Predicting Environmental Impact of Hazardous Liquid Pipeline Accidents: Application of Intelligent Systems
Publication: Journal of Environmental Engineering
Volume 146, Issue 2
Abstract
In case of failure, hazardous liquid pipelines can have adverse environmental consequences. This study presents a method to predict the occurrence of certain environmental impacts resulting from hazardous liquid pipeline accidents. Explanatory variables, including pipe diameter, commodity transported, and incident area type, are used to train an adaptive neuro-fuzzy inference system (ANFIS). Three impact types are analyzed: water contamination, soil contamination, and impact on wildlife. Results show that the model can accurately predict whether a pipeline segment with given design characteristics could lead to adverse environmental impacts due to failure (14%, 6%, and 3% error for soil and water contamination and impact on wildlife, respectively). This model can be used in pipeline design and risk management planning to minimize the potential for environmental consequences. However, more comprehensive and robust reporting requirements beyond simple occurrence would improve our ability to prioritize these mitigative actions.
Introduction
In the United States (US), the hazardous liquid pipeline network has grown by 15% in miles and 30% by operator count since 2010 (PHMSA 2018a). The Pipeline and Hazardous Material Safety Administration (PHMSA) regulates approximately of hazardous liquid pipelines and a total of 529 hazardous liquid pipeline operators.
Pipelines have the lowest failure rates among fuel transportation methods, on the order of and accidents per km per year for onshore and offshore pipelines, respectively (Belvederesi 2017; Green and Jackson 2015; Singleton 2013). However, the receiving environments, communities, and oil companies can be significantly impacted by a single accident (Belvederesi et al. 2018; TransCanada 2017; PHMSA 2018b; EPA 2015). The objective of this study is to predict environmental consequences due to hazardous liquid pipeline accidents based on pipeline explanatory variables, such as commodity transported, accident location, nominal diameter, nominal wall thickness, and material-release method. These explanatory variables were selected following a thorough statistical analysis of the relationships between pipeline design and failure (Belvederesi et al. 2018), which informs on the data availability and significance. Moreover, this set of explanatory variables represents the basic pipeline design and the information most commonly gathered by regulatory agencies after pipeline accidents worldwide (Belvederesi 2017), which helps model reproducibility in other countries. An adaptive neuro-fuzzy inference system (ANFIS) was developed to predict water and soil contamination and adverse impact on wildlife based on the aforementioned explanatory variables. Conventional approaches to system modeling, which include probabilistic analysis, are often unsuitable for complex and uncertain systems, especially in environmental analysis (Suparta and Alhasa 2016; PreventionWeb 2018). By using a fuzzy inference system (FIS) and artificial neural networks (ANN) together, it is possible to control and overcome the complexity and uncertainty of these systems (Jang 1991, 1993). However, when using FIS, there are no mathematical methods to transform the experience and knowledge of human experts into the required if–then rule-based approach, and the lack of adaptability of the learning algorithm in tuning membership functions to minimize the error often limits the applicability of FIS. To address these limitations, ANFIS was employed in this study. ANFIS is classified as artificial intelligence (AI) because its capability for processing nonlinear and complex information is similar to the human brain. ANFIS uses fuzzy conditional statements (if–then rules) to capture the imprecise modes of reasoning behind the human ability to make decisions in an environment governed by uncertainty and imprecision. The set of if–then rules is based on data, which represent human knowledge derived from past experience. Membership functions must be developed from the available data on which the model is based (fuzzification). While membership functions used in FIS are generated by human expertise, membership functions for ANFIS are generated by adaptive neural networks (NN). In addition to ranges, ANFIS also generates probability distribution types and parameters.
