Neural Network Based Method for Compensating Model Error
胡伍生,孙璐
DOI: https://doi.org/10.3969/j.issn.1003-7985.2009.03.024
2009-01-01
Abstract:Two traditional methods for compensating function model errors, the method of adding systematic parameters and the least-squares collection method, are introduced. A proposed method based on a BP neural network (called the H-BP algorithm) for compensating function model errors is put forward. The function model is assumed as y=f (x1, x2, …, x(subscript n)), and the special structure of the H-BP algorithm is determined as (n+1) ×p×1, where (n+1) is the element number of the input layer, and the elements are x1, x2, …, x(subscript n) and y' (y' is the value calculated by the function model); p is the element number of the hidden layer, and it is usually determined after many tests; 1 is the element number of the output layer, and the element is Δy=y0-y' (y0 is the known value of the sample). The calculation steps of the H-BP algorithm are introduced in detail. And then, the results of three methods for compensating function model errors from one engineering project are compared with each other. After being compensated, the accuracy of the traditional methods is about ±19 mm, and the accuracy of the H-BP algorithm is ±4.3 mm. It shows that the proposed method based on a neural network is more effective than traditional methods for compensating function model errors.