多维意象下产品造型满意度预测模型研究  

Research on the prediction model of product modeling satisfaction under multi-dimensional imagery

作　　者：孙利[1] 覃忠志 吴俭涛[1] 袁思琪 王巧玲 程永胜[2] SUN Li;QIN Zhong-zhi;WU Jian-tao;YUAN Si-qi;WANG Qiao-ling;CHENG Yong-sheng(School of Art and Design,Yanshan University,Qinhuangdao 066004;School of Design and Creativity,Kah Kee College,Xiamen University,Zhangzhou 363105)

机构地区：[1]燕山大学艺术与设计学院,河北秦皇岛066004 [2]厦门大学嘉庚学院设计与创意学院,福建漳州363105

出　　处：《机械设计》2023年第1期127-134,共8页Journal of Machine Design

基　　金：河北省科学技术研究与发展计划项目(1921820D);河北省社会科学基金资助项目(HB20YS003);福建省社会科学规划项目(FJ2021CO98);国家社会科学基金艺术学资助项目(2021BG02536)。

摘　　要：为均衡有效地满足用户多维复合意象的情感需求,提升多维意象下产品造型与用户满意度的匹配程度,提出一种结合熵权法与BP神经网络的产品造型满意度预测模型。以老年代步车为例,建立了产品造型元素集和多维意象集,采用语义差分法获取了样本评价值,并采用熵权法计算多维意象权重;将样本造型元素进行编码作为第1层BP神经网络模型的输入变量,将多维意象加权评价值作为第1层输出变量和第2层BP神经网络输入变量;将用户满意度评价值作为第2层BP神经网络输出变量,并通过K-fold交叉验证法训练和测试双重BP神经网络预测模型,验证结果显示:预测模型满意度MSE值小于0.01,表明该模型能有效映射多维意象下产品造型与满意度间的隐性关联。运用该模型对300个新方案进行预测,快速决策出最佳方案,为设计人员提供参考。In order to meet the user requirements on multi-dimensional perceptual imagery, improve matching degree of product modeling and user satisfaction under multi-dimensional imagery, a product modeling satisfaction prediction model was proposed based on entropy weight analysis and BP neural network. Taking the old scooters as an example, the morphological feature element set and typical image set were established. Semantic difference method was used to obtain sample evaluation value. Entropy method was used to calculate multi-dimensional imagery weight. The sample morphological feature elements encoding was used as the input variable of the first-layer BP neural network model. The weighted evaluation value of multi-dimensional imagery was used as the first-layer output viable and the second-layer BP neural network input variable. The satisfaction evaluation value was used as the second-layer neural network output variable. The dual BP neural network prediction model was trained and test by K-fold cross-validation method. The test results showed that the MSE value of the prediction model was less than 0.01,indicating that the model could effectively predict the implicit correlations between product modeling and satisfaction under multi-dimensional imagery. The model was used to predict 300 new schemes, and the optimal scheme could be quickly decided to provide references for designers.

关 键 词：产品造型设计 满意度预测模型 熵权法 BP神经网络 多维意象 

分 类 号：TB472[一般工业技术—工业设计]