The Denoising Technology Of Large Laser Pulse Waveform Based On CNN

Inertial confinement fusion(ICF)is one of the main ways of controlled fusion.The research of ICF is based on high-power laser pulse emission experiment of large solid laser device.In order to improve the success rate of experiment,reduce unnecessary losses,the automation and intelligence of analysis and control system of laser device is one of the most critical subjects.The measured waveform of laser pulse emitted by each experiment of the device is the biggest basis for the design of each component parameter of the intelligent control system.Its precision directly affects the control ability of the whole system,which is the premise for the accurate and reliable realization of each function.Limited by the space structure of the device,suppression and compensation of the waveform’s distortion cannot get a perfect result simply by hardware,while it is difficult to use the traditional denoising method to process the waveform data,due to nonlinear transmission of lasers which leads to plenty of influencing factors and indeterminacy.In this case,there are still a lot of distortion in waveform data used in project development.Under such background of this project,the strong fitting and generalization ability of Artificial Intelligence for complex models has been attached great importance in this paper,then the Convolutional Neural Network method in machine learning has been adopted to process data.In response to the problems including high emission costs,lack of the measured waveform data,and the truth of no real value,use the simulation program to simulate waveform from initial state to the oscilloscope measurement process.The generated data is used to train the network.Then use the measured waveform to test network’s training effect.Based on the above technical route,the following works have been carried out:(1)Refer to the actual transmission characteristics of the large device,research the shape feature of the measured pulse waveform data,set the simulation waveform generating rules,designs the simulation waveform generator,and augment the data according to the need of network training.Using the program to generate the simulate waveform,laying a foundation for the next step of network training;(2)In view of the CNN’s well capability of feature extraction,decide that use CNN to realize the project requirements,design the specific structure and complete the training parameters.Verify the network by the simulate waveform,then test it by measured waveform.While training effect is not ideal,summed up the causes of the results by changing the network convolutional depth;(3)Inspired by the Fully Convolutional Networks,choose the U-Net model to be the basic architecture of the denoising network.Design the specific structure again to complete training,then verify and test the network in the same way.Finally,the effect is far better than the previous common structure,achieving the research purpose of this paper.