Research On Elevation Anomaly Model Based On Seamless Partitioning Technology

In recent years,satellite positioning technology has been used more and more widely because of its advantages such as high efficiency,high accuracy,and strong real-time performance.However,if the GNSS elevation is to be applied in actual engineering construction and scientific research,the position of the geoid-like level must be accurately determined.The high-precision,high-resolution regional-like geoid model is of practical significance for changing the traditional geodetic model and using GNSS technology instead of low-level leveling.This paper focuses on the subject of geoid-like level refinement.It focuses on the introduction of wavelet neural network theory and its application in elevation anomaly fitting,and addresses the shortcomings of the large-area geoid-like level refinement and subdivision fitting method.A seamless partitioning technique is proposed.The main content and research results of this article include the following aspects:(1)Introduced the common methods of geoid-like level refinement,and analyzed the geometric method,gravity field model method and combination method in detail.This paper focuses on how to use geometric method to build a geoid-like model with higher accuracy without gravity data.(2)The wavelet neural network constructed by using the wavelet-based Morlet function as the activation function of the hidden layer of the neural network shows good results in the application of elevation anomaly fitting.The experimental results of an engineering example show that the error in the fitting of the traditional quadric surface model is 23.1cm,and the accuracy of the wavelet neural network model reaches 11.6cm,which is 49.8% higher than the accuracy of the quadric surface model.(3)The quadric surface model is intuitive and specific,and the results are unique,while the wavelet neural network has good fitting accuracy but cannot be expressed by intuitive formulas.Therefore,this paper uses the advantages of the quadric surface model and the wavelet neural network model to improve the wavelet neural network based on the fitting data of the quadric surface model.The experimental results show that the fitting accuracy is further improved,reaching 8.6 cm,which is 62.8% higher than the quadric surface model accuracy.(4)Aiming at the shortcomings of the partition fitting method,a large-area-like geoid refinement method for seamless partitioning is proposed.The experimental results show that the seamless partitioning technique used for elevation anomaly fitting has a fitting accuracy of6.9 cm,which is a 70.1% improvement over the quadric surface model accuracy.Both the seamless partitioning technique and the improved wavelet neural network model show good accuracy and can complement each other,which is suitable for practical engineering applications.