提出了一种基于最小二乘支持向量机(LS-SVM)和粒子群优化技术(PSO)相结合的电力负荷预测方法。以历史负荷数据、气象因素等作为输入,建立预测模型,对未来日最大负荷进行预测。该模型利用结构风险最小化原则代替传统的经验风险最小化,以充分挖掘原始数据的信息,并应用粒子群算法优化最小二乘支持向量机的参数,提高了预测模型的训练速度和预测能力。实际算例表明,使用上述方法进行电力负荷预测,具有良好的可行性和有效性,与BP神经网络法的预测结果相比,前者具有更高的精度和更强的鲁棒性。
In the paper,a method based on the least square support vector machine(LS-SVM) and PSO algorithm is proposed for power system load forecast.By this method,a daily load forecast model is developed by using structure risk minimization(SRM) instead of the traditional ERM to mine the original data for more information,adopting SVM parameters optimized by PSO algorithm to improve the training speed and forecast ability,and taking model inputs of historical loading and atmospheric data.Forecasting results show that this method is feasible and effective and that the robustness and forecast accuracy of the model are better than the method of BP neural network.