基于改进BP网络的装甲装备机动性能评估模型

Evaluation Model for Flexibility Effectiveness of Armored Weapon System Based on Improved BP Neural Network

收敛速度慢、易陷入局部极值是传统的BP神经网络难以避免的问题,最终可能导致网络训练失败。在量化装甲装备机动性能指标的基础上,采用遗传算法对BP神经网络权值进行优化,用自适应梯度下降法对传统BP神经网络进行训练,从而建立装甲装备机动性能评估模型,并通过二次训练得到评估值。仿真结果表明该改进网络收敛速度明显优于传统网络,能有效避免局部极值问题。

inevitable problems in traditional BP neural networks like slow convergence and easy in local extremum may finally lead to abortive net training. On the basis of quantifying the index of armored equipment＇s flexibility performance, we ,optimized the weight of BP neural network by using genetic algorithm; trained traditional BP neural network through adaptive gradient descent algorithm, and established evaluation model for flexibility performance of armored equipment; while trained network to obtain evaluation value. Simulation results show the convergent speed of the improved network is much better than that of traditional neural network, and can effectively avoid local extremum.