The literature offers some examples where ANFIS or similar approaches were adopted as intelligent systems in different fields of application. Alizadeh et al. (2018) demonstrated how machine learning can assist environmental management and monitoring tools by providing accurate predictions for water quality parameters. Wijayasekara and Manic (2014) studied how to develop membership functions using classical statistical methods. They investigated data set coverage, complementarity, and relative dissymmetry, and used this information as metrics for understandability in the generated membership functions (Wijayasekara and Manic 2014). Liu and Zhang (2015) developed an ANFIS model-based data-driven approach to control the welding process. They concluded that, compared with linear models, to control welding processes—welding speed and weld pool characteristics parameters—an iterative local ANFIS model using k-means clustering provides better results for modeling performance (Liu and Zhang 2015). Sharkawy et al. (2014) compared three algorithms to predict the surface roughness of end-milling process: a radial basis function neural network (RBFN), ANFIS, and a genetically evolved fuzzy inference system (G-FIS). Among these techniques, they found that RBFN is the most successful at performing surface roughness assessment (Sharkawy et al. 2014). Dastorani et al. (2010) studied how to predict missing flow data of gauging stations using data from nearby hydrometric stations to train an ANN and an ANFIS model. In this study, the researchers concluded that ANFIS is superior for estimating missing data, although ANN produces results with a high level of accuracy (Dastorani et al. 2010). Badde et al. (2015) compared FIS and ANFIS to predict compressive strength of ready-mix concrete (RMC). ANFIS with Gaussian membership functions could predict the 28-day compression strength of ready-mix concrete with satisfactory performance, according to the authors (Badde et al. 2015). ANFIS was also applied in the oil and gas industry to assess and optimize pipeline system performance, predict corrosion rates of steel pipelines, and reduce energy consumption for gas transportation. For example, Baghmolaei et al. (2014) developed both FIS and ANFIS models to minimize fuel consumption of gas turbines, and compared these two techniques with ANN. The results of modeling by ANN, FIS, and ANFIS showed that ANN with the genetic algorithm code has the best performance (Baghmolaei et al. 2014). He et al. (2012) studied how RBF and ANFIS could predict the corrosion rate of underground steel pipelines. Results showed that ANFIS can more accurately predict corrosion rate considering changing corrosion factors (He et al. 2012). Baghban et al. (2017) proposed a study that aimed to develop an ANFIS-based model to predict the breakthrough curves for rhamnolipid adsorption over activated carbon. The models’ accuracy (absolute average deviation of 1.98%) shows that ANFIS would be a more reliable method to predict breakthrough curves compared with ANN and group method data handling (GMDH). Shahrak et al. (2018) applied ANFIS to model water vapor uptakes in different porous metal–organic framework materials, showing a high coefficient of determination (). A multilevel adaptive neuro-fuzzy inference system was developed by Nabavi-Pelesaraei et al. (2019) to predict various environmental, energy, and economic indices of large-scale food production systems, showing high accuracy with coefficients of determination between 0.91 and 0.98.
Predicting environmental consequences due to hazardous pipeline failures can be challenging. To develop a predictive model, it is necessary to have a robust, complete, and informative data set containing information gathered from past accidents. Belvederesi et al. (2017, 2018) investigated the limitations that derive from inadequate data and their implications for accident response actions, and conducted a thorough statistical analysis of trends in hazardous liquid pipeline accidents in relation to the pipeline design characteristics. Major findings led to the development of the model presented in this study.
Methods
Data Source
The Pipeline and Hazardous Material Safety Administration (PHMSA), US Department of Transportation (DOT), regulates approximately of the country’s inter- and intrastate pipelines (Belvederesi et al. 2017) and gathers information about oil and gas pipeline accidents that occur in the US. Although PHMSA has collected and made available information about hazardous liquid pipelines since 1986, the data available are temporally inconsistent in terms of reporting criteria. The definition of an accident has changed over time and, for this reason, several pre-2010 accidents were not included in the data set because they did not meet the reporting criteria. The quality and quantity of information provided by PHMSA have increased over time. To ensure consistency and significance for model training and, therefore, predictions, this study focuses on information provided by PHMSA between January 1, 2010, and October 31, 2018, because it was collected under the same requirements and reporting criteria, enabling a more robust analysis of hazardous liquid pipeline failures in the US (Belvederesi et al. 2018). Moreover, this study considers both offshore and onshore gathering and transmission hazardous liquid pipelines regulated by PHMSA.
Details regarding the environmental consequences of pipeline accidents are collected by PHMSA in 21 descriptive database fields, including wildlife impact (i.e., fish, birds, and terrestrials), soil contamination and remediation, and water contamination (i.e., surface water, groundwater, drinking water, and public water). However, the database reports information in the form of a Yes/No statement, and details regarding the number and species of animals impacted, the volume of soil contaminated, and the concentration of chemicals and contaminants are missing. To quantitatively predict the environmental consequences of pipeline accidents, input data regarding the magnitude of the impact must be provided to the model. However, there are no data available to quantify the degree of severity of environmental impacts of pipeline accidents (e.g., area affected by the spill or number of animals involved); therefore, the information collected about whether the accidents’ impacts on wildlife, water, and soil were due to pipeline failures can be used only as descriptive input to the model.
Supporting details regarding PHMSA’s hazardous liquid scope, definitions, reporting criteria, and other general information can be found in title 49, subtitle B, chapter 1 (subchapter D), part 195 of the Code of Federal Regulations [C.F.R. (2019)].
The analysis presented in this paper considers five input variables used to predict the environmental outputs:
1.
Commodity transported
a.
Biofuels (biodiesel, and fuel-grade ethanol);
b.
Carbon dioxide;
c.
Crude oil;
d.
High-vapor liquids (HVL) and other flammable or toxic fluids that are a gas at ambient conditions, such as liquefied petroleum gas (LPG) and natural gas liquid (NGL), anhydrous ammonia, natural gasoline, refinery-grade propylene (RGP), etc.; and
e.
Refined and/or petroleum product (non-HVL) that is a liquid at ambient conditions (diesel, jet fuel, kerosene, crude condensate, etc.).
2.
Location
a.
Aboveground;
b.
Tank, including attached appurtenances;
c.
Transition area (soil/air interface); and
d.
Underground.
3.
Nominal diameter over nominal wall thickness (D/t)
4.
Maximum operating pressure (MOP)
5.
Release mode
a.
Leak (seal or packing, connection failure, crack, pinhole, etc.);
b.
Mechanical puncture;
c.
