基于灰色关联和混合神经网络的光伏功率预测

Photovoltaic Generation Forecasting Model Based on Gray Correlation Degree and Hybrid Neural Network

为了提升光伏发电功率预测模型的精度和增强其对多变天气的适应能力,提出了采用基于天气预报的灰色关联相似日样本选取与混合神经网络相结合的光伏发电功率预测模型.相似日选取以辐射强度的影响因素为依据,采用晴天理论太阳辐射强度、空气污染指数、云量、湿度4个变量,通过灰色关联选出与预测日较为接近的历史数据构成样本子集.建立混合神经网络,对选出的样本子集进行天气要素扩充并训练模型,代入预测日特征向量完成预测.为检验该模型的精确性和鲁棒性,通过实例与常见BP神经网络模型进行预测结果对比,显示了新模型在光伏发电功率预测的良好应用前景.

In order to improve the precision and the adaptive capability to weather variations of thephotovoltaic generation forecasting, a model based on gray correlation is adopted to select similar days andhybrid neutral network. The similarity degree in net radiation, AQI, cloud amounts and humidity, which in原fluence solar radiation, are obtained with grey correlation analysis method in the process of choosing train原ing samples. According to chosen training samples provided by gray correlation, hybrid neutral network istrained and evaluated. A practical example is given to demonstrate the accuracy and robustness, and theresult shows our model has promising applications.