Journal Article

Distributed Photovoltaic Power Energy Generation Prediction Based on Improved Multi-objective Particle Algorithm

Qian, Xiaorui, Xu, Ming, Chen, Feifei, Ma, Hanbin, Hong, Huawei, Xiao, Yuanzheng

2025

Overview

Accurate prediction of distributed photovoltaic (DPV) power generation is crucial for stable grid operation, yet existing methods struggle with the non-linear, intermittent nature of solar power, and traditional machine learning models face hyperparameter selection and overfitting challenges. This study developed a highly accurate DPV power prediction method by optimizing a Long Short-Term Memory (LSTM) network's hyperparameters using an improved Multi-Objective Particle Swarm Optimization (MO-PSO) algorithm. A hybrid LSTM-PSO model was created, where the LSTM network served as the core prediction model, and the improved MO-PSO algorithm optimized its hyperparameters, enhancing generalization and avoiding overfitting. The LSTM-PSO model significantly improved prediction accuracy compared to traditional methods. Key results from two power stations included a maximum deviation of 6.2 MW at Power Station A, a peak time deviation of less than 0.1 MW at Power Station B, and a prediction interval error controlled below 30 MW at an 80% confidence level. The optimized LSTM-PSO model effectively captures DPV power generation dynamics, and the superior performance metrics demonstrate its potential for intelligent grid management. However, limitations include prediction accuracy under extreme weather and computational efficiency for large datasets. Future work will focus on broader applicability and more efficient algorithm variants.

• Publisher: European Alliance for Innovation (EAI)
• Subject: Accuracy / Algorithms / Confidence intervals / Deviation / Distributed generation / Distributed photovoltaic power generation / Intelligent Power Grid / Long short-term memory network / Machine learning / Multiple objective analysis / Particle Algorithm / Particle swarm optimization / Performance measurement / Power plants / Power prediction / Prediction models