Anomaly detection is critical to ensure cloud infrastructures’ quality of service. However, due to the complexity of inconspicuous (indistinct) anomalies, high dynamicity, and the lack of anomaly labels in the cloud environment, multivariate time series anomaly detection becomes more difficult. The existing approaches are rarely effective in meeting these challenges. In this paper, we propose a novel convolutional adversarial model, convolutional-adversarial-training-based integrated anomaly detection with explanation framework (CAT-IADEF), for multivariate time series anomaly detection in the cloud. We adopt three convolutional neural networks to learn sequence features and adversarial training to amplify “slight” anomalies while enhancing the robustness of the model. The dynamic threshold is determined in real time by the peaks over threshold (POT) method to improve detection accuracy. In addition, anomaly explanation is also conducted efficiently by analyzing anomaly score vectors. Experiments with seven data subsets from various public datasets show that CAT-IADEF outperforms state-of-the-art methods. The average F1 score on the seven datasets is 0.907, which is 6.5% higher than the state-of-the-art model and up to 22.1% higher than the baseline method. Furthermore, the proposed anomaly explanation framework is also integrated into various models to verify its effectiveness on the experimental datasets.

The rapid expansion of multivariate time series (MTS) data has made deep learning a central tool for forecasting across industrial and scientific domains. As these architectures move into settings such as clinical decision support, industrial monitoring, and financial risk management, researchers need to examine not only predictive accuracy but also computational cost and the kinds of explanations that a model can support. This survey re-examines deep learning models for MTS forecasting through the requirements of efficiency and explainability. Our contributions are threefold: (1) We synthesize the literature through a dual-axis scalability view over sequence length L and variate dimension N , together with a three-question explainability taxonomy covering temporal importance, variate importance, and pattern decomposition. (2) We review dominant architectural families, including Transformer variants, frequency-domain methods, linear and channel-independent backbones, state space models, and foundation-style forecasters, to describe how different designs occupy different efficiency–explainability profiles under specific data and deployment assumptions. (3) Building on this synthesis, we identify key open challenges—including the absence of standardized explainability benchmarks for time series, the interpretability gap in state space models, and the need to advance from correlational to causal explanations—and offer reporting considerations to guide more comparable future research.