Abstract: Aiming at the problems of modal decomposition aliasing and difficult selection of long short term memory network parameters in the traditional power load forecasting model, this paper proposes a new model, namely the long short-term memory network model based on variational modal decomposition and improved pelican optimization algorithm. Firstly, variational modal decomposition is used to decompose the power load data into multiple modes with low complexity, and the subseries are recombined by permutation entropy to reduce the prediction difficulty. Then, two strategies, logistic chaotic mapping, fusion Cauchy mutation and reverse learning, are introduced to improve the global optimization ability. Then, the improved pelican optimization algorithm is used to optimize the long short term memory network parameters to improve the generalization ability and practical operation of the model. Finally, the reconstituted submodes are predicted and superimposed respectively to obtain the final prediction results and compared with the prediction models of different optimization algorithms. Experimental results show that the long short term memory network model of variational mode decomposition-permutation entropy-improved pelican optimization algorithm has higher prediction accuracy and stability, and can effectively predict short-term power load.

Key words: variational mode decomposition, permutation entropy, pelican optimization algorithm, long short-term memory network, short-term load forecasting