中文摘要：

位移是重力坝变形监测的重要物理量,对其进行准确预测是确保大坝安全运行的前提.目前已经有许多预测方法,但是大部分方法都存在易落入局部极小、收敛速度慢和收敛对初值敏感等问题.为解决或减小这些问题,提高预测精度,将多种群遗传算法（MPGA）与反向传播（BP）神经网络算法结合起来,提出一种适用于重力坝变形预测的多种群遗传神经（MPGA-BP）网络算法.实例计算证明,该算法能够有效克服BP神经网络收敛速度慢、易出现局部极小值的缺点和遗传算法的早熟收敛问题,在进行重力坝变形预测中具有更高的收敛性和精度.

英文摘要：

Displacement is an important physical quantity of gravity dam deformation monitoring and its accurate prediction is the premise of ensuring safe operation of the dam. There are already a lot of predic- tion methods at present, but most of them have problem such as easy to fall into local minimum, slow to converge, and sensitive to the initial value. In order to resolve or reduce these problems and improve the prediction accuracy, the multiple population genetic algorithm （MPGA） was combined with the back-prop- agation （BP） neural network algorithm to present algorithm suitable for prediction of gravity dam deform- ation, namely the multiple population genetic algorithm back-propagation （MPGA-BP） neural network al- gorithm. It is shown by calculation of examples that this algorithm can effectively overcome the problem in BP neural network algorithm and the problems of premature convergence of GA, and has a higher conver- gence and accuracy in gravity dam deformation prediction.