基于遗传算法优化反向传播网络的汽车造型评价

Genetic Algorithm Optimized Automotive Styling Evaluation Based on Backpropagation Neural Network

摘要:

针对汽车造型评价存在由于主观性高而导致可靠性低的问题，运用遗传算法原理对评价方法进行了优化。通过遗传算法的优化，基于反向传播（BP）神经网络的汽车造型评价，减小了评价误差。通过问卷调研构建数据集，使用汽车的18个评价对象作为输入，整车评价作为输出，创建了融合遗传算法的反向传播（GA-BP）网络结构，并在Matlab中进行了仿真预测。研究结果表明，经过优化的BP神经网络预测值的相对误差均值由6.7 %下降至1.7 %，显著提升了汽车造型评价的可靠性，具有更好的预测能力和实际应用潜力。

Abstract:

This paper addresses the challenge of low reliability in automotive styling evaluation， which is often affect by high subjectivity by optimizing the evaluation method using principles of genetic algorithms. By integrating a genetic algorithm（GA） with a backpropagation （BP）neural network ， the evaluation error is effectively reduced. A dataset was constructed through questionnaire surveys， with 18 automotive styling evaluation factors as inputs and the overall automotive evaluation as the output. A GA-BP network structure was developed， and simulation predictions were conducted in MATLAB. The results show that the mean relative error of the optimized BP neural network decreases from 6.7 % to 1.7 %， significantly enhancing the reliability of automotive styling evaluation. This demonstrates the enhanced predictive accuracy and practical applicability of the proposed method.