Transformations  in
Business & Economics

SHORT-TERM LOAD FORECASTING BASED ON IMPROVED PARTICLE SWARM OPTIMISATION AND LONG SHORT-TERM MEMORY NETWORK²
Qingyou Yan, Yonghua Wang, Guangyu Qin, Jingyao Zhu, Zilvinas Zidonis

ABSTRACT. Short-term power load forecasting is one of the most important issues for market participants under the context background of Chinese power market reform. However, the instability of the load series makes the forecast difficult. In order to improve the accuracy of the forecast, a hybrid model is established in this study, which combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), permutation entropy (PE), improved particle swarm algorithm (IPSO), and Long Short-Term Memory neural network (LSTM) together. Firstly, the original load sequence is decomposed into several intrinsic mode functions (IMFs) and a residual using CEEMDAN to reduce the volatility of the load sequence affected by complex indicators. Then, the complexity of each IMF is analysed by PE to generate new subsequences with obvious differences in complexity. Finally, the IPSO is used to optimise the learning rate, the number of iterations, as well as the number of hidden layer neurons in LSTM. The final prediction load can be obtained by summing the prediction results of all sub-sequences. It turns out that the proposed model is superior to the other six comparison methods, with the lowest MAE, RMSE, MAPE and Theil's inequality coefficient (TIC), and the highest R² .

KEYWORDS:  short-term load forecasting, complete ensemble empirical mode decomposition with adaptive noise, permutation entropy, improved particle swarm algorithm, short-term memory neural network.

JEL classification:  C63,C83,C90.

²Acknowledge Funding:  This research was funded by National Social Science Foundation of China (19ZDA081), 2018 Key Projects of Philosophy and Social Sciences Research, Ministry of Education, China, (18JZD032), China Scholarship Council Joint PhD Program (202006730045).