Other (details given in the database, such as “Pump malfunction due to oiler not operating correctly” or “Release occurred from a damaged pipe tee connection on the 2” sump discharge line which connects to the return header and the Transmix line”);
d.
Overfill or overflow; and
e.
Rupture.
In this study, the output is environmental impact, divided into three categories: (1) adverse effect on wildlife, (2) water contamination, and (3) soil contamination. The output type for the three models is presented in the same form provided by PHMSA (Yes/No statement on impact on wildlife, soil, and water) because no information regarding the magnitude of the consequences is provided. For calculation purposes, the outcome “Yes” is given as 1, and the outcome “No” is given as 2.
ANFIS
ANFIS can be described as a feedforward neural network with multilayer perceptron (Suparta and Alhasa 2016; Jang 1993). This means that it does not have a feedback link within its architecture, and data and incoming signals are allowed to move in one direction only. The learning algorithm plays the important role of modifying the parameters in the network to adapt to its environment. Two types of learning processes have been widely adopted in the literature: supervised and unsupervised. In supervised learning, a set of input variables is entered into the model as a sample pattern that has been marked or labeled. Each incoming signal to the single neuron spreads along the network until it reaches the end layer of neurons in the output layer. In the final layer, the output is generated and compared with the output pattern. Conversely, unsupervised learning does not have guidelines or target output in the learning process. The network simply receives many samples of inputs and then associates the sample set randomly to some classes or categories. In other words, the output will have some sort of similar characteristics to the input stimulus.
This study considers ANFIS using a grid partitioning and classification method (supervised learning algorithm) and adopting the Takagi-Sugeno type inference system. A hybrid algorithm that combines least-squares estimator and the gradient descent method is adopted. This means that, during the training process, a forward and backward propagation algorithm from Layer 1 to Layer 5 and vice versa (Fig. 3) serves to correct the parameters of the membership functions (in this study, membership functions are set as Gaussian and triangular distributions). At the same time, the gradient descent method is used to find the nonlinear function minimum, resulting from the weights generated by the fuzzy rules. Fig. 3 shows the structure of the fuzzy reasoning mechanism for this study, where it is possible to see where the forward–backward propagation applies from Layers 1 to 5 and back to Layer 1. Fig. 4 shows the application of the gradient descent method, where the minimum error represents the bottom of the curve.
In Layer 1, for each input variable there is a set of membership functions that adapts to function parameters. The output from each node is a degree of membership value that is given by the input of the membership functions. , , and are triangular membership functions [Eq. (2)] because of the categorical nature of the input variables (categorical variables can only take one of a limited and fixed set of values in a group) and because it leads to shorter computational times, and and are Gaussian membership functions [Eq. (1)] because data for these input variables are ordinalwhere = degree of membership functions for the given fuzzy set; = one of the input variables; and , , and = parameters of a membership function that can change the shape of the membership function.
(1)
(2)
In Layer 2, every node is fixed (nonadaptive), and the circle node is labeled as . The output node results from the multiplication of incoming signals and is delivered to the next node. It represents the firing strength for each rule. The T-norm operator with general performance (AND) is applied to obtain the output, because all the explanatory variables occur simultaneouslywhere = output that represents the firing strength of each rule; and = number of membership functions per variable.
(3)
In Layer 3, every node is fixed (nonadaptive) and the circle node is labeled as . Each node is the calculation of the ratio between the th rules firing strength and the sum of all firing strengths. It is also called the normalized firing strength
(4)
In Layer 4, every node is an adaptive node to an output, with a node function defined aswhere = normalized firing strength from the previous layer; and () = a parameter in the node. The parameters in this layer are referred to as consequent parameters.
(5)
In Layer 5, the single node is a fixed (nonadaptive) node that computes the overall output as the summation of all the incoming signals from the previous node. This circle node is labeled as .
(6)
The first and the fourth layer contain the parameter that can be modified over time. The first layer contains a nonlinear set of premises parameters, while the fourth layer includes linear consequent parameters. To update both parameter types, a learning algorithm is necessary so that they can adapt to the model’s environment. A hybrid algorithm is used in this study.
The hybrid learning algorithm can be divided into two parts: the forward propagation and the backward propagation. During the forward propagation, the premises parameters (, , and ) in the first layer must be steady. A recursive least square (RLS) estimator method is applied to repair the consequent parameters in the fourth layer. Because the consequent parameters are linear, the RLS estimator method can be applied to accelerate the convergence rate in the hybrid learning process. After the consequent parameters are obtained, the backward propagation allows for comparison between the generated output and the actual output through the adaptive network input of initial data. The error identified during the comparison between the generated and actual output is propagated back to the first layer. At the same time, premises parameters in the first layer are updated using gradient descent. One level of hybrid learning is called epoch. With the hybrid learning algorithm, which combines RLS estimation and the gradient descent methods, the convergence can be reached faster than using the backpropagation algorithm only, because the dimensional search space is reduced. Further details regarding the hybrid learning algorithm can be found in Suparta and Alhasa (2016).
