基于随机森林和改进局部预测的短期电力负荷预测

Short term load forecasting method based on random forests and improved local predictor

为提高短期电力负荷预测的精度,提出结合集成学习和局部预测思想的负荷预测新方法。首先,构建随机森林模型,从训练数据集中进行有放回随机抽样,对其中的决策树进行预训练。然后,采用改进的局部预测方法,将电力负荷时间序列嵌入到高维相空间中,通过动态时间规整算法确定待预测样本的邻域,从中选取相似样本对随机森林预训练模型进行微调,以提供更准确和可靠的预测结果。对随机森林、线性回归、K最邻近法、支持向量机等负荷预测方法进行了比较。结果表明:随机森林模型具有更好的预测精度和稳定性。采用所提出的改进局部预测方法进行微调后,模型的预测精度得到进一步提高。

In order to improve the accuracy of short-term power load forecasting,a new method combining integrated learning and local predictor is proposed.Firstly,a random forest-based model is built and pre-trained by a sample drawn with replacement from training dataset.Then,an improved local prediction method is used to embed the electricity load time series into a high-dimensional phase space.The neighborhood of the samples to be predicted is determined by using the dynamic time warping algorithm.Similar samples are selected to fine tune the pre-training random forest model to provide more accurate and reliable prediction results.Load forecasting methods such as random forest,linear regression,k-nearest neighbor method and support vector machine are compared by real data.The results show that the random forest-based model has better prediction accuracy and stability.The prediction accuracy of the model is further improved by local predictor.