基于小生境遗传算法优化串级神经网络的台区理论线损预测

台区作为电力系统的最末端部分,节点数目多、网络架构复杂,导致不易精确计算其理论线损问题。在海量实际低压台区运行历史数据的基础上,首先研究影响台区线损的相关因子,建立台区线损特征指标集合。然后利用皮尔逊相关系数分析方法从中提取出高损关键特征指标作为线损预测模型输入。再融合集成学习和梯度提升树思想,建立基于串级BP(back propogatiow)神经网络的台区线损预测模型,并采用小生境遗传算法对预测模型的初始参数进行优化从而提高模型训练效率。最后通过实际数据算例仿真,并与其他预测方法结果进行比较,验证所提台区线损预测模型具有优秀的训练效率和预测泛化能力。

Theoretical Line Loss Forecast Based on Cascade Neural Network Optimized by Niche Genetic Algorithm

As the terminal part of power system, low-voltage distribution network has the characteristics with large number of nodes, and complex network structure, leading to the difficulty of calculating the theoretical line loss accurately. Based on real operational data, the characteristic index set of the line loss was established based on the analysis on the relevant factors affecting the line loss of low-voltage distribution network firstly. Then, Pearson correlation coefficient analysis method was applied to extract the key characteristics of high loss as the input of line loss forecast model. Combined with the idea of ensemble learning and gradient lifting tree, the line loss forecast model of low-voltage substation area based on cascade BP neural network is proposed here, with initial parameter optimized by niche genetic algorithm to improve the training efficiency of forecast model. Compared with other forecasting algorithms, simulation result based on realistic data shows the proposed forecast model for low-voltage distribution network line loss has excellent training efficiency and forecast generalization ability.