To ensure its best performance, reduce calculation errors, and avoid model overfitting, two data sets are entered to train and validate ANFIS, and one additional data set is used to test the model. The training data set contains the sample of data used to fit the model. The model sees and learns from these data. The validating data set, which in this study differs from model testing, includes a sample of data used to provide an unbiased evaluation of a model fit on the training data set while tuning model parameters [, , and in Eqs. (1) and (2)]. The evaluation becomes more biased when the validation data set is incorporated into the model configuration. In other words, the model does not use this data set to train itself, but uses this information to consider uncertainty in the input–output relationship and avoid overfitting. In this study, training and validating data sets have the same sample size to obtain a more thorough representation of the training data and improve the model’s capability to deal with input–output variability and uncertainty. Hence, each accident training set (i.e., pipe D/t, commodity, location, pressure, and release mode) corresponds an accident validating set to guarantee avoiding model overfitting, because not every pipeline with a given set of design characteristics always leads to a certain type of consequence in case of failure.
The testing data set, which differs from the validating data set, includes the sample of data used to provide an unbiased evaluation of a final model fit on the training data set. It contains data that span the various classes that the model would face when used in the real world to make predictions. Table 1 provides the general assumptions for this model and, in particular, the sample size for each ANFIS model and the model settings adopted for all three impact types analyzed (water and soil contamination and impact on wildlife). The ANFIS settings in Table 1 were selected because their combination led to the highest model accuracy and, therefore, the most reliable predictions over the testing data set.
Model assumption/setting | Soil contamination | Water contamination | Impact on wildlife |
---|---|---|---|
Number of samples for training | 357 | 358 | 356 |
Number of samples for validating | 357 | 357 | 356 |
Number of samples for testing | 100 | 100 | 100 |
Membership functions type and number | |||
Commodity type | Triangular, 4 | Triangular, 4 | Triangular, 4 |
Accident location | Triangular, 4 | Triangular, 4 | Triangular, 4 |
D/t | Gauss, 5 | Gauss, 5 | Gauss, 5 |
MOP | Gauss, 5 | Gauss, 5 | Gauss, 5 |
Release mode | Triangular, 4 | Triangular, 4 | Triangular, 4 |
Number of rule nodes | 1,600 | 1,600 | 1,600 |
Number of epochs | 10 | 10 | 10 |
Output type | Linear | Linear | Linear |
The model error for each environmental impact type is calculated using the AIP and AVP as follows:where = predicted value; = actual value; and = sample size of the testing data set or number of input–output relationships. The average validity/invalidity percent provides an accurate tool to estimate the model performance in case of categorical output (i.e., Yes/No fashion), as other model performance tests, such as the coefficient of determination or relative error, generally pertain to ordinal output variables.
(7)
(8)
Results
In this section, results are reported for each output type: soil contamination, water contamination, and impact on wildlife. Table 2 summarizes AIP and AVP for each model. Similarly, Table 3 offers a detailed review of ANFIS performance in predicting whether the environmental contamination occurred according to actual data. These results are discussed in the next section.
Model identification | AIP (%) | AVP (%) |
---|---|---|
Soil contamination | 14 | 86 |
Water contamination | 6 | 94 |
Impact on wildlife | 3 | 97 |
Outcome | Soil contamination (%) | Water contamination (%) | Impact on wildlife (%) |
---|---|---|---|
Actual “No” | 26 | 76 | 88 |
Actual “Yes” | 74 | 24 | 12 |
Predicted “No” | 18 | 72 | 85 |
Predicted “Yes” | 82 | 28 | 15 |
Wrongly predicted “Yes” | 11 | 5 | 3 |
Wrongly predicted “No” | 3 | 1 | 0 |
Linguistic variables (biofuels, crude oil, HVL, refined petroleum products, carbon dioxide, aboveground, tank, transition area, underground, leak, mechanical puncture, other, rupture, and overfill or overflow) are converted to categorical variables for calculation purposes. The following legend of these variable names helps in interpreting the results reported in this section:
1.
Commodity type
a.
Biofuels
b.
Crude oil
c.
HVL
d.
Refined petroleum product
e.
Carbon dioxide
2.
Location
a.
Aboveground
b.
Tank
c.
Transition area
d.
Underground
3.
Release mode
a.
Leak
b.
Mechanical puncture
c.
Other
d.
Rupture
e.
Overfill or overflow
4.
Output
a.
Yes
b.
No
Tables 4–6 provide the testing data set used to verify the model’s performance for each analysis in the sections below. Each line contained in Tables 4–6 represents a hazardous liquid pipeline failure. Each line, or failure, provides commodity transported, location where the accident occurred, diameter-to-wall-thickness ratio, operating pressure, and mode with which the material was released. The predicted output is compared with the actual output and reported as a green tick in case of matching outcomes and a red X in case of mismatch (1 stands for Yes and 2 stands for No).
