سیستم پشتیبانی تصمیم میکرو تونل زنی (MDS)با استفاده از مدل مارکوف مخفی عصبی خودبازگشت (اتورگرسیو)

Microtunneling is a trenchless technology method used for installing new pipelines. The inherent advantages of this method over open-cut trenching have led to its increasing use. This paper presents a general model for microtunneling decision support system (MDS) that can be used as a basis for developing more effective microtunneling design and construction. The model objectives are to: (1) develop a description of local geology that reflects the uncertainty of the information on which it is based and (2) provide the input data necessary for other decision support systems. MDS is composed of two main modules: (1) geology prediction model (GPM) module which is based on Neural-Autoregressive Hidden Markov Model and (2) excavation method selection module to select appropriate excavation method based on GPM result. In order to validate the proposed model, a microtunneling project: Zhong-he drainage water tunnel in Taiwan, was used as a case study. The result shows that the MDS model achieves these objectives to a satisfactory degrese.

Tunnels are vital options for modern transportation system. At the same time, tunnels are expensive underground structures where a variety of risks are encounters in every phase of the project delivery process. Comprehensive and realistic tunneling plans must strive for optimal decision that minimize time and cost while addressing important tunneling risks (Likhtruangsilp & Ioannou, 2004). One of the most important decisions in tunneling is to determine the optimal sequence of excavation method and support system along the tunnel profile (see Fig. 1), so that the time and cost of tunnel construction can be optimized. However, these decisions are greatly influenced by geologic uncertainty and variability. Generally soil conditions are unknown because soil samples taken from vertical boreholes show only the soil present in the discrete borehole location. Therefore, it contributes the project uncertainty (Riwanpura, AbouRizk, & Allouche, 2003).Uncertainty behavior of ground condition is inevitably in tunneling construction process. Many studies have been conducted to build decision support system (DSS) for tunneling projects. Chung, Abraham, and Gokhale (2004) build DSS for micro-tunneling. It is used to evaluate whether micro-tunneling will be economically feasible and suggest appropriate micro-tunneling methods. A concept for decision support in the selection of a type of shield tunneling machine and of an appropriate construction technology is presented by Kakoto and Skibniewski (1991). Likhtruangsilp and Ioannou (2004) and Karam, Karam, and Einstein (2007) build DSS based on three interrelated models: probabilistic geologic prediction, probabilistic tunnel cost estimates and risk sensitive dynamic decision model. The decision support system proposed in this paper is focused on probabilistic geologic prediction model (GPM). However, the result of the prediction model can be used as input data for other decision support systems such as equipment and support selection and scheduling support system. Several geological prediction models have been developed. Chan (1981), Ioannou (1987) and Sutanto (2008) used Discrete-state Markov process and Bayesian updating for modeling uncertainty of geological parameters along the tunnel line. Adi and Leu (2009) using Hidden Markov Model (HMM) to predict the geology parameters in the microtunneling project; however in their model, the relationship among soil parameters are assumed independence. Hu and Huang (2007) used 2D conditional Markov process to predict the soil transition and get the probabilistic risk index by Monte Carlo simulation. Neural Network approaches is also used by several researchers to predict geological condition in the tunnel construction, such as: geological hazards at the tunnel face (Alimoradi, Ali, Reza, Mojtaba, & Fshin, 2008) and prediction of tunnel settlement (Santos & Celestino, 2008). Neural Networks are suitable for use in the prediction of geological parameters because it can analyze non-linear patterns and trends common to geology. The geological prediction model which is proposed in this paper was designed using Hybrid Neural-Autoregression Hidden Markov Model (Neural-ARHMM). This model is a refinement of an earlier model that used Bayesian updating (Ioannou, 1987) and the Hidden Markov Model (HMM) (Adi & Leu, 2009). Its architecture is a combination of the Autoregressive Hidden Markov Model (AR-HMM) and iterative Back Propagation Neural Network (BPNN), and expectedly produces more accurate predictions of geological profiles.

As illustrated by given example, the Microtunneling Decision Support System (MDS) successfully formalizes and expedites the prediction procedure. The Neural ARHMM model also can accurately predict a geologic description that reflects the uncertainty. Hybrid Neural-ARHMM final output, in the form of geology class probabilistic profile, can be directly used as input for other decision support system such MS Project™.