针对电力负荷数据中存在的不平衡小类样本导致负荷预测精度不高问题，提出一种基于K-means-SyMProD-PCA数据预处理及NPMA-LSSVM模型的电力负荷预测方法。通过改进的K均值(K-means)方法根据电力负荷特性对其进行预分类，并构建分类标签作为输入特征；针对电力负荷分类后的样本类别不平衡问题，采用基于概率分布合成小类样本(synthetic minority based on probabilistic distribution,SyMProD)方法扩充小类样本数据以平衡样本类别；为了消除具有重复信息的特征，基于主成分分析(principal component analysis,PCA)方法提取电力负荷主要特征；最后建立最小二乘支持向量机(least square support vector machine,LSSVM)电力负荷预测模型，采用非线性惯性因子和多项式变异的蜉蝣算法对模型参数进行优化，以提高负荷预测精度。分别采用第9届电工杯建模大赛数据和扬中市2015年1 443家企业的用电量数据作为验证数据，结果表明，结合K-means-SyMProD-PCA负荷数据预处理，NPMA-LSSVM预测模型有效降低了电力负荷预测误差，能够较好地解决不平衡小类样本情况下的中短期电力负荷预测问题，具有一定的适用性。
英文摘要:
Aiming at the problem of low accuracy of load forecasting caused by unbalanced minority class samples in power load data, this paper proposes a power load forecasting method based on K-means-SyMProD-PCA data preprocessing and NPMA-LSSVM model. The improved K-means method is used to find a suitable cluster center to classify the load data, and construct a classification label as the input feature. For the sample category unbalance problem after power load classification, synthetic minority based on probabilistic distribution (SyMProD) method is used to expand the number of minority samples to balance the sample category. In order to eliminate the features with repeated information, principal component analysis (PCA) is used to extract the main features of power load. Finally, the least square support vector machine (LSSVM) load forecasting model is established, and the improved mayfly algorithm with nonlinear inertia factor and polynomial variation is used to optimize the model parameters to achieve accurate load forecasts. Using the 9th Electrician Cup Modeling Contest data set and the power consumption data set of 1443 enterprises in Yangzhong City in 2015 as the verification data, the results show that the load data processed by K-means-SyMProD-PCA input into the NPMA-LSSVM model can reduce the forecasting error effectively, it can better solve the problem of short and medium-term power load forecasting in the case of unbalanced minority class samples, which has certain applicability.