Predicting and optimizing pure electric vehicle road noise via a locality-sensitive hashing transformer and interval analysis

Abstract

Pure electric vehicles (PEVs) lack engine noise; thus, the overall noise level within vehicle cabins are reduced. However, due to the absence of engine noise, previously overlooked noise sources are accentuated and detrimentally affect interior noise quality. Road noise is a predominant PEV noise source and significantly contributes to middle- and low-frequency interior noise levels. A novel approach combining data-driven methodologies and uncertainty analysis to predict and optimize vehicle road noise is proposed. To predict the frequency-domain characteristics of road noise, a refined attention mechanism based on the transformer model with a locality-sensitive hashing algorithm is introduced to enhance efficiency and ensure high accuracy. An interval vector optimization method using interval representations of parameter uncertainty is devised to strengthen the robustness and efficacy of the road noise optimization results. The proposed method is validated through a PEV road test, and the optimized noise conditions demonstrates an improvement exceeding 2 dB.