基于PSO-RNN的光伏发电功率预测研究

针对光伏发电功率预测对电力系统的安全稳定和经济运行问题,本文提出了基于粒子群算法优化脊波神经网络的光伏功率预测模型。采用脊波函数作为隐含层激励函数的神经网络,即脊波神经网络,同时采用粒子群算法优化脊波神经网络的权值,并以实际光伏发电站的历史光伏发电数据和气象数据作为仿真算例,对预测模型进行仿真和测试。仿真结果表明,与BP神经网络预测模型相比,基于粒子群算法优化脊波神经网络预测模型的日平均绝对误差和日最大绝对误差均有所降低,证明粒子群算法优化脊波神经网络的预测模型具有较高预测精度,不仅加快了脊波神经网络收敛速度,而且避免了陷入局部最优解,具有一定的实用性及可行性。该研究为光伏发电功率预测提供了理论参考。

Study of Power Forecasting of Photovoltaic Based on PSO-RNN

For the problem of the security,stability and economic operation of the power system which is influenced by the prediction of the photovoltaic power generation,this paper proposed a photovoltaic power prediction model based on ridgelet neural network which is optimized with particle swarm optimization.The neural network using ridge wave function as the hidden layer activation function is called the ridgelet neural network,at the same time,optimizing weights of ridgelet neural network with particle swarm optimization,and taking the history power generation data and meteorological data of an actual PV power plant as samples to train and test the forecasting mode.The simulation results show that,compared with BP neural network model,the average daily absolute percentage error and the daily maximum absolute percentage error using the ridgelet neural network model optimized with the particle swarm optimization are both decreased,it proved that the ridgelet neural network model optimized with the particle swarm optimization can obtain a higher forecasting accuracy,and can not only speed up the convergence speed of ridge wave neural network,but also avoid falling into local optimal solution,which is practical and feasibility for the study.The study provides a theoretical reference for the prediction of the photovoltaic power generation.