关键词: 光伏发电预测, 深度学习, 支持向量机（SVM）, 长短期记忆网络（LSTM）, 多层感知器-深度信念网络（MLP-DBN）算法

Abstract:

Accurate prediction of photovoltaic（PV） power generation is of great significance for improving stability of power system, ensuring power quality, and optimizing grid operation. In order to solve the problems of low precision and poor performance of the existing PV prediction algorithms, and to utilize the ability of Multi-Layer Perceptron（MLP） to solve nonlinear problems and the advantages of Deep Belief Networks（DBN） to effectively process large amounts of complex data, a PV prediction algorithm combining MLP and DBN（MLP-DBN） is constructed. The basic idea is to make preliminary predictions using MLP model, then input the residuals between observed values and the predicted values into DBN model to forecast, finally modify MLP’s predictions by residual’s predictions. Using the simulation data of PV power generation, model’s predictive performance under different learning rates is explored, and the parameters are searched for optimal settings. Using performance indexes such as root mean square error, average absolute error, and determination coefficient, compared with traditional prediction algorithm Support Vector Machine（SVM） and deep learning algorithm Long Short-Term Memory（LSTM） with higher prediction accuracy, the simulation results show that the performance of MLP-DBN algorithm has been significantly improved, which can provide a high-precision and high-performance algorithm for PV power and effectively solve the problem of PV power prediction.

Key words: photovoltaic generation prediction, deep learning, Support Vector Machine（SVM）, Long Short-Term Memory（LSTM）, Multi-Layer Perceptron-Deep Belief Networks（MLP-DBN） algorithm