Research on short-term photovoltaic power prediction based on LMD-IPSO-LSVM

With the rapid development of human society, resource shortages and environmental degradation have become increasingly pressing issues. To address these challenges, solar energy has garnered significant attention due to its high efficiency, safety, and pollution-free nature. This paper proposes a novel short-term photovoltaic power prediction framework based on an integrated LMD-IPSO-LSVM approach. The model's key innovation lies in its hierarchical decomposition-optimization architecture: First, Local Mean Decomposition (LMD) addresses the non-stationary and nonlinear characteristics of PV power data by decomposing original signals into physically meaningful Product Functions (PFs). Second, an Improved Particle Swarm Optimization (IPSO) algorithm featuring an adaptive inertia weight mechanism is developed to optimize LSVM hyperparameters for each PF component. This strategic integration enables the model to simultaneously capture complex temporal patterns while maintaining superior generalization capability. Experimental validation demonstrates that our IPSO achieves significantly faster convergence (46.3% improvement in convergence speed) and enhanced optimization precision compared to standard PSO, providing a solid foundation for accurate power forecasting. In order to evaluate the proposed methodology, comparative models including standalone LSVM and PSO-LSVM are also established and tested on the same dataset. Experimental results demonstrate that the proposed hybrid model called LMD-IPSO-LSVM achieves high prediction accuracy and better performance compared with other algorithms.