论文部分内容阅读
提出一种多维函数的通用进化逼近方法 .通过构造一类结合采样函数和样条函数优点的基本函数族 ,提出一种单调函数逼近方法 ,并借助采样函数的有界变差特点 ,将该方法推广到一般函数情形 ,这两种函数的逼近都可通过遗传算法完成 .该方法的优点在于可以简单一致地推广到更高维函数的逼近 ,并使逼近复杂度与维数成线性关系 ,降低学习算法难度 .试验表明 ,该方法是有效的 .基于文中单调函数逼近技术提出的一种新的决策策略学习方法已成功地应用于某移动机器人控制器设计中 .
A universal evolutionary approximation method for multidimensional functions is proposed.Through the construction of a family of basic functions that combine the advantages of sampling and spline functions, a monotone function approximation method is proposed. Based on the bounded variation of the sampling function, The generalization of these two functions can be accomplished by genetic algorithm.The advantage of this method is that it can be generalized to the approximation of higher dimensional functions simply and consistently and the linear relation between the approaching complexity and the dimension is reduced And the difficulty of learning algorithm.Experiments show that this method is effective.A new decision strategy learning method proposed based on the monotone function approximation technique has been successfully applied to the design of a mobile robot controller.