Abstract:

An optimization scheme was developed to address the issue of excessive stress accumulation and fatigue damage in wind turbine components caused by mismatched dispatch instructions and wind speed during actual operation. To mitigate the unavoidable interference of noise and delay inherent in long-distance sensor data transmission， a seven-level real-time noise reduction feedforward neural network was developed by combining multiple feedforward neural networks. The collected signals were first denoised before being fed into the controller for further analysis. To verify the robustness of the seven-level real-time noise reduction feedforward neural network under complex scenarios， additional noisy and delayed data were used for testing. The results confirmed that the designed noise reduction neural network met the design requirements. To achieve real-time quantification of the cumulative fatigue damage in wind turbine components， the three-point rainflow counting method was improved. The dispatch instructions for wind turbines were optimized using the Sequential Quadratic Programming （SQP） algorithm. The optimized dispatch instructions were more reasonable， leading to a more uniform distribution of cumulative fatigue damage across turbines compared to pre-optimization. This avoids the excessive accumulation of fatigue damage in individual turbines.

Key words: feedforward neural network, seven-level real-time noise reduction neural network, rainflow counting method, power allocation optimization, SQP algorithm, Optimization of noise mitigation in wind farms