| The atmospheric refraction errors (ARE) are objectivity in the photo-electricity survey information (PESI), the error accurate correction is very difficult for its complexity. Some conventional correction rules are only feasible in the standard atmosphere based on the atmosphere models theory and the foliaceous hypothesis method. The conventional nonlinear modeling and statistics technique can not achieve the high precision correction in the nonstandard atmosphere, but the large numbers of photo-electricity survey works have been achieved in the nonstandard atmosphere. Along with unceasing enhancement of the equipment measuring accuracy and people accuracy requirements, the atmospheric refraction errors correction (AREC) desiderate improving in some high precision photo-electricity survey projects.The neural network theory has many advantages than the conventional nonlinear modeling and statistics technique to resemble the intricate nonlinear object of AREC, therefore, this research is trying to improve the precision of the AREC by using the back propagation neural network (BPNN) model for the PESI. It firstly presents the inland and overseas actuality and introduces interrelated the basic theory and essential method for the AREC study on this research field. Some original researches were carried out for the AREC with the BPNN model in the PESI. These research based on following foundation: the aforementioned elementary theory and essential method, the list data in the Pulkovo refraction table (PRT) and the correlation information of two distortion photo-electricity survey projects, the mathematics basic theory and the computer modeling simulation by the help of Matlab7 BP toolbox.(1). It brings forward a new concept of the high degree fraction form mapping function based on the mapping function (MF) principle of the AREC and the generalization expressions of over ten sorts MF function in mathematics, and transforms the basic MF to the BPNN model. It achieves the modeling and fitting of the models on the PRT firstly. Simulation shows that the BPNN is double of the 4-degree fraction form MF in the accuracy of simulation, and proves that correction precision of the MF modeling indeed has approached theoretic precision which the traditional non-linear modeling revises.(2). It put forward second new concept of the baseline back propagation neural network model (BLBPNNM). It studies the entire laboratory performance test of the photoelectric distance meter and analyzes that the measure errors of radio meteorology parameters to influence precision of the AREC. The BLBPNNM can separate the atmospheric refraction errors from the information of baseline photoelectric distance meter by very high precision.(3). It gives a new compendious mathematics proof without the ill-conditioned fraction form constraint for that both the sample pretreatment and the optimization training algorithm of BPNN are equivalence in the modeling efficiency by using the identity matrix and the basic character of matrix operation. It provides an efficacious sample pretreatment arithmetic that the value of sample divided by their average. Example shows that the arithmetic is more efficacious than the pretreatment function of MATLAB and the modeling efficiency of BPNN can enhance 1～4 times.(4). It performs such researches as the abnormity change rules of atmospheric refraction, how use the PESI and modeling principle, steps and application way of the BLBPNNM. It offers a new application method based on the full-day information consistency check between the modeling baseline and the correction line for the BLBPNNM. Example shows that the method is feasible and daily distortion mean value of the baseline is nearly zero and has separated the atmospheric refraction errors from the information of baseline photoelectric distance meter with very high precision.(5). The application research shows that daily mean correction precision (DMCP) of the baseline can attain 10-4mm/km by using the BLBPNNM, the neighboring baseline length ratio method is 10-3mm/km, the full-day mean value difference expressions is 10-2mm/km, the exact meteorology formula is 10-1mm/km when their full-day information consistency check are better, and the real-time correction effect of BLBPNNM is the best in these methods when the consistency check are not nice. So it presents a new correction viewpoint which all the models can improve the AREC precision in the PESI when their consistency checks are fine. The problem which the correction precision is uncertain by the traditional AREC method will make progress in application, which is hopeful on the new viewpoint.(6). In the high order information mining effect of atmospheric refraction, the math proof and simulation show that the mining power of the BPNN model excelled the MF, but the mining power comes from the hidden lay nerve cell rather than the input lay as a result of the mining effect of BPNN model is not distinct in a similar way which the MF form add high order items. Although both the BPNN and the MF are different in the mining modality, their fundamental principles are consistent in mathematics.(7). This research has developed and enriched "the AREC and the BPNN" in both theory and practice, and has attained the anticipative objectives.
|
PDF Full Text Request