Year | Commodity type | Location | D/t | MOP (psig) | Release mode | Actual output | Predicted output | Predicted matches with actual? |
---|---|---|---|---|---|---|---|---|
2010 | 4 | 4 | 21.43 | 1,453 | 1 | 1 | 1 | Yes |
2010 | 3 | 4 | 45.88 | 1,440 | 2 | 2 | 2 | Yes |
2010 | 3 | 4 | 63.93 | 863 | 1 | 2 | 1 | No |
2010 | 3 | 4 | 23.94 | 1,440 | 2 | 2 | 2 | Yes |
2010 | 3 | 4 | 45.88 | 1,440 | 2 | 2 | 2 | Yes |
2010 | 3 | 4 | 42.47 | 1,424 | 2 | 2 | 2 | Yes |
2010 | 3 | 4 | 26.5 | 1,440 | 1 | 2 | 2 | Yes |
2010 | 3 | 4 | 55.29 | 1,198 | 3 | 1 | 1 | Yes |
2010 | 4 | 4 | 29.45 | 720 | 2 | 1 | 1 | Yes |
2010 | 2 | 4 | 42.18 | 800 | 4 | 1 | 1 | Yes |
2011 | 2 | 4 | 59.36 | 285 | 1 | 1 | 1 | Yes |
2011 | 2 | 4 | 45.88 | 400 | 1 | 1 | 1 | Yes |
2011 | 2 | 4 | 55.17 | 810 | 1 | 1 | 1 | Yes |
2011 | 2 | 4 | 43 | 275 | 1 | 1 | 1 | Yes |
2011 | 3 | 4 | 59.86 | 1,000 | 2 | 1 | 1 | Yes |
2011 | 2 | 1 | 27.17 | 150 | 1 | 1 | 1 | Yes |
2011 | 4 | 1 | 64 | 285 | 1 | 1 | 1 | Yes |
2011 | 2 | 4 | 16.88 | 275 | 1 | 1 | 1 | Yes |
2011 | 4 | 4 | 35.84 | 1,296 | 2 | 1 | 1 | Yes |
2011 | 4 | 4 | 40.86 | 1,195 | 2 | 1 | 1 | Yes |
2012 | 4 | 4 | 42.47 | 1,200 | 4 | 1 | 1 | Yes |
2012 | 2 | 4 | 51.28 | 725 | 2 | 1 | 1 | Yes |
2012 | 4 | 4 | 35.24 | 1,200 | 2 | 1 | 1 | Yes |
2012 | 3 | 4 | 51 | 1,218 | 1 | 1 | 1 | Yes |
2012 | 2 | 4 | 96 | 285 | 1 | 2 | 2 | Yes |
2012 | 3 | 4 | 35.24 | 1,573 | 1 | 2 | 2 | Yes |
2012 | 4 | 4 | 34.5 | 1,349 | 1 | 1 | 1 | Yes |
2012 | 2 | 4 | 21.43 | 285 | 1 | 1 | 1 | Yes |
2013 | 2 | 4 | 80 | 936 | 1 | 2 | 1 | No |
2013 | 2 | 1 | 34.5 | 500 | 1 | 1 | 1 | Yes |
2013 | 2 | 4 | 45.37 | 1,480 | 1 | 1 | 1 | Yes |
2013 | 2 | 4 | 64 | 1,058 | 1 | 1 | 1 | Yes |
2013 | 4 | 4 | 39.38 | 1,865 | 4 | 1 | 1 | Yes |
2013 | 2 | 4 | 30.25 | 1,390 | 1 | 1 | 1 | Yes |
2013 | 2 | 4 | 34.5 | 508 | 1 | 1 | 1 | Yes |
2013 | 3 | 4 | 73.06 | 1,025 | 2 | 2 | 1 | No |
2013 | 2 | 4 | 21.43 | 500 | 2 | 2 | 1 | No |
2013 | 4 | 4 | 30.07 | 250 | 2 | 1 | 1 | Yes |
2013 | 4 | 4 | 46.54 | 273 | 1 | 1 | 1 | Yes |
2013 | 2 | 4 | 29.45 | 366 | 2 | 1 | 1 | Yes |
2014 | 4 | 4 | 53.33 | 690 | 1 | 2 | 1 | No |
2014 | 2 | 4 | 25.64 | 800 | 2 | 1 | 1 | Yes |
2014 | 2 | 4 | 64 | 275 | 1 | 1 | 1 | Yes |
2014 | 2 | 4 | 48 | 0 | 1 | 2 | 1 | No |
2014 | 2 | 4 | 48 | 285 | 1 | 1 | 1 | Yes |
2014 | 4 | 4 | 35.24 | 1,440 | 1 | 1 | 1 | Yes |
2014 | 2 | 1 | 24.84 | 275 | 1 | 1 | 1 | Yes |
2014 | 2 | 4 | 28.88 | 308 | 1 | 2 | 1 | No |
2014 | 4 | 4 | 28.88 | 720 | 1 | 1 | 1 | Yes |
2014 | 2 | 4 | 50 | 720 | 1 | 1 | 1 | Yes |
2014 | 2 | 4 | 32 | 275 | 1 | 1 | 1 | Yes |
2014 | 3 | 4 | 24.84 | 1,198 | 1 | 1 | 1 | Yes |
2014 | 2 | 4 | 80 | 936 | 4 | 1 | 1 | Yes |
2014 | 2 | 4 | 64 | 1,315 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 24.84 | 397 | 1 | 1 | 1 | Yes |
2015 | 3 | 4 | 35.24 | 1,440 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 32 | 275 | 1 | 1 | 1 | Yes |
2015 | 2 | 4 | 28.67 | 1,440 | 1 | 1 | 2 | No |
