Abstract:

Short-term load forecasting with high accuracy is the foundation for situation awareness of distribution network.To mine the complex uncertainty information in power load data sequence,a novel load forecasting method is proposed based on quadratic mixed mode decomposition and long short-term memory (LSTM) neural network with moth to fire optimization (MFO) algorithm seeking optimum parameters.Firstly,combined with the integrated empirical mode decomposition (EEMD) and variational mode decomposition (VMD),the quadratic mixed mode decomposition model is built to deal with load data sequence,which can extract relatively stable subsequences in power load data and decrease the influence of disordered uncertainty components in high frequency sequences on load prediction accuracy.Secondly,LSTM optimized by MFO algorithm is put forward to accurately predict short-term load trend including decomposition subsequences.Finally,the generalization ability and the prediction accuracy of the model are verified by the prediction on the basis of load data at node in an actual distribution network.

Key words: short-term load forecasting, long short-term memory (LSTM), moth to fire optimization (MFO), mixed mode decomposition, uncertainty