改进型LSTM网络光伏发电功率预测研究

Research on photovoltaic power generation prediction based on improved LSTM network

针对现有光伏发电功率预测技术存在的未能充分考虑气象因素、提取特征不充分等导致预测精度较低的问题,基于深度学习理论,提出一种基于改进型LSTM网络的光伏发电功率预测方法。根据长短期记忆神经网络的特点,从循环神经网络(RNN)推导出其一般计算过程,阐述该预测方法的优越性和可行性。提出基于改进型长短期记忆(LSTM)网络的光伏发电率预测模型,该模型充分考虑并优化神经网络带来的过拟合问题,且引入RMSProp算法获取模型最佳的损失函数值,确保得到最佳的预测结果。综合考虑对光伏发电功率产生影响的多种气象因素,并将气象因素做标准化处理后作为模型的初始输入量,在Spyder软件上对预测模型进行仿真验证。最后将上述模型与单一输入因素进行比较,结果显示充分考虑气象因素的预测结果明显优于单一因素,仿真结果证明该模型具有较好的预测精度。

In view of the fact that the existing photovoltaic power prediction technology fails to fully consider the meteorological factors and insufficient extraction characteristics,the prediction accuracy is low.Based on the deep learning theory,a photovoltaic power generation prediction based on the improved LSTM network method is proposed.According to the characteristics of long short-term memory neural networks,the general calculation process is derived from the recurrent neural networks,and the superiority and feasibility of the prediction method are expounded.A photovoltaic power generation rate prediction model based on improved long short-term memory(LSTM)network is proposed.The model fully considers the over-fitting problem caused by neural network optimization,and introduces RMSProp algorithm to obtain the optimal loss function value of the model to ensure the best prediction results.Considering a variety of meteorological factors that affect the power generation of photovoltaic power generation,and standardizing the meteorological factors as the initial input of the model,the prediction model is simulated and verified on the Spyder software.Finally,the above model is compared with a single input factor.The results show that the prediction result of meteorological factors is obviously better than the single factor.The simulation results show that the model has better prediction accuracy.