Design and optimization of a test case generation algorithm for real-time embedded systems based on adaptive Q-Learning
Testing real-time embedded systems requires intelligent strategies that balance test coverage, timing constraints, and resource limitations. The traditional test case generation methods, such as random testing and conventional Q-learning, often fail to adapt to dynamic workloads and maintain real-ti...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | en |
| Published: |
Springer Nature Limited
2026
|
| Subjects: | |
| Online Access: | http://ir.unimas.my/id/eprint/51519/1/s10515-026-00598-w.pdf http://ir.unimas.my/id/eprint/51519/ https://link.springer.com/article/10.1007/s10515-026-00598-w https://doi.org/10.1007/s10515-026-00598-w |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Testing real-time embedded systems requires intelligent strategies that balance test coverage, timing constraints, and resource limitations. The traditional test case generation methods, such as random testing and conventional Q-learning, often fail to adapt to dynamic workloads and maintain real-time responsiveness. To address these limitations, an automated test case generation method based on adaptive Qlearning (AQL) is proposed in this study; the method is specifically designed for real-time embedded software. The proposed method introduces dynamic parameter
adjustment and adaptive time-window control schemes to optimize multiple objectives including test coverage, resource utilization, and empirical real-time performance under varying workloads. Experiments were conducted on an ATV dashboard-embedded platform, and AQL was compared with random testing (RT) and traditional Q-learning (QL). The results demonstrated that AQL achieved significant
performance improvements: the statement coverage level reached 92%, the average CPU utilization rate decreased to 63%, and under experimental loads, the deadline miss rate remained below 2% across all scenarios (e.g., 1.2% under high CPU load), while faster response times were achieved. A statistical analysis (ANOVA, p < 0.01) confirmed the significance of these improvements. In summary, the proposed AQL method provides an efficient and scalable intelligent solution for testing embedded systems in real time. Its feedback-driven adaptive structure effectively overcomes
the static limitations of the conventional reinforcement learning approaches, offering both academic innovation and practical potential for testing intelligent software
in resource-constrained real-time environments. |
|---|