution transformers,κ-Medoids clustering is used to cluster the data of power load data set.A short-term load forecasting model based on deep neural network(DNN)and long short-term memory network(LSTM)is constructed.It is verified in Wuhan distribution network system with 1000 subsets of substation data.The verification results show that the proposed κ-Medoids clustering can fit the average parameters of a single transformer forecasting model on the basis of reducing the training time by 44％.DNN and LSTM forecast models track the actual load with mean absolute percentage error(MAPE)of 7.32％and 11.15％,respectively.
万方数据