An Improved Self-Organizing Migration Algorithm for Short-Term Load Forecasting with LSTM Structure Optimization

Establishing an accurate and robust short-term load forecasting (STLF) model for a power system in safe operation rational dispatching is both required beneficial. Although deep long memory (LSTM) networks have been widely used applications, it still has some problems to optimize, such as unstable network performance optimization time. This study proposes adaptive step size self-organizing migration algorithm (AS-SOMA) improve the predictive of LSTM. First, LSTM prediction developed, which divides structure seeking into two stages. One number hidden layer layers, other optimizes neurons, time step, learning rate, epochs, batch size. Then, logistic chaotic mapping method were proposed overcome slow convergence stacking local optimum SOMA. Comparison experiments with SOMA, PSO, CPSO, LSOMA, OSMA on test function sets show advantages improved algorithm. Finally, AS-SOMA-LSTM solve STLF problem verify effectiveness Simulation that AS-SOMA exhibits higher accuracy speed standard set strong ability application