Wind Measurement With Arc Ultrasonic Sensor Array Based On WMNet

Wind parameter measurement is very important for resource utilization and daily maintenance of wind turbine.At present,the traditional wind parameter measurement methods are developed based on array signal processing.This kind of methods can utilize spatial information,but there are problems of high computational complexity and low real-time performance.In order to solve this problem,this paper proposes a neural network for wind measurement based on convolutional neural network(CNN),which is called wind-parameter measurement network(WMNet)in this paper.This network is combined with arc ultrasonic array to realize wind parameter measurement.This method has high precision,low computational complexity,and the ability of real-time wind parameter measurement.The main contents of this paper are as follows:1.Aiming at the problems of high computational complexity and low real-time performance of traditional array signal processing algorithms,a neural network model for extracting wind speed and wind direction parameters is proposed,and two WMNet models based on CNN are built to extract wind speed and wind direction characteristics respectively,transforming the estimation problem into a feature classification problem to improve the real-time performance of wind parameter estimation.Through simulation experiments,the feasibility and related performance of WMNet algorithm were analyzed,and the computational load and real-time performance comparison experiments were conducted with multiple signal classification(MUSIC)algorithm to verify the advantages of WMNet algorithm in reducing computational complexity.2.In view of the limited computing power of the mobile terminal platform and the gap between the file size of the weight parameter of WMNet algorithm and the memory size of the mobile terminal platform,the model pruning optimization of WMNet was carried out based on the depth separable convolution(DSC)under the condition of acceptable performance attenuation.Through simulation experiment,the feasibility and related performance of DSC-WMNet are analyzed,and the computational load analysis and real-time comparison experiment between DSC-WMNet,WMNet and MUSIC algorithm are carried out.The analysis proves that the accuracy attenuation of DSC-WMNet within the acceptable range can greatly reduce the computational load of the algorithm.It lays a theoretical foundation for the algorithm migration of mobile terminal platform.3.A semi-physical hardware experiment platform was built to verify the feasibility of the wind parameter measurement method proposed in this paper.First,the overall structure and various components of the experiment platform were described,and the performance of the wind tunnel of the experiment platform was verified.Then the preprocessed data is applied into the proposed algorithm to verify the feasibility and engineering realizability of the proposed algorithm.Finally,the real-time performance is verified by comparison experiment,and the error analysis and discussion of the measurement results are carried out.Through the analysis of the algorithm,the designed simulation experiment and the algorithm performance comparison experiment,the measured experiment verifies that the success rate of wind speed estimation of WMNet algorithm reaches 100% when the signal-to-noise ratio is greater than 12 d B,and the success rate of wind speed estimation reaches 100% when the signalto-noise ratio is greater than 14 d B.Compared with MUSIC algorithm,WMNet algorithm reduces the computational load by 65.16% and improves the real-time performance by 86.70%;DSCWMNet algorithm reduces the computational load by 98.40% and improves the real-time performance by 93.50%.It is proved that the proposed algorithm can effectively reduce the computational complexity and improve the real-time performance of the wind measurement system on the premise of ensuring the estimation accuracy.It has theoretical and practical significance for constructing clean energy,low-carbon,safe and efficient energy system and improving energy utilization efficiency.