Study On The Judging And Compesating Methods For Model Errors

Data processing needs to build math models. Generally, math models built are sometimes not consistent completely with realities. As a result, model errors come out inevitably. Especially, in modern times, with the high technology and high automatization level in data collection, observed value errors in routine surveys are not just accidental errors. In this paper, judging and compensating methods for model errors in survey data processing are analyzed, based on basic theory about model errors.In this paper, mathematic statistics is mainly used to judge model errors. The compensation methods analyzed here are methods of systematic parameter added, systematic weight added, least-squares collection, semiparametric regression and neural networks. In the last of the paper, two math models, include plane fitting method and conicoid fitting method, are particularly analyzed based on some projects. Problems that whether the two math models of GPS height conversion have model errors, and how to compensate the model errors are analyzed. The results of the example analyses indicate that if the landform is not very complex, the better GPS height conversion equation can be found by add model compensation to the math model, so that the forecast precision improve. The model error compensation method based on neural networks is much better.