Responsible Workload Sequence Anomaly Detection with Bi-View Feature Fusion

Chunhong Liu, Dehao Wang, Li Duan, Kun Wang, Junna Zhang

Abstract: Workload sequence anomaly detection is a key technology for Artificial Intelligence for IT Operations (AIOps), credible and accurate detection results can effectively improve Operation and Maintenance (O&M ) efficiency. However, the change patterns of cloud platform workload sequences are complex and diverse. This makes it difficult to comprehensively capture the sensitive features of these sequences. Meanwhile, during anomaly detection, the results are transmitted between different departments in the enterprise’s AIOps. As a result, important parameters are at risk of being maliciously tampered with. To address this problem, this paper proposes a blockchain-based responsible workload sequence anomaly detection method with bi-view feature fusion. The method uses CNN to extract the local features of the sequence and Transformer to capture the global dependencies of the sequence, and fuses the extracted local and global features to accurately characterize the workload sequence. Then, the single classification hypersphere model is trained in the composed feature space to obtain the anomaly detection model. In the process of anomaly detection parameters and results feedback to the O&M personnel, a responsible detection environment based on blockchain is established. We combine on-chain and off-chain operations, detection parameters and results are recorded on the blockchain, while the anomaly detection process is carried out off-chain. In this paper, three benchmark datasets are used to test the effectiveness of the model, and a generalized performance test is performed using real cloud platform monitoring log datasets. The experimental results show that the performance of the proposed anomaly detection model is better than the latest anomaly sequence detection methods, and the proposed blockchain-based responsible detection scheme is feasible.