Tight control of the output energy is required in high-power laser devices. The main amplifier provides the most dominant energy gain, whose output needs to be predicted accurately. However, due to its complex structure and time-varying performance, the prediction results using traditional physical model-fitting methods are biased. In this paper, we propose a physical knowledge-based neural network, with an analytical model as the backbone and multidimensional influencing factors introduced by neural networks as input, to achieve accurate prediction. The method combines the powerful characterization ability of neural networks and the interpretability of physical models, which significantly improves the accuracy by considering the coupling effects of several factors and measurement errors. The relative deviation of the method's prediction results improves 65.9% compared to the traditional physical model and 57.9% compared to the pure neural network. The model provides a correction approach for similar problems of oversimplified physical models and can be exploited to aid model development of other measurable processes in physical science.
The energy accuracy of laser beams is an essential property of inertial confinement fusion (ICF). However, the energy gain is difficult to be predicted and controlled precisely due to the dramatically-increasing complexity of huge optical systems. Artificial neural network is a numerical algorithm with valuable flexibility that maps inputs to output values, which provides an approach to figure out this issue. In the study, a novel method combining deep neural networks and the Frantz- Nodvik equations is proposed to predict the output energy of the main amplifier in the high-power ICF laser system. To improve the prediction performance, the artificial neural network counts in more related factors that are neglected in traditional configurations. Dynamic state parameters describing amplification capacity are output by neural network and constrained by physical prior knowledge. The experimental results show that the proposed method provides a more accurate prediction of output energy than the conventional fitting approaches, from 6.5% to 4.2% on relative deviation. The study investigates the methodology of combining neural networks with physical models to reproduce a complex energy gain process and to represent a nonlinear unresolvable model, which can be exploited to aid model development of other measurable processes in physical science.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.