State prediction of secondary power equipment based on data fusion

The stable operation of secondary power equipment is crucial to the reliability of power systems. Traditional maintenance methods are limited by insufficient accuracy and lag, making it urgent to improve state prediction technology. However, the monitoring data of secondary equipment presents characteristics of multi-source, heterogeneity, and noise, which hinder accurate prediction. To address this issue, this study proposes a state prediction method for secondary power equipment based on data fusion. The method first collects multi-source data and performs preprocessing. Then, a two-level fusion strategy is designed: data-level fusion integrates synchronous sensor data through weighted average and aligns asynchronous data with time markers; feature-level fusion combines time-frequency features of sensor data and text features of maintenance records. Finally, a long short-term memory (LSTM) network is constructed to predict equipment state using fused features. Experimental results show that the proposed method achieves the lowest MAE (0.021), RMSE (0.028), and MAPE (2.3%) on the test set, which are 28.9%, 31.7%, and 30.3% lower than the single-source LSTM method, respectively.