ESTIMATION MODEL STRUCTURING DEVICE, INFORMATION PROCESSOR, AND RECORDING MEDIUM

摘要

PROBLEM TO BE SOLVED: To structure a nonlinear polynomial type estimation model which obtains output data from arbitrary input data and is no inferior in precision with a small calculation quantity by improving the conventional GMDH arithmetic method. SOLUTION: Two input variables are selected at random (S110) out of input variables corresponding to (n) pieces of input data, and a partial polynomial is generated (S120) by using those two input variables and data for learning. To evaluate the generated partial polynomial, an error in squaring is calculated (S130) by using data for evaluation. Only partial polynomials reducing the error in squaring are stored and the outputs of the partial polynomials are added to the input data to enable chaining (S140, S150: YES at S160). Those processes are repeated until an end condition is met (S170).