基于BP神经网络代理模型的交互式遗传算法

针对交互式遗传算法存在用户评价噪声和审美疲劳的问题，提出一种基于误差反向传播神经网络用户认知代理模型的交互式遗传算法。通过构建用户评价噪声模型，形成进化个体动态模糊区间适应值，以刻画用户认知随机不确定性；在用户认知确定阶段历史评价信息基础上，构建误差反向传播神经网络代理模型，给出一种新的适应值估计策略；通过度量均方误差，实现代理模型的管理与更新。将所提方法应用于蜡染风格图案设计，并与其他典型算法对比。结果表明，该方法能够有效优化进化个体适应值质量、降低用户审美疲劳。

Interactive Genetic Algorithm Based on BP Neural Network and User Cognitive Surrogate Model

To solve the problem of evaluation noise and user fatigue in the Interactive Genetic Algorithm(IGA),an error Back Propagation(BP)neural network user cognitive surrogate model is proposed to improve the performance of IGA.First,because user perceptions are uncertain,a model for noises is built to form the adaptive value with individual dynamic fuzzy interval.Then user cognitive surrogate model based on BP neural network is constructed by historical evaluation information when the user’s cognition is certain and a new adaptive value estimation strategy is presented.By measuring Mean Squared Error(MSE),the management and update of the agent model are realized.This method is applied to selection system of pattern design with batik style,the results show that it can effectively optimize the quality of adaptive value with individual and reduce user fatigue.