Industrial system anomaly detection method and device based on sensing and flow data fusion

The invention discloses an industrial system anomaly detection method and system based on sensing and flow data fusion. The method comprises the steps that sensing data and flow data are obtained, the two data are sliced, aligned and subjected to feature extraction to form training sample sets, and then the training sample sets are sampled to construct a plurality of respective sub-training sample sets; local training based on a sensing local reconstruction error and a flow local reconstruction error is carried out on a plurality of sensing local models and a flow local model by using a plurality of sub training sample sets corresponding to various data; extracting a sensing local reconstruction error and a flow local reconstruction error corresponding to the trained sensing local model and the flow local model as input of a global model, and performing global training on the global model based on the global reconstruction error; and the trained sensing local model, the traffic local model and the global mode