A Genetic Programming-Driven & v" L% |7 D7 \5 Q5 |2 x$ u
Data Fitting Method ) O2 t3 E* f) u5 e
6 F6 E9 ^$ e+ I* P6 S
$ A+ Q% ?+ L8 L) t* M4 h) ?7 L
6 W7 \' E7 Z" J i" e4 R& r
Data fifitting is the process of constructing a curve, or a set of mathematical functions, that has
$ Y4 }) T$ G: T) H, V1 Y4 o& Z7 Cthe best fifit to a series of data points. Different with constructing a fifitting model from same type of function,
/ h" Y; ~* O r" c0 C( gsuch as the polynomial model, we notice that a hybrid fifitting model with multiple types of function may have
% T7 N; K5 g, B' {+ P6 v( T: o: `8 Z4 Ya better fifitting result. Moreover, this also shows better interpretability. However, a perfect smooth hybrid- Z! r/ C% }5 u) M3 G( ?9 k/ _+ F
fifitting model depends on a reasonable combination of multiple functions and a set of effective parameters.5 |* j" A+ Q7 w# `3 A
That is a high-dimensional multi-objective optimization problem. This paper proposes a novel data fifitting
, |6 S0 e* G. m, ~2 \ Emodel construction approach. In this approach, the model is expressed by an improved tree coding expression
2 G! ?) ^7 T" J0 K2 o) |and constructed through an evolution search process driven by the genetic programming. In order to verify! j) R. `8 }$ E
the validity of generated hybrid fifitting model, 6 prediction problems are chosen for experiment studies. The
, E* v9 y O6 q& _3 c' u% [experimental results show that the proposed method is superior to 7 typical methods in terms of the prediction3 C h2 T0 Q" s9 b+ G1 e8 a
accuracy and interpretability
& s2 o' G6 E5 U' |
+ d( R) a8 M, V3 U
' Z7 j3 l" H: s" @+ U; N
( X) a8 v1 E! j7 N0 A( d3 n' m+ v, [
) w& u3 \& ]( K, y B( r7 M
|