A Genetic Programming-Driven
/ S O* m+ ]6 H7 q( fData Fitting Method , _* d7 W3 \6 ^* R/ l* v
! e2 l- V9 o9 {: g# W& s
3 c! E- h2 F( }( W$ E. y l, \9 I
% x; x; h7 o0 M0 w: YData fifitting is the process of constructing a curve, or a set of mathematical functions, that has
$ n3 {6 q" l% B9 c' |) S: jthe best fifit to a series of data points. Different with constructing a fifitting model from same type of function,
: g4 Z" B. R$ t' v5 q5 l2 p8 Osuch as the polynomial model, we notice that a hybrid fifitting model with multiple types of function may have
5 U& F! L" {, s+ C. x; W7 x& na better fifitting result. Moreover, this also shows better interpretability. However, a perfect smooth hybrid
( ?: q! |8 {( u' J# n6 Y, R; u; Mfifitting model depends on a reasonable combination of multiple functions and a set of effective parameters.
5 s& A* @- S8 n3 U# I# DThat is a high-dimensional multi-objective optimization problem. This paper proposes a novel data fifitting/ B3 Z9 C9 c1 w/ M: l& L4 u1 F7 l
model construction approach. In this approach, the model is expressed by an improved tree coding expression
# R6 @, T0 ?& [ ^% F" Eand constructed through an evolution search process driven by the genetic programming. In order to verify/ l) W2 z- {3 p# U; [& U% O
the validity of generated hybrid fifitting model, 6 prediction problems are chosen for experiment studies. The
3 V6 U" w2 l$ {6 n9 q$ U1 w( I9 |' Qexperimental results show that the proposed method is superior to 7 typical methods in terms of the prediction
0 B* o0 @" z; F c: `1 v1 r9 raccuracy and interpretability
4 M7 a4 \8 u. n; p; j/ X& W/ `: K* ]5 f9 T, T( @4 X6 c) k
7 H+ q' X: b' ^2 N0 Q0 d$ o6 Y
, b" c5 p+ j" u
& j- H- X7 C# [) V% f/ E2 Y% K- m$ W$ L! E
|