基于遗传算法优化的Elman网络时滞补偿研究

Time Delay Compensation Employing Elman Network Optimized by Genetic Algorithm

神经网络时滞补偿算法在时滞补偿方面一般应用于特定激励下固定时滞的补偿问题,适用范围有待进一步改善。针对主动质量阻尼器控制系统,考虑不同地震波和时滞的影响,本文提出了一种基于遗传算法(GA)优化的神经网络时滞补偿方法。该方法通过跟踪理想状态的结构响应,采用GA对Elman神经网络进行初始权重和阈值的优化训练,并与经典的移相法和泰勒一阶法的时滞补偿效果进行了对比分析。研究结果表明:理想状态下AMD可以显著减小结构的位移和加速度响应。在175 ms、225 ms和275 ms的大时滞情况下,结构位移峰值、层间位移角、加速度均值和峰值指标与理想状态的指标差值均分别在19.4%、15.5%和22%之内,适用于不同地震波和时滞下的结构主动控制。

Neural network time delay compensation algorithm is generally applied to fixed time delay compensation problems under specific excitation,and its application scope needs to be further improved.Considering the influence of different seismic waves and time delays,a neural network time-delay compensation method based on genetic algorithm(GA)optimization was proposed for active mass damper control system.By tracking the ideal structural response,the Elman neural network was trained with GA to optimize the initial weight and threshold,and the delay compensation effect was compared with the classical phase shift method and Taylor's first-order method.The results showed that AMD could significantly reduce the displacement and acceleration response of the structure under ideal conditions.In the case of 175 ms,225 ms and 275 ms,the difference of the peak displacement,interstory drift angle,mean acceleration and peak acceleration index were within 19.4%,15.5%and 22%,respectively,which are suitable for active structural control under different seismic waves and time delays.