AI-Driven Performance Degradation Identification via Self-Supervised Spatiotemporal Graph Modeling in Microservice Systems

This paper proposes a self-supervised performance degradation identification model to address the challenges of high-dimensional heterogeneous data, complex dependency structures, and dynamic non-stationarity in large-scale microservice architectures. The model takes multi-source monitoring data as input and first performs semantic alignment among different metrics through multidimensional feature embedding and projection layers. An adaptive dynamic graph convolutional network is then employed to capture the topological dependencies and interaction features among service nodes, constructing time-varying structural representations. In the temporal modeling stage, a gated recurrent unit-based embedding mechanism is introduced to jointly characterize long-term dependencies and local fluctuations of performance evolution, while a residual fusion structure enhances the stability of feature propagation. To improve feature discrimination under unsupervised conditions, the model adopts a contrastive learning optimization strategy and utilizes a temperature adjustment mechanism to strengthen the distinction between positive and negative samples in the latent space, enabling adaptive aggregation and recognition of degradation patterns. Furthermore, multiple hyperparameter sensitivity experiments are conducted to systematically evaluate the effects of learning rate, residual coefficient, temperature parameter, and monitoring sampling interval on model performance. Experimental results show that the proposed model outperforms mainstream methods in accuracy, precision, recall, and F1-score, achieving efficient and stable identification of performance degradation in complex microservice systems under unsupervised settings, thus providing a practical solution for intelligent operations and maintenance.