Average Velocity of Debris Flow Forecast Based on Genetic Programming Combined with Rough Set Theory

Abstract

Accurate prediction is an essential ingredient in geotechnical engineering, which plays a significant role in the management of key engineering of countries. This paper presents a novel approach of genetic programming (GP) integrating with rough set (RS) to solve prediction problems in geotechnical engineering. Firstly, RS is considered as the fore-treatment tool of genetic programming to eliminate the redundant and noise information in training data, then, genetic programming is used to establish the prediction model. In order to verify the reliability of this method, the genetic programming model based on rough set is applied to predict debris flow velocity, taking the debris flow depth, the gradient of the channel, the debris flow density and the average grain size as the input factors, a total of 50 debris flow events are investigated in the JiangJia gully, which used for building and verifying the RS-GP model. After attribution reduction by RS, 4 groups of the noise data and weak interdependency data of training samples are eliminated, then, the rest 41 groups of the measured data are selected as the training database and 5 groups of data are used as testing data to establish the GP model. Finally, the modified DongChuan empirical equation and BP approach are also used for comparison and validation, whose results show that the RS-GP predicted debris flow velocity values are very close to the measured values, the maximum error is 2.97%, the average error is 0.9% .Therefore, the prediction accuracy of RS-GP prediction model is higher than those of empirical equation and BP, which indicate the model brought up in this paper is efficient and effective.