Runtime WCET Estimation Using Machine Learning

Abstract

Accurate task execution time estimation is vital for efficient and dependable operation of safety-critical systems. However, modern automotive functions' complexity challenges conventional estimation methods. To address this, we propose a novel technique that combines execution time and job sequence data using a multi-layer perceptron (MLP) neural network. Leveraging MLP's capabilities, our approach achieves impressive 99.7% prediction accuracy with a mere 38.33 μs latency. Integrating our technique into safety-critical systems optimizes resource allocation and scheduling, enhancing performance and reliability. Importantly, our method extends beyond automotive systems, finding potential in diverse safety-critical domains. By precisely estimating task execution time, we enhance operational efficiency and decision-making in complex systems.