基于机器学习的光伏发电功率预测——以金华市为例

Photovoltaic power generation prediction based on machine learning-taking Jinhua City as an example

不可再生能源的大量消耗,对新能源发电提出更高的要求。国家倡导大力发展新能源发电,且光伏发电在新能源中占有非常重要的位置,然而光伏发电功率不稳定,因此实时准确预测光伏发电功率对新能源消纳和存储具有非常重要的意义。当前存在很多光伏发电功率预测算法,大量研究揭示深度学习方法能更准确地预测光伏发电功率。收集了金华市3个光伏发电站每5分钟时间窗口的发电功率,并通过网络爬虫得到相应的天气数据,再采用机器学习算法对各类天气的光伏发电功率进行预测。研究结果揭示深度学习算法(如双向长短期循环神经网络Bi-LSTM)比传统机器学习算法更好,也更稳定,且晴天和多云天气比阴雨天预测得更准确。

The massive consumption of non-renewable energy puts forward higher requirements for new energy power generation.The country advocates vigorously developing new energy power generation,and photovoltaic power generation occupies a very important position.However,photovoltaic power generation is unstable,so real-time and accurate prediction of photovoltaic power generation is of great significance for new energy consumption and storage.At present,there are many prediction algorithms for photovoltaic power generation,and a large number of studies have revealed that deep learning methods can more accurately predict photovoltaic power generation.This paper collects the power generation of three photovoltaic power stations in Jinhua City and the data are obtained for every 5 minutes.Furthermore,we obtains corresponding weather data through web crawlers,and then uses machine learning algorithms to predict photovoltaic power generation in various weather.Prediction results reveal that deep learning algorithms(such as bidirectional LSTM)are better and more stable than traditional machine learning algorithms.In addition,we find that,comparing with overcast and rainy days,it is more accurate to predict photovoltaic power generation for sunny and cloudy days.