Design and Implementation of NPC AI based on Genetic Algorithm and BP Neural Network

Abstract

In real game scenes, there may be a problem that it is impossible to collect a large amount of data to train NPC's correct behavior. This paper implements a method of training the neural networks to control game NPC behavior based on an improved genetic algorithm. This method optimizes the weights of the fixed network structure through the genetic algorithm, realizes the self-evolution of the neural network, and improves the fitness function and the selection and crossover mutation method in the traditional genetic algorithm, so as to be suitable for game production. Testing in two real game scenes shows that game NPC can acquire intelligent behavior capabilities using this algorithm.