Abstract
With the continuous growth in the scale and complexity of modern information systems, achieving high-precision system state perception and efficient resource scheduling still faces significant challenges. Especially under conditions of dynamic loads, data loss, or sudden peaks, existing methods usually have problems such as limited classification accuracy, slow strategy response, and insufficient robustness. To address these issues, this study proposes a decision support model based on Convolutional Neural Network (CNN). Through an end-to-end mechanism, the model effectively integrates system state recognition and strategy generation, and constructs a closed-loop optimization and feedback framework to realize system-level resource scheduling optimization. Experimental results on two public datasets, Azure VM Workload Traces and Google Cluster Data, demonstrate the model’s effectiveness. The model achieves a state recognition accuracy of 96.2%, an F1 score of 0.961, an average policy response time of 135.5 milliseconds, and reduces scheduling failure rate from 6.1% to 2.2%, representing a 63.9% improvement. Furthermore, the model maintains over 90% accuracy and policy success rate even under challenging conditions such as data loss and sudden workload surges. These results confirm the model’s strengths in real-time performance, accuracy, and robustness, highlighting its strong potential for practical deployment.