【作者】 王振树；

【导师】 李林川；

【作者基本信息】 天津大学 ， 电力系统及其自动化， 2009， 博士

【摘要】 本论文研究电力负荷动特性的支持向量机建模方法、模型结构中核函数的选取、贝叶斯证据框架的参数优化选择、精确数据的获取以及计及负荷模型的无功优化算法等关键性问题。该研究解决了难以对复杂的用电负荷结构进行负荷建模,负荷模型结构不灵活、泛化能力不强、模型不准确的问题,以及考虑负荷模型的复杂的综合无功优化问题。论文的主要内容如下:负荷模型和参数的准确度对电力系统数字仿真结果影响很大,提出了利用贝叶斯证据框架的支持向量机(SVM)负荷建模方法。该方法将负荷节点看着“黑箱”,应用SVM回归理论建立节点处的非机理负荷模型结构,选用高斯径向基核函数、采用贝叶斯证据框架的三个准则进行模型参数的优化选择。该方法能够灵活地改变模型结构、对参数进行辨识和优化,建立了能够反映负荷特性的非机理负荷模型。广域测量系统(WAMS)具有异地高精度和高密度同步测量、高速通信等特点,能够实时地提供大量反映系统特性的动态数据,提出了利用WAMS信息和SVM的负荷建模。仿真结果表明,模型待辨识参数少,计算速度快,泛化能力好,所建模型精确、能够较准确地描述实际负荷。对于大扰动事件,提出了利用电力故障录波系统信息(PFRMS)和SVM的负荷建模。创建的PFRMS满足了负荷建模精确数据来源的要求。利用故障录波信息重演负荷特性的暂态过程,通过实测曲线和负荷模型仿真曲线的比较,进行负荷动特性模型的校验。针对电网中负荷大小及其变化趋势对无功优化的影响,提出的电力系统无功优化以全调度周期网损最小,改善系统电压质量以及控制设备动作次数最少为目标,优化中计及负荷变化、利用粒子群与模拟退火相结合算法的综合无功优化。对几个测试系统进行了仿真计算表明,该算法原理简单易实现,计算效率高。更多还原

【Abstract】 This dissertation studied the following issues: power load modeling based on support vector machine, selection of kernel function in model structure, optimization of model parameters by using Bayesian evidence framework, access to accurate data and reactive power optimization algorithm considering load model. The study solved the problem of establishing model for complex power load. Traditional load model is inflexible and inaccurate, and didn’t have good generalization ability. All these problems had been solved in this dissertation. It also studied the synthetic reactive power optimization considering load model. The main contents of the dissertation are as follows:The accuracy of load model and its parameters have great effect on power system digital simulation results. This dissertation presented a support vector machine (SVM) load modeling method by using Bayesian evidence framework. This method regarded load bus as“black box”, and applied SVM regression theory to establish non-mechanism load model. Gaussian radial basis kernel function was used. It adopted three inference levels of Bayesian evidence framework to optimize the model parameters. This method could change model structure flexibly, also could identify and optimize model parameters. It established a non-mechanism load model that can reflect load characteristics.The wide area measurement system (WAMS) ,which is mainly characterized by synchronous measurement and high-speed communication with high-precision and high-density among differernt places, could provide real-timely massive dynamic data that reflect the system characteristics, therefore, this dissertation proposed using the SVM and information acquired WAMS to establish load model. Simulation results showed that this model has less identification parameters. It owned rapid calculation speed and good generalization ability, so it could describe the actual load characteristics more accurately. As for large disturbances, this dissertation used SVM and information of power fault recording and monitoring system (PFRMS) to establish load model. The established PFRMS could meet the needs of providing accurate data. Utilizing the recorded fault data to replay the transient process, the dynamic load model could be verified by the comparison between measured curves and simulation ones.Considering that load quantity and its chang trend produce certain influence to reactive power optimization, this dissertation presented a reactive power optimization which aimed at minimizing network loss during the whole electric power dispatching period,improving voltage quality and reducing action number of control devices. Taking load changes into account,particle swarm algorithm(PSO)was combined with simulated annealing algorithm(SA)to optimize reactive power synthetically. Simulation results for several testing systems demonstrated that this cooperative algorithm was simple and can be easily realized with high efficiency.更多还原