基于SVR-LSTM-BP的分布式光伏短期出力预测方法研究

A research on short-term distributed photovoltaic power prediction model based on SVR-LSTM-BP

提出一种分布式光伏短期出力组合预测方法,以BP神经网络耦合支持向量回归(Support Vector Regression,SVR)和长短期记忆(Long Short-Term Memory,LSTM)神经网络实现组合预测。首先分别构建两个单一模型:以高斯径向基函数为核函数的支持向量回归模型和三层长短期记忆神经网络,并分别预测,再通过三层BP神经网络将前两个单一模型的预测结果耦合并输出,以提高预测的准确度。利用江苏某光伏发电装置采集的真实数据进行仿真验证,得出结论:SVR-LSTM-BP模型的准确度与SVR模型相比有显著提高,而与LSTM模型接近,稳定性则比LSTM模型有一定提高。

A new combined prediction model for short-term distributed photovoltaic output is proposed,using support vector regression(SVR),long short-term memory neural network(LSTM)and backpropagation neural network(BP).Two simple models using SVR with Gaussian radial basis function and 3-layer LSTM are constructed,respectively.Then,to increase the prediction accuracy,the outputs of the two simple models are combined by a 3-layer BP neural network.Numerical experiments based on the real data of a photovoltaic power station in Jiangsu Province shows that the SVR-LSTM-BP model has a significantly improved prediction accuracy than that of the SVR model,which is close to that of the LSTM model.The stability of the SVR-LSTM-BP model is slightly improved than that of the LSTM model.