针对电能质量暂态扰动信号时频局部化信息量较广难以简洁灵活提取有效细微特征以及匹配追踪算法计算量大的问题，提出一种应用于电能质量扰动特征参量提取及压缩重构的匹配时频原子框架及其遗传优化改进算法。在Gabor过完备时频原子库离散基础上，采用匹配追踪方法(matching pursuit，MP)对扰动信号进行时频原子自适应分解，并通过遗传算法(genetic algorithm，GA)对时频原子参量进行优化计算，从而降低匹配追踪搜索过程的复杂度，获得最佳匹配电能质量扰动信号特征的时频原子参量化解析表示以及匹配特征重压缩构波形。算例仿真表明，该框架重构信噪比高达50 dB，均方误差数量级为0.001，能量恢复系数达到0.99以上，与小波(包)相比，具有更优良的压缩重构性能及多分辨能力，遗传优化时频参量的改进算法，基本保持了 MP 优良的压缩重构性能，计算复杂度缩减率为95.8%，算法收敛性得到提高，匹配扰动信号计算效率提高80~100倍，满足电能质量扰动分析要求。“,”A novel framework using Time-frequency Atom Decomposition, as well as its improved method optimized by genetic algorithm is proposed to decompose and reconstruct power quality disturbances and extract feature parameters of disturbances in this paper, In allusion to the problems that transitory power quality power quality disturbances is high in local time-frequency domain and is difficult to be extracted effective and fine features of the signals flexibly and sententiously,as well as the computation of matching pursuit is huge. Given a redundant Gabor dictionary of time-frequency atoms, we decompose a power disturbance signal into dominant atoms in the frequency-time distribution by Matching Pursuit, which are selected in order to best match the signal structure. In particular, the time-frequency atoms parameters are optimized by Genetic Algorithm to reduce the complex rate of finding the best atoms. And thus some best time-frequency atoms matching features of disturbance signals, and also its reconstruction waveform, which can be described in parametric analysis, are obtained. The simulation results show that using such a framework the signal to noise ratio (SNR) of reconstruction of matching feature of single disturbance can reach up to 50 dB, the order of magnitude of mean square error is 0.001, and energy recovery coefficient is up to 0.99. Relative to the decomposition based on wavelet package, such method is with more excellent signal reconstruction compression and multi-resolution analysis performance. Using the improved method optimized by genetic algorithm, the Computational complexity reduction rate is up to 95.8%, thus the efficiency of matched disturbance feature and convergence performance are improved further to meet the demand of power quality analysis.