| With the vigorous development of China’s economy,the full use of the ocean is becoming more and more important for the country.Ships play an important role in developing marine resources and protecting the life line of trade.Therefore,in recent years,the research on ships is also increasing.Ship underwater radiated noise is an important branch of ship research.Among them,underwater radiated noise data is the cornerstone of the research on underwater radiated noise of ships.However,at present,the underwater radiated noise data is less open,the acquisition cost is high,and the experiment is difficult.Therefore,how to generate the simulated underwater radiated noise more quickly and more widely has become one of the important directions.In recent years,artificial intelligence and deep learning have emerged a new wave and development.Deep learning technology has some applications and achievements in many fields.How to use deep learning technology to realize the simulation generation of ship underwater radiated noise is the key of current research.In order to solve the problem of low accuracy and large error of traditional convolution network model in modulation spectrum parameter extraction,the Temporal Convolutional Network method is adopted in this thesis,and a modulation spectrum parameter extraction method based on Multi-dimensional Jump Temporal Convolutional Network(MJTCN)is proposed.In this method,a multi-dimensional input method in time domain and time-frequency domain is proposed to make the model learn the characteristics of different dimensions;the concept of residual error is introduced to make the model not easy to disappear and explode gradients;multiple output tasks share most of the hidden layers to reduce the amount of computation;jump expansion causal convolution is proposed to expand the receptive field of the model.Aiming at the problems of poor generality and high computational complexity of traditional continuous spectrum simulation generation methods,the Generative Adversarial Nets method is adopted in this thesis,and a simulation generation method of continuous spectrum signal based on Long noise Conditional Wave Generative Adversarial Nets(LCWave GAN)is proposed.In this method,a multi-resolution STFT auxiliary loss function is introduced to adapt to the data characteristics of continuous spectral component noise,and a random sampling mechanism is proposed and designed to generate underwater radiated noise for a long time.In order to solve the problem that the frequency shift is not considered in the previous methods of underwater radiated noise simulation,a method of underwater radiated noise simulation based on time delay filter is proposed in this thesis.In this method,the Doppler frequency shift caused by ship motion is simulated by introducing a time delay filter,and the underwater radiated noise propagation loss is calculated according to the real-time distance,so as to generate the simulated underwater radiated noise.In summary,in this thesis,the simulation method of ship underwater radiated noise combined with deep learning technology can generate the simulated underwater radiated noise more conveniently and universally on the basis of high degree of simulation,and has achieved good results in the comparative test.Through the research of this thesis,it makes a contribution to expand the data set of ship underwater radiated noise,so that researchers can generate the required simulated ship underwater radiated noise data more quickly and with lower cost. |
