Physical Spatial-Temporal Weighted Graph Construction for FTU's Health Condition Assessment
Yujie Liu, Jin Zhang, Yuxin Li, Fengyuan Zhang, Yitao Fei, Jie Liu · 2024
To ensure the dependable operation of hydropower generation, substantial research efforts have been dedicated to the health condition assessment (HCA) of Francis turbine units (FTUs). Typical HCA strategies encompass the creation of health benchmark model (HBM) and the calculation of performance degradation indicators (PDIs). However, Previous investigations have primarily concentrated on either the temporal or spatial relationships within pulsation signals and working condition parameters, ignoring the complex interactions between these relationships. In addition, a simplistic averaging of temporal and spatial relationships can lead to the loss of critical information. This paper proposes a physical spatial-temporal weighted graph construction for the health condition assessment of FTUs. Initially, a simulation model of the FTUs is developed, utilizing computational fluid dynamics theories to produce simulated pulsation signals. Then, actual pulsation signals are used to refine the simulation model, achieving a high-fidelity representation. Subsequently, the pulsation signals are modeled using graph data to capture both physical spatial relationships and essential features. On this basis, time-related statistical parameters are employed to allocate weights to each identified relationship. Furthermore, the integration of the feature relationships with physical spatial relationships leads to the creation of a physical spatial-temporal weighted graph, enhancing its signal representation capability. Ultimately, this graph is fed into the graph convolutional model to obtain HBM. To assess the performance, the PDI is calculated by extracting the Euclidean distance between the comprehensive features derived from the HBM-predicted labels and the actual signals observed in the degraded state. Validation experiment is conducted to verify the effectiveness of proposed method through the utilization of real-time monitoring data and working condition parameters from the FTU deployed within the hydropower station.