Introduction: One of the destructive consequences of the earthquake was the failure of the infrastructure structures and facilities of each region, such as the fuel transmission system. In general, the fuel transmission system in cities consists of pipes buried in the ground for fuel transfer and various tanks on the ground for the storage and storage of petroleum materials. With the occurrence of earthquakes and the movement of the earth's tectonic plates, we have witnessed two main destructive phenomena: 1- failure of fuel transfer pipes 2- creating cracks in the pipes (leakage of oil materials). Based on the position of the fuel networks system, which consists of pipes buried in the soil, it is expected that after the pipe breaks and leaks, the contents inside the pipes of the fuel transmission system will move in the soil as a pollutant, and in addition to soil pollution based on the coefficient hydraulic transmission, has entered the groundwater resources and caused the pollution of these valuable resources. Considering the nature of the earthquake, which is classified as a random phenomenon, an accurate prediction of its time, place, and intensity cannot be provided, and also based on the numerous parameters involved in the performance of the fuel transmission system under seismic loading, which has wide uncertainties. There is a need to carry out extensive studies (vulnerability analysis, risk analysis) on the fuel network under different earthquake scenarios and consider getting the conditions of uncertainty assessment on the pollution of groundwater resources by the leakage of oil pollutants resulting from the failure of the fuel transmission system should be implemented.
Methods: In this research, a comprehensive model has been presented to evaluate the pollution of underground water resources based on the vulnerability of the fuel transmission pipeline under the conditions of an earthquake. In the research model, the seismic vulnerability of the buried pipeline has been developed by considering the conditions of uncertainty and using the machine learning method. The results of the validation of the developed numerical models have shown acceptable performance and high accuracy in predicting the seismic vulnerability of the pipe. Also, the aquifer pollution assessment using the DRASTIC model has been implemented. In this research, 5 main machine learning algorithms, each of which has different computational branches, have been used to develop a model for predicting the seismic vulnerability of a buried pipeline.In order to consider the conditions of uncertainty in the input data of stress prediction models and the possibility of seismic damage that directly affects the results, the conditions of uncertainty in the input data of the model have been taken into account regarding the main goal of assessing the probability of seismic vulnerability in the line The buried pipe is under different earthquake scenarios, taking into account the uncertainty in the input data values of the model, the most critical conditions for the possibility of pipe vulnerability are extracted. Based on this, 4 main parameters: 1- Depth of working, 2- Modulus of elasticity, 3- Soil density, 4- Magnitude of earthquake in numerical models based on the conditions of uncertainty in the initial input values and using the probability distribution function and the coefficient of variation. The Monte Carlo method is used for simulating critical conditions (the highest probability of pipe seismic failure) have been extracted. In this research, the number of Monte Carlo simulation repetitions is considered equivalent to 100,000 repetitions.
Results: The results of the performance of the machine learning model based on different algorithms indicate the amount of error in predicting the stress on the buried pipe compared to the exact values based on the initial data set. Based on the results, it is clear that the GPR models have a better performance than other computational models of machine learning. Based on the results, it is clear that among the algorithms of the GPR method, the Exponential GPR algorithm has the minimum error rate, so the machine learning model developed based on this algorithm has an MSE error equal to 0.00329 and the correlation R equal to 0.999. The results of the performance of the machine learning model for predicting seismic vulnerability based on different algorithms showed that GPR models (Gaussian Process Regression Models) had better performance than other computational models of machine learning and had a minimum error rate, so the machine learning model developed based on this algorithm has an MSE error equal to 0.00329 and a correlation value R equal to 0.999. Based on the results of the drastic model, the pollution of the Tehran aquifer based on the first seismic scenario (earthquake event with magnitude Mw = 5) showed that the pollution zoning of the aquifer has changed and the area of the area with moderate pollution has increased by about 9% based on the ML prediction model. According to the results of the forecasting models, which indicate the number of failures 2 and the number of leaks 15 under the first seismic scenario in the pipeline, the amount of 10% changes in the medium vulnerability level can be investigated.
Based on the results of the implementation of the second seismic scenario (earthquake with a magnitude of 6 on the Richter scale) based on the seismic damage prediction model, the ML prediction model showed an increase in the level of moderate pollution by 35% and an increase in the level of high pollution by 5%. The zoning of Tehran aquifer pollution based on the third seismic scenario (earthquake event with magnitude Mw = 7) showed that the underground water sources of Tehran city have been heavily polluted due to the leakage of fuel products from the fuel pipeline in such a way that according to the model ML, the area of low pollution level is 18%, medium pollution level is 45% and high pollution level is 37%. Based on the results obtained in the event of an earthquake with a magnitude of 7 on the Richter scale, the amount of aquifer pollution based on the ML prediction model indicates an increase of 25% in the medium pollution level and 19% in the level of high pollution. There has been a lot of pollution.