Research On High Resolution Reconstruction Of Ocean Temperature Field Combining Ocean Model Prediction With Moving Node Observation

Ocean temperature field plays an important role in acoustic target detection,underwater acoustic communication,and acoustic positioning of underwater targets.Under the influence of various complex ocean dynamic processes,the ocean temperature field has a strong spacetime variation.Ocean water temperature can usually be measured directly by instrumentation.However,direct measurements can only obtain local,short-term point or line temperatures,and cannot obtain temperature data for the whole field.The ocean dynamic model can predict a wide range of temperature field information,but its prediction accuracy and resolution are difficult to apply to subsequent acoustic calculations due to its limitation of physical scale and computational complexity.Therefore,for a long time,ocean environmental monitoring has been confronted with the contradictions between resolution and working range,space-time scale and resources.Based on the local high-precision temperature data observed by moving nodes and the whole-field low-resolution temperature field predicted by the ocean model,several water temperature field reconstruction models and algorithms based on multilevel interpolation model,regression model,removal and recovery technology and radial basis function(RBF)interpolation method are presented in this paper to achieve the high resolution,accuracy and real-time temperature field reconstruction of the whole field.The models and algorithms are validated by ocean experimental data.The specific work in this paper is as follows.To meet the application requirements of high resolution,high precision and fast reconstruction of ocean three-dimensional temperature field,the existing high resolution multilevel spectral interpolation(HRMSI)algorithm achieves fast solution by fast Fourier transform(FFT),which limits the order of Fourier basis function to the current grid size,resulting in incomplete field feature extraction.For this problem,this paper presents a sparse Bayesian learning(SBL)interpolation method of multilevel Fourier basis,which expands the Fourier basis orders of each dimension in each iteration of the original algorithm,and uses the SBL algorithm to sparsely solve the feature components,thus improving the reconstruction accuracy slightly.Considering that the multilevel Fourier basis SBL interpolation method is computationally complex and has reached the performance limit of Fourier basis.Therefore,by further replacing Fourier basis with RBF with better spatial correlation,the multilevel RBF interpolation model can better extract the related information between location and temperature of the temperature field,and the accuracy and speed of field reconstruction can be improved to some extent.To solve the problems of high reconstruction time,inadequate application of observed data,and limited resolution of reconstruction in multilevel interpolation model,a data fusion regression model and a removal and recovery regression model which can reconstruct temperature field at once are presented.Both types of regression models can use RBF interpolation and Kriging interpolation as model solvers.The data fusion regression model can fuse the two types of data and quickly reconstruct the high-resolution temperature field.The simulation results show that the accuracy of RBF interpolation applied to the model is close to that of Kriging interpolation,and the reconstructed speed is much faster than that of Kriging interpolation.Removal and recovery regression model can learn the difference between the observed data and the model data,and it can reconstruct the whole field and improve the accuracy,which is much better than other models.To solve the problem of over-fitting of the model when the coverage of node observation is low,a variety of filling methods of residual components are proposed to control the fitting process,which further improves the stability and flexibility of the model reconstruction.This paper uses the 2020 ocean experimental data and forecast data to validate the reconstruction of three types of models.Based on the experimental scene,a vertical profile and a horizontal profile in the experimental sea area are reconstructed,the effect of profile reconstructions of each model is analyzed,and the observation data of autonomous underwater glider(AUG)are predicted,and the accuracy and speed of various reconstruction algorithms are analyzed.The experimental data processing results show that the multilevel RBF interpolation model can only improve the reconstruction accuracy of the sea area near the AUG route,with the highest reconstruction complexity and the lowest resolution.The RBF fusion regression model can control the horizontal mapping range of AUG observation data by adding scale factor,which can improve the reconstruction accuracy of distant ocean region to a certain extent,with the lowest reconstruction complexity and higher resolution.Removal and recovery regression model reconstruction works best,and for its overfitting problem,a depth-based residual filling method is proposed to improve the model,resulting in a more realistic reconstructed result,with with lower complexity,the highest resolution and the best data fitting characteristics,which is the best model for ocean temperature field reconstruction in the three types of models.