2015 | 4 | 4 | 45.88 | 1,252 | 1 | 1 | 1 | Yes |
2015 | 2 | 4 | 64 | 275 | 1 | 1 | 1 | Yes |
2015 | 2 | 4 | 25.5 | 1,200 | 1 | 2 | 1 | No |
2015 | 2 | 4 | 57.89 | 1,016 | 1 | 1 | 1 | Yes |
2015 | 4 | 4 | 43 | 1,076 | 1 | 1 | 1 | Yes |
2015 | 2 | 4 | 38.46 | 1,360 | 2 | 1 | 1 | Yes |
2015 | 4 | 4 | 113.88 | 657 | 1 | 1 | 2 | No |
2015 | 3 | 4 | 46.87 | 1,440 | 1 | 1 | 1 | Yes |
2015 | 2 | 4 | 48 | 400 | 1 | 1 | 1 | Yes |
2016 | 4 | 4 | 128.11 | 584 | 1 | 1 | 1 | Yes |
2016 | 4 | 4 | 42.67 | 175 | 2 | 1 | 1 | Yes |
2016 | 2 | 4 | 24.84 | 397 | 1 | 1 | 1 | Yes |
2016 | 2 | 4 | 51.28 | 1,440 | 1 | 2 | 1 | No |
2016 | 2 | 4 | 48 | 1,440 | 2 | 1 | 1 | Yes |
2016 | 3 | 4 | 42.70 | 1,440 | 1 | 2 | 1 | No |
2016 | 3 | 4 | 59.92 | 1,440 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 18.26 | 250 | 1 | 1 | 1 | Yes |
2016 | 4 | 4 | 31.91 | 1,102 | 1 | 1 | 1 | Yes |
2016 | 3 | 4 | 35.23 | 1,440 | 1 | 2 | 2 | Yes |
2016 | 2 | 1 | 80 | 285 | 1 | 1 | 1 | Yes |
2016 | 2 | 4 | 64 | 285 | 1 | 1 | 1 | Yes |
2016 | 3 | 4 | 42.55 | 1,361 | 1 | 2 | 2 | Yes |
2016 | 4 | 4 | 26.78 | 1,300 | 3 | 1 | 1 | Yes |
2016 | 2 | 4 | 64.10 | 1,212 | 1 | 2 | 2 | Yes |
2017 | 4 | 4 | 16.88 | 275 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 106.76 | 701 | 1 | 1 | 1 | Yes |
2017 | 3 | 4 | 32 | 1,090 | 1 | 1 | 1 | Yes |
2017 | 3 | 4 | 67.99 | 1,480 | 1 | 1 | 1 | Yes |
2017 | 2 | 4 | 36.53 | 1,464 | 1 | 1 | 1 | Yes |
2017 | 2 | 4 | 36.53 | 1,000 | 2 | 1 | 1 | Yes |
2017 | 4 | 4 | 51 | 1,468 | 1 | 1 | 1 | Yes |
2017 | 3 | 4 | 23.94 | 1,440 | 1 | 2 | 2 | Yes |
2017 | 4 | 4 | 44.87 | 1,162 | 1 | 1 | 1 | Yes |
2017 | 4 | 4 | 31.91 | 720 | 2 | 1 | 1 | Yes |
2017 | 2 | 4 | 36.53 | 1,480 | 2 | 1 | 1 | Yes |
2017 | 2 | 4 | 53.33 | 575 | 1 | 1 | 1 | Yes |
2018 | 2 | 4 | 53.19 | 780 | 1 | 1 | 1 | Yes |
2018 | 4 | 4 | 63.83 | 880 | 1 | 1 | 1 | Yes |
2018 | 2 | 4 | 120.99 | 275 | 1 | 1 | 2 | No |
2018 | 3 | 4 | 24.84 | 1,080 | 4 | 2 | 1 | No |
2018 | 2 | 4 | 28.88 | 640 | 1 | 1 | 1 | Yes |
2018 | 3 | 4 | 42.55 | 1,440 | 1 | 1 | 1 | Yes |
Year | Commodity type | Location | D/t | MOP (psig) | Release mode | Actual output | Predicted output | Predicted matches with actual? |
---|---|---|---|---|---|---|---|---|
2010 | 2 | 4 | 92.53 | 809 | 4 | 2 | 2 | Yes |
2010 | 2 | 2 | 53.33 | 285 | 1 | 2 | 2 | Yes |
2010 | 4 | 4 | 27.4 | 275 | 1 | 2 | 2 | Yes |
2010 | 3 | 4 | 45.88 | 1,440 | 2 | 2 | 2 | Yes |
2010 | 4 | 4 | 56.94 | 0 | 3 | 1 | 1 | Yes |
2010 | 4 | 4 | 62.81 | 1,050 | 1 | 1 | 1 | Yes |
2010 | 2 | 4 | 51 | 492 | 1 | 2 | 2 | Yes |
2010 | 4 | 4 | 64 | 1,035 | 1 | 1 | 1 | Yes |
2010 | 2 | 4 | 48 | 200 | 1 | 2 | 2 | Yes |
2010 | 2 | 4 | 51 | 275 | 1 | 2 | 2 | Yes |
2010 | 2 | 4 | 26.78 | 852 | 1 | 2 | 2 | Yes |
2010 | 2 | 4 | 42.18 | 800 | 4 | 2 | 2 | Yes |
2011 | 2 | 4 | 64.06 | 262 | 1 | 2 | 2 | Yes |
2011 | 4 | 4 | 42.47 | 1,200 | 2 | 1 | 1 | Yes |
2011 | 4 | 4 | 51 | 1,186 | 1 | 1 | 1 | Yes |
2011 | 3 | 4 | 31.91 | 1,130 | 1 | 2 | 2 | Yes |
