电力系统量测最优配置及状态估计算法研究

发布时间：2018-05-18 00:08

本文选题：状态估计 + 量测装置配置　；参考：《华北电力大学(北京)》2017年硕士论文

【摘要】：状态估计作为电力系统潮流计算、短路电流计算和稳定性分析的基础,是对从现场获取来的数据的第一次处理。随着现代科技的发展,现场采集数据的设备种类越来越多,设备精度越来越高,同时,配电网相对于输电网结构更加复杂,大量的分布式电源、环网甚至微网,出现在配电网中。因此,本文在新型量测设备PMU和传统量测设备在含环配电网中的混合配置方案上进行了深入研究。含环配电网中的量测装置优化配置,实际上就是用花费较少的量测系统来获得较高精度的状态估计结果,同时必须保证系统的可观性。本文以新兴的和声搜索算法为框架,以PMU和传统量测装置的成本较低和量测精度较高为目标,得到多目标pareto解集前沿,用获取得到的量测进行状态估计,以状态估计的估计精度来衡量目标函数中的量测精度。文章在仿真过程中,采用的是IEEE33关闭联络开关来实现含环配电网,并在关闭9节点和15节点之间、8节点和21节点之间的联络开关的系统上进行了仿真实验,得到了理想的配置方案。状态估计有着成熟的算法框架,在静态状态估计中使用较多的是最小二乘状态估计,但是由于量测量和状态量之间是非线性关系,最小二乘状态估计需要借助高斯牛顿反复迭代,计算成本较高。随着神经网络这种专门解决非线性问题的技术的出现,使用神经网络训练一个网络用于计算状态估计已经成为一种可能。本文就是利用稀疏自编码器(SAE)和前馈(BP)神经网络结合,同时使用粒子群算法(PSO)调整网络参数,最终在IEEE14的基础上训练得到一个计算状态估计的网络,和传统的最小二乘状态估计相比,本文使用提出的神经网络训练得到的网络计算状态估计不仅可以节省每一次的计算时间,还可以得到更高的状态估计精度。
[Abstract]:As the basis of power flow calculation, short-circuit current calculation and stability analysis, state estimation is the first processing of the data obtained from the field. With the development of modern science and technology, there are more and more kinds of equipment to collect data on the spot, and the precision of the equipment is more and more high. At the same time, the distribution network is more complicated than the transmission network structure, and a large number of distributed generation, ring network and even microgrid, Appear in the distribution network. Therefore, the hybrid configuration scheme of new measuring equipment PMU and traditional measuring equipment in distribution network with ring is studied in this paper. In fact, the optimal configuration of measuring devices in the distribution network with ring is to obtain a high precision state estimation result with a less cost measurement system, and at the same time, the observability of the system must be guaranteed. In this paper, the new harmonic search algorithm is used as the framework, the cost of PMU and the traditional measuring device is low and the measurement precision is high, the frontier of multi-objective pareto solution set is obtained, and the obtained measurements are used to estimate the state. The measurement accuracy in the objective function is measured by the estimation accuracy of state estimation. In the course of simulation, the IEEE33 switch is used to realize the distribution network with ring, and the simulation experiment is carried out on the system of closing the connection switch between 9 and 15 nodes, 8 nodes and 21 nodes. An ideal configuration scheme is obtained. State estimation has a mature algorithm framework. The least square state estimation is used more frequently in static state estimation, but because of the nonlinear relationship between the measurement and the state quantity, The least square state estimation needs to be iterated repeatedly by means of Gao Si Newton, and the calculation cost is high. With the emergence of neural network, which is a special technique to solve nonlinear problems, it is possible to use neural network to train a network to calculate state estimation. In this paper, we combine sparse self-encoder (SAE) with feedforward (BP) neural network, and use particle swarm optimization algorithm (PSO) to adjust the network parameters. Finally, we train a computational state estimation network on the basis of IEEE14. Compared with the traditional least square state estimation, the proposed neural network training can not only save the time of each calculation, but also obtain higher state estimation accuracy.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM930

【参考文献】