A multiparameter monitoring method for pumped storage power stations based on integrated sensing and distributed fiber optic sensing

Abstract

Monitoring algorithms for pumped storage power plants primarily rely on single-parameter monitoring mechanisms, where independent sensing devices are deployed to collect physical data such as vibration and temperature of key equipment, with anomaly detection achieved through preset thresholds. Due to the difficulty in conducting correlation analysis of multi-parameter coupled features, monitoring accuracy is often suboptimal. This study proposes a multi parameter monitoring method for pumped storage power stations based on integrated sensing and distributed fiber optic sensing. Integrated sensing and communication distributed optical fiber sensing technology is introduced to simultaneously capture multi-physical information such as strain and temperature through a single optical fiber, and a dynamic correlation model is constructed to enable multi-physical sensing. The Welch method is employed to estimate the power spectral density of vibration and temperature signals, and the spectral entropy of each parameter signal is calculated to quantify its complexity. A spatiotemporal graph neural network is constructed, modeling the topological structure of power plant equipment as a graph structure, with multi-parameter time-series data serving as node features, enabling collaborative prediction of vibration states and temperature trends, thereby achieving multi-parameter monitoring of the power plant's operational status. In experiments, the monitoring accuracy of the proposed method was validated. The final test results indicate that when applying the proposed method for multi-parameter monitoring of pumped storage power plants, the credible interval coverage for temperature stabilizes around 0.96 after 200 iterations, demonstrating satisfactory monitoring performance.