Multi-Source Heterogeneous Data Fusion Method for IoT Terminals

To address the potential inaccuracies in multi-source heterogeneous data collected by Internet of Things terminals, this paper proposes a data fusion technique based on SOM-FCM Enhanced Bidirectional LSTM Sequence-to-Sequence Model. The proposed method first employs a two-layer clustering algorithm, self-organizing map-fuzzy clustering mean, to achieve dimensionality reduction through clustering. Subsequently, the bidirectional long short-term memory sequence-to-sequence algorithm is applied for anomaly detection, and the attribute-associated features of heterogeneous perceptual data are integrated to train a heterogeneous data fusion model offline. Finally, the overall data accuracy is enhanced through fusion processing. To evaluate the effectiveness of the proposed method, it is benchmarked against multiple regression methods, radial basis function neural network structures, and Dempster-Shafer evidence theory. Simulation results demonstrate that the method achieves the lowest mean square error across all four datasets. These findings confirm that the proposed algorithm effectively detects anomalous data, reduces measurement errors, and produces sampled values that closely approximate the actual true values.