The Use of Different Models to Investigate the Threat of the Third Line of the Gas Transmission Network by Domain Instability

The area under study (Tehran gas line 3) is one of the most important and sensitive areas in terms of threats to gas supply due to the instability of domains and tectonic activities. In this research, to achieve the research objectives, different data from different sources and different criteria were used. For example, geology, height, distance from the slope fault, slope direction, distance from the river, land use, soil, distance from the road, precipitation, land cover, and height were used. Risk assessment using five fuzzy models, network analysis, Fuzzy network, multilayer perceptron, and random forest method were analyzed. Because the traditional methods of risk assessment are based on mathematical functions and need more knowledge of experts and are less practical, intelligent systems were used which, in addition to being easy to use and analyzing relationships, provided more appropriate results. The results of the investigations showed the comparison of landslide and earthquake risk with different models shows that the risk of landslide is higher with the RF model. Still, with the use of the ANP model, the studied area shows relatively higher risk. In the Fuzzy model, the highest percentage belongs to the low-risk class and the lowest percentage belongs to the medium-risk class. In the fuzzy-ANP model, the relatively high-risk class shows the highest percentage and the high-risk class shows the lowest amount in the range. In the MLP model, most of the study range has medium risk and the high-risk class range has the lowest amount in the range. Therefore, in the RF model, the highest percentage belongs to the relatively high-risk class and the lowest percentage belongs to the low-risk class. According to the evaluations made from the results of the ANP model, the systematic error (MBE) of this model is -0.20336 and the absolute error of the model is 0.209895. The RMSE error was 0.131107. According to the evaluations of the results of the Fuzzy model, the systematic error (MBE) of this model is -0.23687 and the absolute error of the model is 0.25511. The RMSE error rate was 0.162122.

Risk assessment in threats that have a spatial aspect and are related to the characteristics of the environment and the pipeline bed is considered essential. Due to the nature place of risk, it seems that it is possible to establish a connection between the process of risk estimation and the geographic information system, and by using different models, the high-risk areas can be specified. According to the issues, in this research, the process of environmental risk estimation has been investigated with a geographic information system and using hybrid-fuzzy algorithms. After positive results, by assessing the risk of gas pipelines, valuable information such as risky components can be determined and a suitable reaction and strategy can be used to reduce and even eliminate it. To achieve the goal, the appropriate technique has been used in the research. which can assess the existing risks more accurately and reliably. Also, planners and managers should act with a wider horizon and a lower risk factor toward the optimal management of gas transmission lines.

In this research, five models were analyzed for risk assessment. These models were fuzzy analysis, network analysis, fuzzy network, multi-layer perceptron, and random forest method. Given that the traditional methods of risk assessment are based on They are mathematical functions and lack the knowledge of practical meter experts, intelligent systems were used which, in addition to easy use and relationship analysis, provided more appropriate results. In this research, the network analysis method was used. The Multilayer Perceptron (MLP) model was also used in this research. This method requires less investigation in estimation or statistical methods to analyze the accuracy of data. The most important advantages of MLP are high learning potential, robustness against noise, non-linearity, parallelism, error tolerance, and high capabilities in task generalization. The overall goal of this model is to find a system to minimize the total error for the relevant training data by The training algorithm. The networks between these layers are with different weight values in the interval [1 and -1]. The result of input values, weighted values, and bias values are obtained from equation (1)based on the aggregated performance.
Equation (1)                                               
n shows the total number of input points, I_(i) the input variable, β_j the bias value, ω_ij the connection weight.
Random forest model (RF) was another model used in this research. In the structure of the RF method, the importance of the variables was determined in the model and the variables that have a greater role in each tree and the final model were identified. It was considered from the bag and these data played the role of experimental data to evaluate that tree. Based on this model, the most weight is given to the rainfall criterion and the least to the soil criterion. The measured values were used in the test stage. To validate the model, the statistical indices of root mean square error (RMSE), mean error of exploitation (MBE), and mean absolute error (MAE) were used. The relations of these indices are in the form of equations 2 to 4:
Equation(2)                 RMSE =       Equation (3) MBE =                Equation(4) MAE =                
In the above relationships, O_i and P_i are the observed and estimated values at time i, and t is the number of days.

In this study, it was tried by using different models, while the level of risk is checked, a comparison is also made between the models. In the scope of the study, The comparison of landslide and earthquake risk with different models shows that the risk of landslide and earthquake is higher based on the RF and ANP models. The analyses conducted by other researchers show that the MLP model and the ANES model show better results. The results show the difference between the output values of the model and the values. The result of smart tracking is as target points for model evaluation. The MBE error shows whether the model error is generally positive or negative, that is, whether the model estimates the data more than the real values or less than the real values. This error includes any systematic errors in the design, collection, analysis, interpretation, and dissemination of data that lead to incorrect estimates. This error is one of the systematic errors in that the increase of sample points does not affect the reduction or increase of the error. The systematic error (MBE) of this model is 0.002812.

In the ANP model, parts of the pipeline based on earthquake criteria are 0.363 km long with a low vulnerability is 20.317 km. Also, 258.60 km is with relatively high vulnerability and 29.062 km is with high vulnerability. Based on the results of the review of landslide criteria of the ANP model, 0 % of the area is in the low-risk class, 17.28% is in the risk class average, 73.14% in the relatively high class, and 9.58% in high-risk class. According to the landslide criterion using the ANP 008 model, 19.19 km of the studied area with medium vulnerability, 80.454 km with relatively high vulnerability, and 10.538 km with high vulnerability. According to the landslide index results of the MLP model, 9.78% of the area is in the low-risk class, 47.17% in the medium-risk class, 36.95% in the It is relatively high and 6.10% is in the high-risk class. According to these results, most of the region is in the class with relatively high risk. Based on the results of the MLP model earthquake criteria, it was found that 5.093 km with low vulnerability, 29.48 km with moderate vulnerability, 713.7 km 42 km with relatively high vulnerability, and 32.714 km with high vulnerability. Therefore, according to the results, it can be said that in areas with high risk, it is necessary to use higher class pipes, periodic studies and investigation of physical factors and an environment that is effective in causing damage should be done to reduce their vulnerability.