| Automotive Software Abnormal Detection Method Based on Deep Learning |
| Qiujun Zhao, Ziwei Tian, Weinan Ju, Shuhua Zhou, Yu Su |
| Intelligent Communication Research Department, China Automotive Technology & Research Center Co. Ltd, CATARC Software Testing (Tianjin) Co. Ltd, No. 68, Xianfeng East Road, Dongli District, Tianjin, 300300, China |
|
|
|
Received: February 14, 2025; Revised: April 17, 2025 Accepted: April 17, 2025. Published online: July 4, 2025. |
|
|
|
| ABSTRACT |
|
With the rapid development of automotive intelligence and networking, the complexity and dynamism of automotive software systems have significantly increased, placing higher demands on software reliability. This paper proposes a new method for automotive software abnormal detection that integrates service grid and deep learning (DL) to effectively address the problems of high-dimensional, strongly correlated, and nonlinear behavior patterns in traditional abnormal detection methods. This method constructs a lightweight automotive service grid architecture, achieving transparent monitoring and flexible management of communication between services, providing comprehensive and multi-dimensional data support for abnormal detection. At the service grid control level, a fusion model of graph convolutional network (GCN) and recurrent neural network (RNN) is used for features extraction, and an end-to-end DL model is deployed to capture and deeply analyze telemetry data in real time, quickly identify, and effectively respond to various abnormal behaviors. The experimental results show that this method has achieved an accuracy rate of over 90% in abnormal detection of performance, safety, functionality, configuration, and compatibility, offering an effective means to ensure software reliability in the automotive intelligence and connectivity era. |
| Key Words:
Automotive software · Abnormal detection · Service grid architecture · Deep learning · Graph convolutional network · Recurrent neural network |
|
Corresponding Author.
Ziwei Tian , Email. 
Preview 
Full text via DOI 
Download Citation
Print
