新建牵引变电所的负荷预测及变压器容量优化配置

为了获得新建牵引变电所的负荷情况并校验优化所内牵引变压器的配置容量，将高斯混合模型用于牵引变电所实测数据聚类，然后引入神经网络对新建牵引负荷进行匹配分类. 依据聚类和分类结果，结合概率密度及蒙特卡洛抽样方法，实现新建电气化铁路牵引负荷的预测. 根据热传递原理和相对老化计算，建立新建牵引变电所牵引变压器温升与寿命损失的差分方程模型，对新建牵引变电所的牵引变压器容量进行优化配置. 通过对大量牵引变电所实测数据的分析，聚类后伪-F统计量达12.81，匹配分类后伪-F统计量进一步上升至12.90，表明本文聚类分类方法效果良好. 通过牵引变压器建模，将算例中变压器容量利用率从60%提高到96%，即使考虑安全裕度适当提高安装容量也能使容量利用率达到75%，实现了变压器容量的优化，充分利用了变压器的温度指标和寿命损失. 

Load Forecasting and Transformer Capacity Optimization for Newly-built Traction Substation

In order to predict the load of a newly-built traction substation and optimize the traction transformer capacity, Gaussian mixture model was employed for clustering the measured data of traction loads, and then the neural network is introduced to match and assign the new traction load. According to the results of clustering and matching, the new load process for new electrified railway is evaluated by the probability density and Monte Carlo method. Then based on the theory of heat transfer and the calculation of aging rates, the difference equation models of the temperature rise and loss of life is proposed to optimize the capacity of new transformer. After analyzing large amount of measured data from traction substations, the pseudo-F value of the clustered data is 12.81, and rises to 12.90 when new load is matched and assigned, indicating that the clustering and assigning methods are effective. The capacity utilization ratio in case study rises from 60% to 96% by modeling the traction transformers. Even with safety margin, the capacity should be expanded and the capacity utilization ratio is also 75%, which achieves the goal of capacity optimization and makes the best of the temperature rise and loss of life.