基于NGO-LSTM神经网络的短期光伏功率预测

Short-Term Photovoltaic Power Prediction Based on NGO-LSTM Neural Network

摘要: 为了进一步提高光伏发电功率预测的精度，本文提出了一种用于短期光伏发电功率预测的NGO-LSTM模型。选择在时间序列问题处理上具有良好性能的长短期记忆(LSTM)神经网络，并通过全局搜索能力强、算法鲁棒性高的北方苍鹰算法对LSTM网络隐含层神经元个数、学习率和训练次数等超参数进行优化，得到NGO-LSTM模型。采用印度安得拉邦某光伏发电场功率数据进行算例分析，仿真结果表明，NGO-LSTM模型比BP、GA-BP和LSTM具有更高的预测精度、更好的预测稳定性。可为调整电网计划和配电，优化发电效益，帮助光伏电站运维管理提供可靠参考。

Abstract: To further improve the accuracy of photovoltaic (PV) power generation forecasting, this paper pro-poses a NGO-LSTM model for short-term PV power generation forecasting. The model selects the long short-term memory (LSTM) neural network, which has good performance in time series pro-cessing, and optimizes the hyperparameters such as the number of hidden layer neurons, learning rate, and training times through the Northern Goshawk Optimizer (NGO) algorithm, which has strong global search ability and high algorithm robustness, to obtain the NGO-LSTM model. Case studies are conducted using power data from a PV power generation field in Andhra Pradesh, India. Simulation results show that the NGO-LSTM model has higher prediction accuracy and better pre-diction stability than BP, GA-BP, and LSTM models. This model can provide reliable references for adjusting grid plans and distribution, optimizing power generation benefits, and assisting in the operation and management of PV power stations.