基于改进混合密度网络的毁伤效应预测方法

Prediction Method of Damage Effects Based on Improved Mixture Density Network

提出一种基于改进混合密度神经网络的毁伤效应预测方法,解决了现有智能毁伤效应预测方法中仅能输出点预测结果,但难以量化毁伤效应预测结果的不确定性问题。采用鲁棒性更好的t分布作为混合分量,利用混合密度网络生成概率密度函数,以反映毁伤效应预测中的不确定性,并根据给定置信水平获得区间预测结果。仿真实验表明,获得的概率密度函数可以较为准确地拟合蒙特卡洛仿真模拟结果,与现有的毁伤效应预测方法相比,可以更好地指导作战筹划。

A damage effect prediction method based on an improved mixture density neural network was proposed.The method aimed to solve the problem that current intelligent damage effect prediction methods could only output point prediction results,but fail to quantify the uncertainty of damage effect prediction results.A more robust t distribution as a mixture component and a mixture density network was used to generate a probability density function that reflects the uncertainty in the prediction of damage effects.Interval prediction results could be obtained according to a given confidence level.Simulation experiments demonstrated that the probability density function obtained by the proposed method could more accurately fit Monte Carlo simulation results and better guide combat planning than existing damage effect prediction methods。