Event and Slack Time Prediction With Deep Learning for Proactive Runtime Scheduling in Time-Triggered Systems
Dongchen Li, Daniel Chidiebere Onwuchekwa, Roman Obermaisser · 2025
Time-triggered systems are extensively employed in various safety-critical applications. Adapting to the changing environment by dynamically adjusting the schedule has become a critical research topic in this field, thereby enhancing system efficiency and reducing energy consumption. The two types of existing adaptive scheduling systems either require substantial resources or exhibit long adaptation delays. By integrating the advantages of both systems, proactive speculative scheduling systems are designed to address these issues. The key to achieving it lies in predicting events from environmental inputs and the historical system states. Based on the prediction results, the schedule can be proactively computed to respond to environmental changes. Our goal is to predict the event sequence, event source sequence, and slack time of all tasks in the next scheduling cycle based on the environmental inputs and system states in the current cycle. In this paper, we leverage two deep learning models, DA-LSTM and Priorest, originally applied in different domains, which are capable of predicting events and slack times in parallel. Based on our dataset and goal, the structures of the models have been modified. We designed a specialized loss function, weighted mean square error (WMSE), to enhance the prediction performance of slack time. The experimental results demonstrate that both models can predict events and slack time. Both models achieved approximately 73% coverage in event sequence prediction.