Improved BP neural network short-term load forecasting based on semi-Markov chain

Short-term load forecasting in power systems is crucial for their stable operation and economic benefits. Traditional forecasting methods often fail to effectively handle complex nonlinear relationships, external influences, data randomness, volatility, and time delays. To address these issues, an improved BP neural network (BPNN) model combined with a semi-Markov chain (SMC) is proposed. The model considers multiple factors such as weather, temperature, and date, and optimizes the initial weights and thresholds of the BPNN using particle swarm optimization (PSO) and genetic algorithms (GA), enhancing the model's convergence speed and forecasting accuracy. The optimized BPNN is then used for load forecasting, with the results corrected by SMC. Using real load data from the Jiangsu power grid, BPNN, PSO-GA-BPNN, and MC-PSO-GA-BPNN are set as comparison models. Experimental results show that the proposed combined model offers higher accuracy and stability in short-term load forecasting, effectively addressing the randomness, volatility, and time delays in load data, outperforming the other models.