随着科学的发展及电力电子技术的不断提高，大量高精密仪器仪表被广泛使用，随之而来的各种电能质量问题严重影响着电力系统的供电安全和稳定运行，这就对电力系统的电能质量提出了更高层次的要求。电压暂降作为电能质量的一种，由于其发生频率高、波及面广、造成影响较大等问题，已经成为目前电能质量的重点研究方向。 通过分析国内外研究现状，发现在电压暂降检测方面，S变换作为一种新型时频分析方法，较其他方法具有窗函数可变，时间、频率和幅值分辨率高，对噪声不敏感等优点，正逐渐在相关领域被推广应用；而在用于电压暂降扰动信号识别的分类器方面，虽然BP神经网络在并行处理、自学习、自组织和容错性能上具有显著优势，但由于其本身存在收敛速度慢、容易陷入局部最优等缺陷。在对BP神经网络的优化方面，大量文献已经表明粒子群（PSO）算法对BP神经网络的优化有着很好效果。故本文提出了基于S变换和PSO-BP神经网络的电压暂降检测与识别方法。 首先利用MATLAB/SIMULINK为仿真平台，搭建简单配电网仿真模型，获得6种不同类型的电压暂降扰动源模型，包括线路短路故障模型、感应电动机启动模型、变压器投入模型、多重故障模型、感应电动机重新启动模型、感应电动机启动和变压器投入共同作用模型。其次对6种扰动信号进行S变换获得信号的基频幅值曲线和基频斜率变化曲线，完成对电压暂降幅值和起止时刻的检测，并根据扰动信号的不同特点利用S变换提取所需特征值。最后将提取到的特征值作为训练样本和测试样本输入到PSO-BP神经网络中，完成电压暂降扰动源的分类识别。 仿真结果表明基于S变换的检测方法可以较为准确的检测到电压暂降的幅值和起止时刻；利用PSO-BP神经网络对6种电压暂降扰动源分类识别的准确率高达94.44%，证明了基于S变换和PSO-BP神经网络的方法可以有效的检测和识别电压暂降扰动信号。

With the development of science and the continuous improvement of power electronic technology, a large number of high precision instruments and meters are widely used, and the various power quality problems seriously affect the safety and stable operation of power system., which poses higher-level requirements of the power quality of power system. Voltage sag as a kind of power quality, due to its high frequency of occurrence, wide spreading, great influence and so on, has become a key research direction of the current power quality. Through the analysis of the present situation of domestic and foreign research, it is found that in terms of voltage sag detection, S-transform as a new time-frequency analysis method, having advantages of changeable window function with high time, frequency and amplitude resolution and insensitivity to noises over other methods, is gradually popularized and used in related fields. In respect of voltage sag disturbance signal recognition, although the BP neural network has conspicuous advantages in parallel processing, self-learning, self-organization and fault tolerance performance, it has the shortcoming of slow convergence speed and is easy to encounter local optimization. In the optimization of BP neural network, a lot of literatures have shown that the particle swarm optimization （PSO） algorithm has brought about very good effects in the optimization of BP neural network. Therefore, this paper put forward voltage sag detection and recognition method based on S-transform and PSO-BP neural network. The first step is utilizing MATLAB/SIMULINK as a simulation platform to build a simple simulation model of power distribution network so as to get six different types of voltage sag disturbance source model, including the short circuit fault model, induction motor starting model, transformer input model, multiple fault model, induction motor restarting model as well as the interaction model of induction motor starting and transformer input. The second step is getting the fundamental frequency amplitude curves and fundamental frequency curves of the slope through the S-transform of six kinds of disturbance signal, so as to detect the amplitude of voltage sag and the start-stop time and use S transform to extract the needed characteristic values according to different characteristics of the disturbance signal. The final step is inputting the extracted characteristic values into PSO-BP neural network as the training sample and testing sample to complete the recognition of disturbance source of voltage sag. The simulation results show that detection method based on S-transform can more accurately detect the amplitude and start-stop time of voltage sag. The classification recognition accuracy of the six kinds of voltage sag disturbance sources is up to 94.44% by using PSO-BP neural network, proving that method based on S transform and PSO-BP neural network can effectively detect and identify the voltage sag disturbance signal.