2011 | 2 | 4 | 42.55 | 1,000 | 2 | 2 | 2 | Yes |
2011 | 4 | 1 | 35.84 | 1,147 | 1 | 1 | 1 | Yes |
2011 | 2 | 4 | 12.99 | 600 | 1 | 2 | 2 | Yes |
2011 | 2 | 1 | 76.92 | 125 | 1 | 1 | 1 | Yes |
2011 | 2 | 4 | 16.88 | 275 | 1 | 2 | 2 | Yes |
2011 | 2 | 4 | 45.88 | 1,000 | 2 | 2 | 2 | Yes |
2011 | 2 | 4 | 33.75 | 500 | 1 | 2 | 2 | Yes |
2012 | 2 | 4 | 23.66 | 408 | 2 | 2 | 2 | Yes |
2012 | 3 | 4 | 42.47 | 1,440 | 1 | 2 | 2 | Yes |
2012 | 2 | 4 | 73.18 | 1,378 | 4 | 2 | 2 | Yes |
2012 | 3 | 4 | 42.47 | 1,335 | 1 | 2 | 2 | Yes |
2012 | 4 | 1 | 35.24 | 462 | 1 | 1 | 2 | No |
2012 | 4 | 4 | 36.36 | 200 | 1 | 1 | 1 | Yes |
2012 | 2 | 4 | 64 | 794 | 2 | 2 | 2 | Yes |
2012 | 4 | 4 | 51 | 900 | 1 | 2 | 2 | Yes |
2013 | 4 | 4 | 37.33 | 250 | 1 | 1 | 1 | Yes |
2013 | 3 | 4 | 73.06 | 1,025 | 2 | 1 | 1 | Yes |
2013 | 1 | 4 | 51.62 | 1,953 | 1 | 1 | 1 | Yes |
2013 | 2 | 4 | 64 | 913 | 1 | 2 | 2 | Yes |
2013 | 3 | 4 | 53.19 | 1,307 | 4 | 2 | 2 | Yes |
2013 | 3 | 4 | 42.47 | 1,440 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 23.66 | 720 | 1 | 2 | 2 | Yes |
2013 | 3 | 1 | 23.66 | 720 | 3 | 1 | 1 | Yes |
2013 | 4 | 4 | 32 | 75 | 3 | 1 | 1 | Yes |
2013 | 2 | 4 | 64 | 824 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 34 | 720 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 64 | 945 | 2 | 2 | 2 | Yes |
2013 | 2 | 4 | 42.55 | 1,440 | 1 | 1 | 1 | Yes |
2014 | 2 | 1 | 57.69 | 275 | 1 | 2 | 2 | Yes |
2014 | 4 | 4 | 24.84 | 485 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 40 | 285 | 1 | 2 | 2 | Yes |
2014 | 4 | 4 | 24.84 | 0 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 43 | 1,172 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 28.88 | 308 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 43 | 790 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 80 | 809 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 40.86 | 285 | 1 | 2 | 2 | Yes |
2014 | 4 | 4 | 35.24 | 1,440 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 23.66 | 700 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 22.96 | 568 | 4 | 1 | 1 | Yes |
2015 | 4 | 4 | 43 | 1,076 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 40 | 1,142 | 4 | 1 | 1 | Yes |
2015 | 2 | 4 | 48 | 784 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 113.88 | 657 | 1 | 1 | 1 | Yes |
2015 | 2 | 4 | 27.39 | 10 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 16.88 | 275 | 1 | 2 | 2 | Yes |
2015 | 2 | 1 | 27.39 | 366 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 43 | 1,076 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 40 | 427 | 1 | 2 | 2 | Yes |
2015 | 1 | 4 | 39.20 | 2,220 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 80 | 526 | 1 | 2 | 1 | No |
2015 | 4 | 4 | 42.55 | 615 | 2 | 2 | 1 | No |
2015 | 2 | 4 | 64 | 275 | 1 | 2 | 2 | Yes |
2016 | 3 | 4 | 51.28 | 1,440 | 1 | 2 | 2 | Yes |
2016 | 4 | 4 | 31.91 | 1,102 | 1 | 2 | 1 | No |
2016 | 1 | 4 | 23.94 | 1,765 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 68.57 | 275 | 1 | 2 | 2 | Yes |
