基于PSO-SVR模型的光伏功率预测研究  

Photovoltaic Power Prediction Based on PSO-SVR Model

作　　者：许鸣吉 孙歌 徐焰 赵彬 陈佳瑜 冯陈佳 XU Mingji;SUN Ge;XU Yan;ZHAO Bin;CHEN Jiayu;FENG Chenjia(State Grid Shibei Power Supply Company,SMEPC,Shanghai 200072,China)

机构地区：[1]国网上海市电力公司市北供电公司,上海200072

出　　处：《电力与能源》2023年第2期150-156,共7页Power & Energy

摘　　要：光伏发电功率预测对电网调度工作有着重要的指导意义。以其光伏电站为研究对象,首先分析季节类型、天气类型和气象要素等因素对光伏功率的影响,对历史数据进行分类,选择合适的特征作为训练集;其次,针对支持向量回归机(SVR)模型对内部参数依赖性较高这一现象,设计粒子群优化(PSO)算法对SVR模型的惩罚系数、不敏感损失系数和尺度参数进行寻优,建立基于PSO-SVR方法的光伏功率预测模型;最后,通过实际案例,选择BP神经网络、SVR和PSO-SVR等方法进行罗泾灰场的光伏功率预测并作比较,验证了所提方法能较好地跟踪实测曲线,对光伏功率的预测结果更接近实测值。The prediction of photovoltaic power generation has important guiding significance for power grid dispatching.In this paper,a certain photovoltaic power station is taken as the research object.Firstly,the influences of season,weather and meteorological elements on photovoltaic power are analyzed,the historical data are classified,and appropriate features are selected as the training set.Secondly,in view of the high dependence of support vactor regression(SVR)model on internal parameters,particle swarm optimization(PSO)is designed to optimize the penalty coefficient,insensitive loss coefficient and scale parameter of SVR model,and a PV power prediction model based on PSO method is established.Finally,through a practical case,BP neural network,SVR,PSO-SVR and other methods are selected to predict and compare the photovoltaic power of the Rojing ash yard,which verifies that the proposed method can track the measured curve well,and the predicted result of photovoltaic power is closer to the measured value.

关 键 词：光伏功率预测 特征 支持向量回归机模型 粒子群优化算法 

分 类 号：F426.61[经济管理—产业经济] TP183[自动化与计算机技术—控制理论与控制工程]