Active Warning Method for Time-Series Data Based on Integrated Network Model with Multi-head Residuals

The factory data collected through IoT technology has a certain temporal sequence, and the distance between the initial moment and the final moment data is long, which leads to low accuracy and low real-time warning of the factory’s temporal data. In order to solve these problems, this paper proposes an active warning method for time series data based on the multi-head residual integrated network model (MHR-INM). The method first maps the collected real-time vector data to the corresponding warning space; then extracts the eigenvalues of the vectors through the MHR-INM model, and uses the modified cosine similarity computation method to compute and compare them with the saved time-series eigenvectors of the normal operation of the equipment; and finally compares the results of this computation with the set threshold, thus realizing the active warning of the factory. The experimental results show that the warning accuracy of the model is high compared with the existing methods.