利用退火回归神经网络极值搜索算法求纳什均衡解

针对如何解算n人非合作的动态博弈对策中的纳什均衡解问题,提出一种利用退火回归神经网络极值搜索算法解算纳什均衡解的方法．在动态博弈对策问题中,将每个竞争者视为一个代价函数,利用此算法可以使每个代价函数均收敛于其最小值,从而获得此对策的纳什均衡解．此算法不限制代价函数的具体形式,同时由于摒弃了正弦激励信号,解决了一般极值搜索算法中存在的输出量“颤动”现象和控制量来回切换问题,改善了系统的动态性能．

Nash Equilibrium Solution by Extremum Seeking Algorithm Based on Annealing Recurrent Neural Network

An algorithm is proposed to solve the Nash equilibrium solution for an n-person noncooperative dynamic game by an annealing recurrent neural network for extremum seeking algorithm(ESA).In noncooperative dynamic game,each player is defined as a cost function.Each cost function will fast converge to its minimum point by the algorithm proposed,so that the Nash equilibrium solution can be obtained.ESA combined with the annealing recurrent neural network does not limit the formation of the cost functions or make use of search signals such as sinusoidal periodic signals,which can solve the "chatter" problem of the output and the switching problem of the control law in the general ESA,and improve the dynamic performance of the system.