MULTI-SENSOR DATA FUSION BASED ON GCN-LSTM

Due to the constraints of traditional multi-sensor systems with multiple sources and heterogeneity, the common fusion model is inefficient and error-prone, which is difficult to meet the demand for multi-sensor data fusion. To cope with this problem, this paper proposes a combined graph convolutional network (GCN) and long short-term memory (LSTM) algorithm to construct a feature-level fusion model for sensor data. The model uses GCN to extract features of multi-source heterogeneous data, which solves the problem of difficult fusion of heterogeneous data caused by differences in data types, and LSTM for feature extraction of time series, which solves the problem of gradient disappearance. Finally, the proposed model is validated on an industrial-grade data set, and the test results show the model's effectiveness.