Research On Wave Height Extraction Of Compact High Frequency Radar Based On Neural Network

With the rapid development of marine economy,the problem of marine safety has become increasingly prominent,and it is urgent to further strengthen the prevention and control of marine disasters.Among the Marine disasters mainly including storm surge,sea wave and red tide,sea wave is the Marine disaster with the highest risk factor,which is related to the life safety of coastal and maritime personnel.The all-day,all-weather,large-range,high-precision wave map observation can master the sea state information in real time,which is of great significance to strengthen the risk prevention work for maritime personnel.High Frequency Ground Wave Radar(HFGWR)has the ability of over sight detection and long-term stable operation,and is one of the ideal equipment for coastal wave field observation.The compact HFGWR has been widely used in the world because of its advantages such as compact and light weight,simple site requirements,easy erecting and maintenance and low cost.In the existing wave map extraction methods for compact HFGWR,aiming at the shortcomings of the method based on the fitting relationship between the singular peak and the significant wave height: The simplified fitting relationship makes the model have certain errors and weak generalization;The single echo parameter makes the algorithm unsuitable for transforming complex sea conditions.In this thesis,a significant wave height map extraction method based on Back Propagation(BP)neural network is proposed.With the strong feature learning and nonlinear fitting ability of BP neural network,a more accurate relationship model between radar echo parameters and significant wave height is obtained by training;With the first-order Bragg peak,the second harmonic peak and the ratio of the second harmonic peak to the Bragg peak as the model inputs,the relationship model between multiple echo parameters and significant wave height is constructed to make up for the limitation of inversion of wave height with a single echo parameter.The method presented in this thesis is compared with the linear fitting method based on the ratio of the second harmonic peak to the Bragg peak for the measured radar data,the results show that: The proposed method is superior to the latter in terms of correlation coefficient and root mean square error index between radar inversion wave height and numerical model reference wave height.The feasibility of this method in wave map extraction of HFGWR is verified.In addition,for sea conditions such as a large wave height fluctuation range during typhoon,the algorithm can not cover the actual wave height fluctuation range due to the double limitation of the saturation upper limit and the detection lower limit of wave height by using single-frequency radar echo parameters to retrieve wave height.In this thesis,a method of wave map extraction by combining the echo parameters of dual-frequency radar is presented.Taking the corresponding the first-order Bragg peak,the second harmonic peak,the ratio of the second harmonic peak to the Bragg peak and dual-frequency first-order peak ratio parameters in the echoes of the two operating frequencies as model inputs,the corresponding relational model is obtained by training with BP neural network.Compared with the algorithm of single-frequency network model,the results show that: The dual frequency algorithm has better performance in both high precision wave map coverage and wave height fluctuation range.In contrast,the extraction results of the dual-frequency algorithm are more robust under complex and variable sea conditions.