In recent years, the all-vanadium flow battery (VRFB) has demonstrated a notable trajectory of advancement as a large-scale, long-life energy storage technology, characterised by high safety standards and extended operational lifespan. In the practical operation of vanadium batteries, pump failures represent a significant category of incidents that have the potential to result in irreversible battery failure. The prompt identification and rectification of pump malfunctions can assist in the reduction of associated losses. A fault diagnosis method combining natural selection particle swarm optimisation (NPSO) and support vector machine (SVM) is proposed. By establishing an experimental platform to simulate constant current charging and discharging and wind power-VRFB joint operation conditions, and incorporating three recoverable fault modes, the researchers preprocessed the experimental data and extracted the fault characteristic parameters. The NPSO algorithm is employed to optimise the parameters of the SVM model, and a fault classification model is constructed. The results demonstrate that the classification accuracy of the model reaches 98.85% under simple working conditions and 96.76% under complex working conditions, indicating that the method can effectively enhance the diagnostic accuracy of VRFB pump faults.