摘要
针对短期负荷预测领域中常用的人工神经网络和支持向量机算法存在泛化性能较差,参数选择敏感等缺点,将随机森林回归算法引入电力系统短期负荷预测中,提出了一种基于聚类分析与随机森林的短期负荷滚动预测模型。首先利用K-means聚类算法识别用户用电行为习惯并将用电数据分成多个样本集,然后采用随机森林对每个样本集进行训练,生成对应不同用电行为模式下的预测模型,接着计算待测日与聚类簇中心的相似程度,找到待测日所属的预测模型,不断采集待测日每个时刻最新的负荷数据更新模型的输入,最后滚动预测出各时刻点的负荷数据。仿真算例表明本算法能明显改善预测精度,且具有较强的鲁棒性。
In the field of short-term load forecasting,the common artificial neural network（ANN）and support vector machine（SVM）have the disadvantages of poor generalization performance and being sensitive to parameter selection,so this paper introduces the random forest regression algorithm into the short-term load forecasting,and proposes a short-term load rolling prediction model based on cluster analysis and random forest（K-RF）. Firstly,use K-means clustering algorithm to identify the user＇s habits of electricity consumption and divide the electricity data into multiple sample sets.Then each sample set is trained in a random forest to generate a predictive model corresponding to different electricity consumption patterns. Then calculate the degree of similarity between the date to be predicted and the center of the cluster cluster,find the prediction model to which the day to be predicted belongs,continuously update the input of the model using the latest load data at each moment of the day to be predicted,and finally roll forward the load data at each time point.The simulation example shows that the algorithm can obviously improve the prediction accuracy and has strong robustness.
出处
《智能城市》
2018年第9期9-11,共3页
Intelligent City
