复式Kohonen和改进BP网络的微网光伏发电预测

Micro-grid photovoltaic power generation prediction based on compound Kohonen and improved BP neural network

针对微电网光伏发电波动性和间歇性问题,提出一种基于复式Kohonen和改进BP网络的微网光伏发电预测方法。利用首层有监督Kohonen网络进行天气聚类,分化复杂天气因素的非线性度,减小对二层预测模型的影响。由天气类别而采用改进的BP网络模型,提高模型预测精度,增强模型的在线实时性能。利用现场数据对网络进行训练和预测,仿真结果有效,复现了光伏输出功率变化。

This paper proposes a compound method for forecasting PV power based on Kohonen and improved BP neural network as a solution to the volatility and intermittency problem of photovoltaic power generation in micro-grid. This method functions by using the supervised Kohonen network on the first-level for weather clustering to weaken the nonlinearity behind complex weather factors,and reducing the impact on the second-layer prediction model; in the prediction layer,as is required by each weather category,adopting the correspondingly improved BP network modeling to improve the online real-time performance of the model while enhancing the prediction accuracy of the model; and ultimately train and forecast the network using photovoltaic power plant site data. The simulation provides an effective reproduction of the dynamic variation rule of PV output power.