The rapid development of the Industrial Internet of Things (IIoT) has enabled the communication of numerous devices, resulting in the generation of massive time-series data. The goal of time-series anomaly detection algorithms is to quickly detect anomalous behavior and trigger warnings. We propose a Hybrid Unsupervised Time-series Anomaly Detection method (HUTAD) that fully leverages two different anomaly detection strategies: spatial feature extraction by a variational autoencoder and temporal feature extraction by a variational autoencoder and a recurrent feature extraction LSTM structure with predictive capability. Through a dynamic fusion module based on the self-attention mechanism, we integrate the two methods to obtain the output, analyze the spatio-temporal patterns of the data, and detect potential anomalies. We incorporate adversarial training in our model to further improve its accuracy and robustness. Finally, we use three widely-used large-scale datasets to validate our model. The experimental results show that our method outperforms the baseline methods in F1-Score.