Prediction Of Ionospheric Parameters Based On Depth Learning And Incoherent Scattering Radar

The Earth’s ionosphere has important significance for modern information activities such as wireless communication,navigation,radar detection,etc.The Incoherent Scattering Radar(ISR)is an important large-scale ground-based detection equipment for ionospheric detection and scientific research.It is also one of the world’s competitive technological heights.With the successful construction of the Kunming Incoherent Scattering Radar and the Sanya Incoherent Scattering Radar,China has become an important participant in the competition of incoherent scattering radar technology in the world.Incoherent scattering radar technology has also become one of the hotspots of attention in China’s space science and information science.Traditionally,extracting ionospheric parameters,i.e.parameter inversion,from the echo signals of incoherent scattering radars is mainly based on the physical modeling of non-coherent scattering spectra of ionospheric plasma,introducing priori parameters.The echo power spectral curve is obtained according to the scatter spectrum model,and then iterative parameter modification starts based on the difference between the priori and the measured power spectra,until the error between the priori power spectrum and the measured power spectrum is less than the preset value.However,this method has problems such as slow convergence or divergence,which significantly affects the detection accuracy of incoherent scattering radars.On the other hand,with the rise of artificial intelligence technology,deep learning has been widely used for modern radar signal processing and target key parameter extraction.However,since the parameter inversion of incoherent scattering radars highly depends on the physical modeling of non-coherent scattering spectra,it requires sufficient physical interpretability of the parameter inversion results.Artificial intelligence cannot provide physical details in parameter inversion,and thus has not been applied to the parameter inversion of incoherent scattering radars.This study aims to address the issues regarding the low computational efficiency,complex iterative process,and non-convergence in the traditional parameter inversion method of incoherent scattering radars.Based on the measured data of incoherent scattering radars in the Madrigal database,we established a training set and testing set,and introduced intelligent algorithms and neural network models to conduct ionospheric parameter prediction research based on Kalman filtering,convolutional neural networks,recurrent neural networks,and long short-term memory neural networks.We achieved ionospheric calm and disturbed period parameter prediction based on incoherent scattering radar detection,and used the prediction results as priori parameters for iterative calculations to extract ionospheric parameters.The research results demonstrate that accelerating the parameter inversion process of incoherent scattering radars with deep learning can improve the convergence speed,convergence performance,and inversion accuracy of iterative computations while retaining the physical interpretability of the inversion results.This approach is feasible and may become one of the potential technological routes for the next generation of incoherent scattering radar parameter inversion.