Application Of Machine Learning In GAMIT High Precision GPS Baseline Solution

There are some factors affecting the accuracy of the baseline solution results in the process of global positioning system(GPS)data resolving when using GAMIT high-precision data processing software.These factors are either the software's own limitations or human operations.The accuracy of the baseline solution quality affects the accuracy of the entire control network,and also affects the subsequent GLOBK adjustment processing results and the application research of various deformation monitoring.Therefore,it is necessary to study the factors affecting the accuracy of the baseline solution and explore new methods for improving the accuracy.This paper mainly studies the following three aspects: First,due to the software's ability to handle the number of stations,GAMIT software needs to partition the station when solving large intensive stations.However,the general zoning scheme will result in the existence of both long and short baselines,thus reducing the accuracy of the entire network.Second,when GAMIT software solves GPS data,the setting of different control parameters has a great influence on the resolution accuracy.Improper setting of the solution control parameters will cause the normalized root-mean-square value(NRMS)of the solution result to be too large or the solution to fail.Thirdly,when GPS data is solved by GAMIT software,the selection scheme of different International GPS Service(IGS)stations has certain influence on the calculation accuracy,and the selection process has certain randomness and mechanicality.In view of the above three problems caused by software limitations or human operation,this paper explores the use of machine learning algorithms to solve.First,the K-means++ algorithm and the Hash algorithm are introduced to partition the intensive stations to reduce the impact of short baselines on accuracy and improve the accuracy of the solution.Secondly,analyzing the influence of different control parameters on the accuracy of baseline calculation and selecting four types of solving control parameters with great influence as the discriminating factors,andestablishing the Bayes discriminant model to predict the results of the baseline solution.Thirdly,analyzing the influence of different station selection schemes on accuracy,and using the Generative Adversarial Networks(GAN)algorithm to realize the intelligent selection of IGS sites.The experimental results show that: Firstly,the proposed K-means++ partitioning method can effectively solve the problem of reducing the accuracy of the whole network caused by the presence of long and short baselines.It is more accurate than the general area partitioning method,and it is consistent with the accuracy of the existing spacing partitioning method,but it is more stable and efficient than the spacing partitioning method.Secondly,the established Bayes discriminant model has good prediction performance,and can effectively predict the baseline solution result under the premise of ensuring the accuracy of baseline solution.It can solve the problem of inefficiency caused by repeated solution in the solution of large network.Third,the constructed GAN intelligent station selection model can automatically select the IGS site to ensure the accuracy of the solution and improve the work efficiency.