基于深度学习的超短期光伏精细化预测模型研究

摘要

光伏发电系统的超短期功率预测对电网调度的计划安排及光伏发电系统的优化运行具有重要意义.机器学习、人工智能领域的技术进步为精细化分析光伏功率预测影响因素并提高光伏预测精度提供了有效途径.提出一种基于深度结构网络模型的光伏超短期功率预测方法,首先根据光伏发电系统的机理特征,分析深度学习算法处理光伏预测问题的可行性;然后提出基于深度学习算法的光伏功率预测模型,采用基于受限玻尔兹曼机的深度置信网络提取深层特征完成无监督学习过程,采用有监督BP神经网络作为常规拟合层获得预测结果;并立足于实际需求,建立含离线训练和在线预测的双阶段光伏发电预测系统,分析天气信息及历史信息的输入属性;最后利用光伏发电系统的实际运行数据进行仿真,验证算法准确性和有效性,通过比较深度结构是否包含无监督学习过程,说明其在预测中的重要性.%Accurate ultra short-term power forecasting of photovoltaic (PV) generation is very important for dispatching of power grid and the optimal operation of PV system.The progress of machine learning and artificial intelligence technology provides an effective way for refined analysis of PV power prediction factors and the improvement of forecasting accuracy.This paper proposes an ultra short-term power forecasting method for PV based on the deep structure network model.Firstly,according to the mechanism characteristics of PV generation system,we analyzed the feasibility of deep learning algorithm to deal with the PV power forecasting problem.And then,we presented PV power forecasting mode based on deep learning algorithm which learns the structure internal characteristics by deep belief network based on restricted boltzmann machine to extract deep features as unsupervised learning process,and used the supervised BP neural network as conventional fitting layer to obtain the forecasting results.In addition,based on the actual demand we established a two-stage PV power generation forecasting system including off-line training and on-line prediction,to analyze the input properties of the weather information and historical information.At last,we verified the validity and accuracy of the algorithm through the actual operating data of PV system,and explained its importance in forecasting through comparing whether the depth structure contained unsupervised learning process.