78 / 2015-10-24 17:18:00

Improved MGGP Approach for Spatial Grade Prediction

spatial grade prediction,OK,SVR,Improved MGGP

The demand for a new grade predictor originates from the limitation of conventional methods. Typically, geostatistics methods such as kriging reflect well the spatial variations, but on the other hand they only capture the weighted linear relationship between sampled values and evaluated value. While artificial intelligent methods such as neural network and support vector regression (SVR) have ability to capture non-linear relationships between the input and output data under extreme conditions, they show poor performance when the sample size is small due to lack of spatial variation in sample field. This paper introduces an improved multi-gene genetic programming (MGGP) model for spatial grade prediction and also compares it with other well-known techniques such as ordinary kriging (OK) and SVR. The proposed technique organically combines the self-adaptive nonlinear capturing ability of MGGP about the input-output relationship with the expression ability of spatial variation in geostatistics. The results obtained from case study show that it produces much higher prediction accuracy than either OK or SVR, and has strong generalization ability.