论文部分内容阅读
在有约束(尤其是非线性约束)的微分方程组曲线族求解及最佳化设计计算中所广泛采用的各种惩罚函数法,均因存在高时耗的选代过程,很难为大型数值积分型子优计算所接受。尤其严重的是,对某些约束条件若惩罚不当,就会干扰调优“走向”,导致“假优”的出现。如采取通过随机采集计算终止点的数值解对指定解(给定的终界约束条件)的散布误差,用于随机回归修正相应初值,并对指定解步长过算引起的漂移误差,进行负步长回积修正的方法,就可使计算以最少的时耗(一次回归、一步回积即可)获得指定解的满意结果。这种方法具有很好的通用性,对于大型寻优计算,例如以弹道族计算为基础的战术导弹总体参数最佳化设计计算,更能显示其优越性,且能解决用罚函技术难以解决的问题,可供工程实用。
In the constrained (especially non-linear constraints) differential equations of the family of solutions and optimization of the design and calculation of the widely used method of penalty function, both because of the existence of high time-consuming substitution process, it is difficult for large numerical integral type Suboptimal calculations accepted. In particular, if some of the restraints are not properly punished, they will interfere with the “direction” of adjustment and lead to the emergence of “fake”. For example, by randomly collecting the end point of the numerical solution of the specified solution (given the final constraint) of the distribution error, for the random regression to correct the corresponding initial value, and the designated solution over the calculation of the drift caused by drift error Negative step backfill correction method, you can make the calculation with the least time-consuming (a regression, a step back to the product) can be obtained satisfactory results of the specified solution. This method has good versatility. For large-scale optimization calculation, for example, optimization and calculation of overall parameters of tactical missile based on ballistic family calculation can better show its superiority and can solve the problem of using penalty letter The problem, for engineering and practical.