Cavitation Coefficient Identification Model for an Axial Flow Pump Based on Pressure Signal Feature Extraction and Spider Wasp Optimization Algorithm

This research proposes a quantitative cavitation-coefficient identification method based on a VMD-SWO-BiLSTM model, addressing the current limitations of existing machine-learning-based cavitation diagnostics, which mainly classify cavitation stages rather than identify the cavitation coefficient as a continuous quantity. Pressure fluctuation signals under various cavitation conditions are decomposed using Variational Mode Decomposition (VMD), and sample entropy is extracted to represent signal complexity. The Spider Wasp Optimization (SWO) algorithm optimizes the hyperparameters of a BiLSTM network, forming a composite VMD-SWO-BiLSTM framework. Multi-point input features from unsteady internal flow simulations and experimental measurements are used for model training and validation. Results show that the proposed model outperforms conventional BiLSTM, achieving mean absolute percentage error below 5% based on multi-point pressure signals. Validation with eight experimental datasets yields a maximum absolute error of 0.006 and a maximum percentage error of 4.7%. The proposed method can be applied to online monitoring and intelligent diagnosis of cavitation in axial flow pumps.