深度信念网络在水火电力系统一次调频能力预测中的应用  

Research on Application of Deep Belief Network in Primary Frequency Modulation Capability of Hydrothermal Power System

作　　者：崔伟 李武璟 王聪 薛晨 段建东[2] 魏嵬[3] CUI Wei;LI Wujing;WANG Cong;XUE Chen;DUAN Jiandong;WEI Wei(Northwest Control Center, State Grid Corporation of China, Xi’an 710048, China;School of Electrical Engineering, Xi’an University of Technology, Xi’an 710048, China;Computer Science and Engineering College, Xi’an University of Technology, Xi’an 710048, China)

机构地区：[1]国家电网公司西北分部,西安710048 [2]西安理工大学电气工程学院,西安710048 [3]西安理工大学计算机科学与工程学院,西安710048

出　　处：《西安交通大学学报》2021年第3期99-108,共10页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金资助项目(51877174)。

摘　　要：为了准确地预测电力系统一次调频能力,提出了一种采用深度信念网络的水火电力系统一次调频能力预测方法。以新英格兰39节点系统和某电力公司水火电力系统2016—2019年历史数据作为仿真算例,将系统频率、有功负荷、功率缺额、扰动类型、负荷水平、发电机组惯性时间常数、备用容量与总备用容量作为网络的输入特征值,为网络参数训练提供数据支撑;使用无监督预训练与有监督参数微调相结合来训练网络参数,构建深度信念网络,输出系统一次调频的出力调节曲线,预测水火电力系统一次调频能力。与传统的网络模型进行了对比,结果表明:在新英格兰39节点系统中,最大功率补偿量平均相对误差和均方根误差分别为1.49%和4.04 MW;在电力公司水火电力系统中,最大功率补偿量平均误差和均方根误差分别为1.18%和18.3 MW,均小于循环神经网络的平均误差1.41%和均方根误差21.6 MW。所提深度信念网络的逐层训练解决了传统网络预测方法的参数优化问题,能够避免陷入局部最优和训练周期过长,应可在突发性故障时为调度中心后续制定防护控制故障策略提供信息支撑。In order to accurately predict the primary frequency regulation capability of the power system,a method for predicting the primary frequency regulation capability of hydrothermal power systems using a deep belief network is proposed.System frequency,active load,power vacancy,disturbance type,load level,generator set inertia time constant,reserve capacity and total reserve capacity are used as input characteristic values of the network to provide data support for network parameter training.The combination of unsupervised pre-training and supervised parameter fine-tuning is used to train the network parameters,build a deep belief network,output the output adjustment curve of the system’s primary frequency modulation to predict the primary frequency modulation ability of the hydrothermal power system.Compared with the traditional network model,the results show that:in the New England 39-bus system,the average relative error and root mean square error of the maximum power compensation are 1.49%and 4.04 MW,respectively.In the hydrothermal power system of the electric power company,the average error and the root mean square error of the maximum power compensation are 1.18%and 18.3 MW,respectively,which are smaller than the average error of 1.41%and the root mean square error of the cyclic neural network of 21.6 MW.Compared with the traditional network prediction methods,the layer-by-layer training of deep belief networks solves the problem of parameter optimization and avoids falling into local optimality and long training periods.This model can provide information support for the dispatch center to formulate protection and control failure strategies in the event of sudden failures.

关 键 词：一次调频能力 深度信念网络 水火电力系统 预测 

分 类 号：TM734[电气工程—电力系统及自动化]