A Transformer-LSTM-SVR hybrid model for AI-driven emotional optimization in NEV embedded interior systems

Area of Science:

• Engineering
• Human-Computer Interaction
• Artificial Intelligence

Background:

• The new energy vehicle (NEV) market is rapidly expanding, shifting focus from functionality to user emotion for enhanced experience and brand differentiation.
• Existing research on emotion optimization in NEV interiors lacks efficient nonlinear modeling methods.
• Understanding user emotional satisfaction is crucial for designing compelling and competitive NEVs.

Purpose of the Study:

• To develop an efficient nonlinear modeling method for predicting user emotions in NEV interiors.
• To address the complex relationship between interior design attributes and user emotional satisfaction.
• To provide practical tools for designers to create emotion-oriented and sustainable NEV interiors.

Main Methods:

• A hybrid Transformer-LSTM-SVR model was proposed, integrating attention mechanisms and temporal modeling.
• The Transformer module captured higher-order interactions among multidimensional design parameters.
• A Long Short-Term Memory (LSTM) network enhanced time-series feature capture, with its outputs fused with Transformer representations and fed into a Support Vector Regression (SVR) module.

Main Results:

• The proposed model significantly outperformed benchmark models (SVR, PSO-SVR, PSO-RF, BPNN, GBR) in prediction accuracy, improving it by 12.7% to 23.4%.
• The synergistic integration of LSTM's temporal attention and Transformer's global context modeling improved robustness against noisy user feedback data by 18.9% compared to traditional knowledge engineering (KE) methods.
• Interpretability analysis revealed key design feature differences among user groups.

Conclusions:

• The hybrid Transformer-LSTM-SVR model offers a viable and robust approach for accurately predicting users' emotional needs in NEV interior design.
• This method enhances model robustness and prediction accuracy, addressing limitations of traditional models in handling dynamic, nonlinear relationships.
• The findings support the development of emotion-oriented and sustainable interior design strategies, boosting NEV market competitiveness.