论文部分内容阅读
针对支持向量机(SVM)在短期负荷预测中,根据经验选取参数导致预测精度下降的问题,提出一种基于布谷鸟搜索算法(CSA)优化SVM的短期负荷预测新方法(CSA-SVM)。先以历史负荷、温度、湿度等属性构成训练样本集的输入向量作为SVM的输入,以负荷值作为输出,建立SVM预测模型;再根据训练误差,以CSA对SVM中惩罚因子和核参数进行寻优;最后,按照CSA寻优获得的最优参数建立基于CSA-SVM的预测模型并开展短期负荷预测。实际负荷数据试验显示,相较于SVM模型、粒子群(PSO)优化SVM模型、BP神经网络模型,CSA-SVM具有更高的预测精度,能够满足电力系统短期负荷预测精度需求。
Aiming at the shortcoming of support vector machine (SVM) short-term load forecasting, the parameter selection based on experience leads to the decrease of prediction accuracy. A new short-term load forecasting method (CSA-SVM) based on cuckoo search algorithm (CSA) is proposed. Firstly, the input vector of training sample set is constructed with the attributes of historical load, temperature and humidity as the input of SVM, and the load value is taken as output to establish SVM prediction model. Then according to the training error, CSA is used to search for the penalty factor and nuclear parameter in SVM Finally, the CSA-SVM-based forecasting model and the short-term load forecasting are established according to the optimal parameters obtained through CSA optimization. Compared with SVM model, PSO-optimized SVM model, BP neural network model and CSA-SVM show higher prediction accuracy than the SVM model, which can meet the demand of power system short-term load forecasting accuracy.