基于数据驱动的汽车形态特征进化预测方法

Vehicle form feature evolutionary prediction based on data-driven

针对传统设计方法因依靠定性经验知识而存在的不足,提出一种基于历史数据的汽车形态特征进化趋势预测方法。提取汽车形态特征并将其数值化,获得进化初始序列,使用灰色理论、人工神经网络和马尔科夫链的组合分析模型解决序列样本量小且震荡波动等问题。使用数据替换法和灰色模型初步得到形态特征拟合值,以该拟合值作为输入、实测值作为输出训练BP神经网络,得到形态特征进化预测结果,进而使用马尔科夫链消除由系统随机性产生的误差,得到最终预测结果。以某品牌汽车侧面轮廓线为例论证了该方法的可行性。

To overcome the defeats of traditional design methods due to relying on the qualitative empirical knowledge,a method of vehicle form feature evaluation prediction was proposed.The vehicle form features were extracted and quantized to obtain the initial evolution sequence,and a novel forecasting model based on grey theory,backpropagation neural network and Markov chain was proposed to solve the problem of small sample and data volatility.By using data replacement method and grey model,the fitted values of form features were obtained preliminarily,which was taken as input to train BP Neutral Network（BPNN）with the measured value as output.The evolutionary prediction result of form feature was received,and Markov chain was used to eliminate the errors caused by system randomness,thus the final prediction results were obtained.Validation of the proposed method was performed with the prediction of lateral contour line of a certain brand automobile was taken as an example to verify the effectiveness of the proposed method.