Research On The Theory And Application Of Distributed Computing For Large-scale Survey Adjustment

The rapid development of information technology promotes the change of the way ofmodern measurement data processing. Distributed processing of large-scale measurementadjustment provides a new high-performance handing for current enormous measurement dataprocessing. This paper carried out a more detailed study and practice, the main contents andinnovations are as the following:1. The background and current status of the large-scale measurement data processing areanalyzed and the main concern about the large measurement data processing is the huge GNSSnetwork processing. The current major challenges are pointed out, the theory and method ofdistributed processing of large-scale adjustment are put forward, and the scope of this paper isdiscussed and basic content are defined.2. The concept, characters and advantages of distributed computing technology areintroduced. Start from the classical least squares parameter adjustment model, this paper studiesthe equivalent of the combined solution and sequential solution, and discusses the method ofsub-block least squares adjustment under fixed parameters condition or variable parameterscondition. The consecution of combined solutions suitable the parallel adjustment calculation inthe distributed environment is obtained, and this paper elaborates on the principle of adjustmentmodel direct solution by solving the coefficient matrix generalized inverse. This concept laid thefoundation for parallel decomposition of adjustment models.3. The theory and method of adjustment models block parallel computing is proposed. Afterdescribed a variety of block matrix parallel operations of the adjustment calculation involved insome detail, the adjustment model block parallel processing algorithms are designed from thelevel of normal equations and the level of observation equations. The normal equations levelfocus on the block parallel of Gauss elimination method and Jacobi iterative method, and theobservation equations level studies the block parallel singular value decomposition of thecoefficient matrix by the exchange sequence of index. The accuracy and efficiency of blockparallel adjustment algorithms are verification under the multi-core and multi-machinedistributed environment.4. In order to process the large GNSS network data by using the block parallel adjustmentalgorithms, this concept studies the differences between the non-difference processing anddouble-difference processing from the view of time complexity, the PPP distributed processingand the double-differences distributed processing based on sub-network method are achieved byuse of distributed computing technology. The new network solution algorithm of Ambizap which is based on the fixed point theory is detailed derived. Through PPP distributed processing andindependent baseline double-difference distributed processing, the network double-differencefixed solution can constraints PPP solution and the whole network adjustment can parallelcomputing. The accuracy of calculation of the large-scale GNSS network data has ensured andthe limitation of the calculation scale has broken. By using distributed computing, this algorithmhas good portability and scalability and provides an efficient and economical processing methodfor large GNSS network data processing.5. Based on the theory, method and technique in this paper, the software prototype oflarge-scale adjustment distributed computing which is central concentrated on GNSS dataprocessing is designed and realized. The SINEX files of baseline solution or network adjustmentcan distributed parallel integrated, and the network adjustment which is based on coordinatemodel can parallel computing. The accuracy of the distributed parallel computing of adjustmentmethods and algorithms are verified by this software.