基于神经网络算法的发射场坪承载能力预测方法

为进一步提高发射场坪的快速选定能力，提出基于线性多元回归算法、反向传播（BP）神经网络算法和径向基函数（RBF）神经网络算法的发射场坪承载特性近似模型，对其承载能力进行快速预测。采用了引入参数敏感度的优化拉丁超立方采样方法，构建样本空间并分别建立起竖载荷和发射载荷下的算法预测近似模型；利用上述预测算法，通过某结构参数未知的场坪在起竖载荷下的动力学响应，预测其在发射载荷下的可能最大弯沉值，判断土基弹性模量未知的发射场坪承载能力。结果表明：RBF神经网络算法对起竖载荷和发射载荷下的场坪最大弯沉量均具有最优的预测性能，与有限元计算结果相对比，回归系数分别为0.941和0.983，模型精度较高，具有可信性；预测结果与仿真结果的平均误差为10.46%；对于承载强度较大的场坪，残差范围在±2 mm以内，对发射过程中的场坪选定决策具有参考价值。

The Prediction Method Based on Neural Network Algorithm for the Bearing Capacity of Launching Site

A rapid prediction method of the bearing capacity of launching site is proposed to improve the quick response ability of the missile launching，which is based on linear multiple regression algorithm，back propagation(BP) algorithm and radial basis function (RBF) algorithm. An optimized Latin hypercube sampling method with parameter sensitivity is applied to construct the sample space. The approximate models evaluated by different algorithms under different loads are established and proved to be effective. Evaluation algorithm of the bearing capacity of unknown launching site is established to predict the probable maximum deflection under launch load using dynamical response of the launching site under erection load. The results show that the RBF algorithm has the best prediction performance and the regression coefficients under erection and launch loads are 0.941 and 0.983.The average error between predicted and simulated results is 10.46%. For the launching site with higher bearing capacity，the residual error of the evaluation algorithm ranges in ±2 mm.