基于数据挖掘的SVM短期负荷预测方法研究

支持向量机方法已成功地应用在负荷预测领域,但它在训练数据时存在数据处理量太大、处理速度慢等缺点。为此提出了一种基于数据挖掘预处理的支持向量机预测系统,引用在处理大数据量、消除冗余信息等方面具有独特优势的数据挖掘技术,寻找与预测日同等气象类型的多个历史短期负荷,由此组成具有高度相似气象特征的数据序列,将此数据序列作为支持向量机的训练数据,可减少数据量,从而提高预测的速度和精度,克服支持向量机的上述缺点。将该系统应用于短期负荷预测中,与单纯的SVM方法和BP神经网络法相比,得到了较高的预测精度。

Study on Forecasting Approach to Short-term Load of SVM Based on Data Mining

The support vector machine (SVM) has been successfully applied to the load forecasting area, but it has some disadvantages of very large data amount and slow processing speed. Using advantages of the data mining technology in processing large data and eliminating redundant information, a SVM forecasting system based on data mining preprocess was proposed to search the historical daily load with the same meteorological category as the forecasting day and to compose data sequence with highly similar meteorological features. Taking the new data sequence as the training data of SVM, the data amount was decreased and the processing speed was improved. This approach has achieved greater forecasting accuracy comparing with the method of single SVM and BP neural network.