| 引用本文: | 张程熠,唐雅洁,李永杰,高强,江全元.适用于小样本的神经网络光伏预测方法[J].电力自动化设备,2017,37(1): |
| ZHANG Chengyi,TANG Yajie,LI Yongjie,GAO Qiang,JIANG Quanyuan.Photovoltaic power forecast based on neural network with a small number of samples[J].Electric Power Automation Equipment,2017,37(1): |
|
| 摘要: |
| 基于神经网络的短期光伏预测方法通常需要大量训练样本，对于新投运的光伏电站，历史运行数据的不足使得常规短期光伏预测方法难以应用。针对该问题，提出一种适用于小样本的双层神经网络单步光伏预测方法。根据光伏发电各环节影响因素的解耦特性，将常规单层神经网络拆分为双层网络，使每层网络具有简化的结构；用单步预测代替多步预测，降低神经网络的输入输出维数；基于统计分析，将天气影响因素有效整合到预测模型中，简化输入输出之间的映射关系。使用实际数据对所提光伏预测模型进行训练和验证，结果表明，所提方法可有效减少对训练样本数量的需求，同时保证预测的准确度。 |
| 关键词: 光伏发电 短期功率预测 小样本 神经网络 单步预测 |
| DOI：10.16081/j.issn.1006-6047.2017.01.016 |
| 分类号: |
| 基金项目:浙江省重大科技专项计划项目(2014C01006) |
|
| Photovoltaic power forecast based on neural network with a small number of samples |
|
ZHANG Chengyi1, TANG Yajie1, LI Yongjie1, GAO Qiang2, JIANG Quanyuan1
|
|
1.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China;2.Power Dispatch and Control Center of State Grid Zhejiang Electric Power Company, Hangzhou 310007, China
|
| Abstract: |
| Since the conventional short-term PV (PhotoVoltaic) power forecast based on neural network normally needs a large number of samples for training, it is not suitable for the newly-built PV station with insufficient historical data, for which, a method of PV power forecast based on dual-layer neural network is proposed for a small number of samples. The conventional single-layer neural network is divided into two layers according to the decoupling characteristics of different influencing factors of PV generation and each layer has a simplified structure. Instead of multi-step forecasting, single-step forecasting is adopted to lower the input and output dimensions of neural network. Based on the statistical analysis, the weather factors are effectively integrated into the forecast model to simplify the mapping between input and output. Practical data are applied to train and verify the proposed model and results show that, the proposed method can effectively reduce the demand of training samples while guarantee the forecast accuracy. |
| Key words: photovoltaic generation short-term power forecast a small number of samples neural networks single-step prediction |
