RISK EVALUATION OF SOIL CONTAMINATION PURIFICATION COSTS BY BAYESIAN NEURAL NETWORK METHODS

Abstract

In this paper, a Bayesian Neural Network (in brief, BNN) model is presented to evaluate the risks of purification costs based upon surface survey on soil contamination. The traditional Neural Network (NN) model is applied to determine the network structure. The post stochastic distribution of the weight coefficients of the NN model are estimated by Bayesian estimation methods based upon the predetermined network structure. The BNN model presented in this paper is formulated to estimate the stochastic distribution of the purification costs (purification cost risk). The paper also discusses the methods to revise the BNN model by use of the additional samples newly obtained. The validity of the methodology the presented in this paper is examined based upon the case studied by use of the observations in the reality.