基于嵌套优化的GA-PSO-BP神经网络短期风功率预测方法研究

Research on Short-Term Wind Power Prediction Methods Based on GA-PSO-PB Neural Network with Nested Optimization

中文关键词:风电功率预测  BP神经网络  遗传算法  粒子群算法  嵌套优化

keywords:wind power prediction  BP neural network  genetic algorithm  particle swarm algorithm  nested optimization

中文摘要:短期风电功率预测对于保障电力系统稳定运行具有重要意义。针对单一BP（back propagation）神经网络预测模型难以满足风电功率的强随机波动特性，结合遗传算法（genetic algorithm，GA）和粒子群智能算法（particle swarm optimization，PSO），提出嵌套优化的GA-PSO-BP神经网络短期风电功率预测模型。建立内外双层嵌套的优化机制，内层机制中引入GA算法优化PSO算法学习因子，优化后PSO算法作为外层机制实现BP神经网络阈值和权值的优化。模拟风电数据预测结果表明，比起GA-BP、PSO-BP、长短期记忆网络（long short-term memory，LSTM）预测模型，所提嵌套优化模型在平均绝对误差（mean absolute error，MAE）、均方根误差（root mean squared error，RMSE）、决定系数R2 3个评价维度上均取得了最优值；利用山西某风电场不同月份、不同时段、不同波动特征的实际运行数据进行验证，预测结果表明MAE均小于0.02，R2均大于0.99，所提嵌套优化模型具有较高的预测精度和拟合程度。

Abstract:Short-term wind power prediction is of paramount significance in ensuring the stable operation of power systems. In response to the difficulty of a single BP （back propagation） neural network prediction model in meeting the strong random fluctuation characteristics of wind power， this study proposes a nested optimization approach named GA-PSO-BP neural network for short-term wind power prediction. This novel model integrates the strengths of both genetic algorithm （GA） and particle swarm optimization（PSO）.The proposed model incorporates a dual-layer nested optimization mechanism. Within this structure，the GA algorithm optimizes the learning factors of the PSO algorithm in the inner layer， while the optimized PSO algorithm serves as the outer mechanism to optimize the threshold and weight values of the BP neural network. Simulation results based on wind power data demonstrate that the suggested nested optimization model outperforms GA-BP， PSO-BP， and LSTM prediction models in terms of three evaluation dimensions： Mean Absolute Error （MAE），Root Mean Square Error （RMSE）， and the coefficient of determination（R2）. Verification is conducted using the actual operation data from different months，time periods，and fluctuation characteristics of a wind farm in Shanxi province. The results reveal that the average absolute prediction error is less than 0.02，and the coefficient of determination exceeds 0.99， showing the high prediction accuracy and fitting degree of the proposed model across various operational scenarios.