基于Spiking神经网络的光伏系统发电功率预测  

Power Generation Forecasting for Photovoltaic System Based on Spiking Neural Network

作　　者：陈通[1] 孙国强[1] 卫志农[1] 李慧杰[2] CHEUNG KWOK W 孙永辉[1] 

机构地区：[1]河海大学能源与电气学院,南京210098 [2]阿尔斯通电网技术中心有限公司,上海201114 [3]美国阿尔斯通电网技术公司,华盛顿98052

出　　处：《电力系统及其自动化学报》2017年第6期7-12,44,共7页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金资助项目(51277052;51107032;61104045)

摘　　要：为了提高光伏系统发电功率预测的精度,本文提出一种基于Spiking神经网络(SNN)的预测模型。该神经网络采用精确脉冲时间的编码方式,更接近真实的生物神经系统,具有强大的计算能力。考虑季节类型、天气类型和大气温度等主要影响因素,该模型采用灰色关联分析法选取相似日。本文应用实际光伏发电系统的数据分别对基于SNN、BP人工神经网络(BP-ANN)和支持向量机(SVM)的预测模型进行测试和评估。预测结果表明:SNN预测模型相比于BP-ANN和SVM模型有较高的预测精度和较强的适用性,可以为光伏系统发电功率预测提供一种可行方法。A forecasting model based on Spiking neural network （SNN） was proposed in this paper to improve the forecasting accuracy of power generation from photovoltaic system （PVS） . This neural network uses temporal encoding scheme with precise time of spikes, which is closer to the real biological neural system, thus it has powerful computing capability. Considering the main influencing factors such as season types, weather types and atmospheric temperature, the proposed model uses grey correlation analysis to select similar days. The data from a practical PVS were adopted to test and evaluate the forecasting models based on SNN, back propagation artificial neural network （BP-ANN） and support vector machine （SVM）, respectively. The forecasting results reveal that compared with BP-ANN and SVM models, the SNN model has a relatively higher forecasting accuracy and a more robust applicability, which can provide a feasible way to forecast the power generation from PVS.

关 键 词：光伏系统 SPIKING神经网络 SpikeProp算法 相似日选择算法 发电功率预测 

分 类 号：TM615[电气工程—电力系统及自动化]