Neural Networks And Genetic Algorithms Applied Research In Game Design

All the time, games have been used as experimental test beds for many areas of artificial intelligence. Environment Game is a computer game that can simulate the process of realistic enterprise's investment. In the process of virtual investment, enterprise not only need consider the income question, namely economic performance question, but also should consider the influence to surrounding environment because of the development of enterprise, on the contrary, the change of the environment will also influence the development of enterprise, that is to say that according to long view the continuous development problem of enterprise. In order to make game simulate the process of realistic enterprise's investment well, and improve the game's interesting, this thesis utilizes Neural Network and Genetic Algorithm to find its appropriate parameters and to mimic the player's behavior. The main work as follows:Parameter optimization based on Genetic Algorithm. In the course of investment, though the risk and income are not always in direct ratio relations, but generally speaking, the larger is risk, the larger is income, and the greater is risk, the heavier is perhaps loss certainly. The relation is reflected with parameter of different investment strategy of game. If the parameters is not appropriate, the difference of different investment strategy can not be reflected, and investors are not able to realize the difference each other while using different investment strategy too, certainly game will loss its attraction. This thesis uses the optimization theory of Genetic Algorithm, consider times that top player has been changed by other players,variance of the total points,state of the environment,total times that participated players have been rewarded as the parameters of fitness function. Through finding the appropriate income and loss parameter of different investment strategy in game, so as to reflect better the relations between risk and income in investment, thus make game playing more interesting.Learn the investment behavior based on Neural Network. When using Genetic Algorithm to optimize the parameter in the Environmental Game, Because the evaluation of each chromosome is to be executed by the results of game playing, and normally the number of chromosome is more than 100, this mean at least 100 times is needed in order to execute GAs for scores of generations (one time needs 2 hours generally), It is difficult to find so patient investors, so we propose using the neural network to learn the behavior of a investor, and make use of neural network model to replace investor in GAs, thus can avoid many the same process of simulating investment, make the idea that use genetic algorithm to find appropriate parameters be realized. Some experimental results are shown, from the results, we can observe that neural network can mimic the ways in which player play rather well.Code implement. The Environment Game system is made by author with Visual Basic 6.0. This system has good user interfaces ,easy to use, and has vividly showed the importance of environmental protection. Especially the player can be Neural Network Model in this system.