基于K-Means和GBRT的分布式光伏中短期发电量预测  

Medium-short term power generation prediction of distributed photovoltaic based on K-Means and GBRT

作　　者：刘刚 闵金 宋伟 高迪 栗辉[2] Liu Gang;Min Jin;Song Wei;Gao Di;Li Hui(State Grid Jibei Electric Power Company,Beijing 100056,China;School of Automation,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]国网冀北电力有限公司,北京100056 [2]北京科技大学自动化学院,北京100083

出　　处：《能源与环保》2023年第3期210-215,221,共7页CHINA ENERGY AND ENVIRONMENTAL PROTECTION

基　　金：宁夏回族自治区重点研发计划重大(重点)项目(2019BFG02009);中央高校基本科研业务费专项资金资助项目(FRF-BD-10-012A)。

摘　　要：光伏发电量与太阳辐照强度有直接关系,但太阳辐照强度并不能直接获取,而是需要依据多种气象指标间接计算,所以基于太阳辐照度来预测光伏发电量的方法难以应用。现有的光伏发电量预测方法一般是根据气象条件对单个用户或电厂的实时功率进行建模,而随着光伏用户数目的剧增,为每一用户建立单独预测模型的做法显然不可取。为此,基于K均值算法(K-Means)和梯度提升回归树(GBRT)提出了一种针对多用户建模预测的方法。首先分别将用户和天气进行聚类,然后再分别对同一天气类型下日均实际发电量相近的用户建立发电量预测模型,从而为同时预测多个用户的发电量提供了新思路。通过均方根误差(RMSE)对模型预测效果进行评估,通过一系列对比试验,发现GBRT预测模型的精度较高,在测试集上的RMSE值低至3.17,表明该预测方法具有一定的可靠性。Photovoltaic power generation has the most direct relationship with solar irradiation intensity,but the intensity of solar irradiation cannot be directly obtained,but needs to be calculated indirectly based on various meteorological indicators,so it is difficult to predict photovoltaic power generation based on solar irradiance.The existing photovoltaic power generation forecasting methods generally model the real-time power of a single user or power plant based on meteorological conditions.With the sharp increase in the number of photovoltaic users,it is clearly not advisable to establish a separate prediction model for each user.Therefore,based on K-Means algorithm(K-Means)and gradient promotion regression tree(GBRT),this paper proposes a method for modeling and forecasting for multiple users.Firstly,users and weather are clustered,and then power generation prediction models are established for users with similar power generation capacity under the same weather type,thus,it provides a new way to predict the power generation of multiple users at the same time.This paper evaluates that model prediction effect through root mean square error(RMSE),through a series of tests,it turns out that the GBRT prediction model is relatively high,and the value of the RMSE on the test set is low to 3.17,indicating that the prediction method has certain reliability.

关 键 词：K-MEANS 气象 发电量 多用户 GBRT 

分 类 号：TU4[建筑科学—土工工程]