Method for theoretically calculating curve overfitting degrees in field of comparative hydrology

Abstract

The invention discloses a method for calculating curve overfitting degrees in field of comparative hydrology. According to the method, a computer program and a calculus principle are used for comparing smoothness of two curves which are drawn on the basis of hydrological measured data and have same precisions, namely, judging overfitting problems of the curves; the precisions, at measured data points, of the curves are the same, but the overfitting degrees are different. Through comparing lengths of the two curves which are drawn by using the hydrological measured data and have same precisions, the curve with a relatively long length has relatively great fluctuation, relatively bad smoothness and a relatively severe overfitting phenomenon, and the curve with a relatively short length has relatively small fluctuation, relatively good smoothness and a relatively light overfitting phenomenon, and is a curve according with conditions, namely, the curve has a moderate fitting degree and accords with common laws, so that the curve overfitting problem during hydrological data processing is solved. According to the method, the steps are relatively simple and feasible, complex mathematic formula derivation is abandoned, and a curve with a relatively fitting degree is selected from two to-be-selected curves more concisely and accurately by means of the computer program.

Description

technical field [0001] The present invention relates to a hydrological data processing method, in particular a method that can be used to compare the degree of overfitting of theoretical calculation curves in the field of hydrology. It is suitable for the scientific research of hydrological data processing to solve the problems caused by different calculation data [0002] The problem of judging the degree of overfitting of the drawn curve. Background technique [0003] The overfitting problem is a common problem in the field of scientific research. Simply put, to solve any problem, you need to follow the principle of specific analysis of specific problems, and use the specific method to solve a problem to solve other problems rashly. It is easy to make mistakes, that is, the general method principles are not summarized in the process of solving the problem, which is the problem of overfitting. In scientific research, it is often necessary to draw curves based on calculate...

AI Technical Summary

Problems solved by technology

These two methods to solve the overfitting problem have their own emphasis. The bias frequency method is more inclined to analyze from the perspective of error, while the Bayesian method focuses on analysis from the perspective of prior distribution, but in general , no matter which method is used, the overfitting problem is solved by adding norms. Both methods are biased towards derivation with formulas. The calculation of the method to solve the problem is very complicated and the feasibility is poor.

Images