基于差分进化优化RBF网络的小电流接地多判据选线方法

Multi-criteria Line Selection Method for Small Current Grounding Based on Differential Evolution Optimization of RBF Network

本文采用RBF网络为基础构建小电流接地系统的选线模型。通过种群更新机制、进化参数自适应等方法提高差分进化算法(DE)的训练能力,进而应用DE优化RBF网络参数。采用ATP仿真软件建立小电流接地配电系统的仿真模型,采样不同故障条件下的零序电流信号,提取稳态和暂态的多种故障特征分量输入优化好的RBF神经网络,结果表明经过DE训练的RBF网络收敛速度快,输出误差小,选线正确率高于传统RBF神经网络,且不受各种故障条件的影响。

A line selection model is built based on the radial basis(RBF) neural network in this paper. Methods such as the population update mechanism and evolutionary parameter adaptation are used to improve the training ability of the differential evolution algorithm(DE), and DE is used to optimize the RBF network parameters. A simulation model of small current grounding distribution system is established by the ATP simulation software. The zero-sequence current signals under different fault conditions are sampled, and various steady-state and transient-state fault feature components are extracted and input into the optimized RBF neural network. The results show that the trained RBF network by DE has fast convergence speed, small output error, and the correct line selection rate is higher than the traditional RBF neural network, which is not affected by various fault conditions.