2016 | 3 | 4 | 51.28 | 1,198 | 1 | 2 | 1 | No |
2016 | 2 | 4 | 42.67 | 275 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 71.17 | 794 | 1 | 2 | 2 | Yes |
2016 | 3 | 4 | 57.18 | 1,300 | 4 | 2 | 2 | Yes |
2016 | 2 | 4 | 96 | 0 | 3 | 2 | 1 | No |
2016 | 2 | 4 | 64.10 | 720 | 1 | 2 | 2 | Yes |
2016 | 1 | 4 | 34.5 | 1,671 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 64 | 1,160 | 1 | 2 | 2 | Yes |
2016 | 3 | 4 | 42.70 | 1,440 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 38.46 | 361 | 1 | 2 | 2 | Yes |
2017 | 2 | 1 | 48 | 1,440 | 1 | 2 | 2 | Yes |
2017 | 4 | 4 | 40 | 275 | 3 | 2 | 2 | Yes |
2017 | 2 | 4 | 48 | 285 | 1 | 2 | 2 | Yes |
2017 | 4 | 4 | 48 | 990 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 36.53 | 1,480 | 2 | 1 | 1 | Yes |
2017 | 2 | 1 | 90.67 | 1,400 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 83.33 | 799 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 24.84 | 640 | 1 | 1 | 1 | Yes |
2017 | 2 | 4 | 64 | 610 | 2 | 1 | 1 | Yes |
2017 | 3 | 1 | 12.29 | 1,450 | 1 | 1 | 1 | Yes |
2017 | 4 | 4 | 16 | 275 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 24 | 275 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 24.84 | 1,012 | 1 | 2 | 2 | Yes |
2018 | 3 | 4 | 64 | 1,440 | 1 | 2 | 2 | Yes |
2018 | 4 | 4 | 36.53 | 1,150 | 1 | 1 | 1 | Yes |
2018 | 2 | 4 | 64.26 | 0 | 1 | 1 | 1 | Yes |
2018 | 4 | 4 | 32 | 950 | 1 | 2 | 2 | Yes |
Year | Commodity type | Location | D/t | MOP (psig) | Release mode | Actual output | Predicted output | Predicted matches with actual? |
---|---|---|---|---|---|---|---|---|
2010 | 2 | 4 | 45.88 | 1,440 | 2 | 2 | 2 | Yes |
2010 | 3 | 4 | 55.29 | 1,345 | 4 | 2 | 2 | Yes |
2010 | 2 | 4 | 53 | 500 | 2 | 2 | 2 | Yes |
2010 | 3 | 4 | 55.29 | 1,198 | 3 | 1 | 1 | Yes |
2010 | 4 | 4 | 29.45 | 720 | 2 | 1 | 1 | Yes |
2010 | 2 | 4 | 51 | 275 | 1 | 2 | 2 | Yes |
2010 | 3 | 4 | 42.47 | 1,440 | 1 | 2 | 2 | Yes |
2010 | 4 | 4 | 45.88 | 960 | 2 | 2 | 2 | Yes |
2010 | 4 | 4 | 26.78 | 625 | 1 | 2 | 2 | Yes |
2010 | 2 | 4 | 53.33 | 275 | 1 | 2 | 2 | Yes |
2010 | 2 | 4 | 70.51 | 1,050 | 4 | 2 | 1 | No |
2011 | 2 | 4 | 38.53 | 780 | 1 | 2 | 2 | Yes |
2011 | 3 | 4 | 40 | 672 | 1 | 2 | 2 | Yes |
2011 | 2 | 1 | 76.92 | 125 | 1 | 2 | 1 | No |
2011 | 3 | 4 | 34.5 | 1,440 | 3 | 2 | 2 | Yes |
2011 | 2 | 4 | 64 | 990 | 1 | 2 | 2 | Yes |
2011 | 2 | 4 | 23.66 | 464 | 1 | 2 | 2 | Yes |
2011 | 3 | 4 | 34.5 | 1,440 | 1 | 2 | 2 | Yes |
2011 | 2 | 4 | 96 | 300 | 1 | 2 | 2 | Yes |
2011 | 3 | 4 | 26.5 | 1,050 | 2 | 2 | 2 | Yes |
2011 | 4 | 4 | 60.61 | 25 | 1 | 2 | 2 | Yes |
2011 | 4 | 4 | 42.49 | 1,150 | 2 | 1 | 1 | Yes |
2012 | 2 | 4 | 28.88 | 560 | 1 | 2 | 2 | Yes |
2012 | 4 | 4 | 35.24 | 720 | 1 | 2 | 2 | Yes |
2012 | 4 | 4 | 49.26 | 1,342 | 4 | 2 | 2 | Yes |
2012 | 4 | 4 | 51 | 900 | 1 | 2 | 2 | Yes |
2012 | 4 | 4 | 32 | 1,440 | 1 | 2 | 2 | Yes |
2012 | 2 | 4 | 122.09 | 150 | 1 | 2 | 2 | Yes |
2012 | 2 | 4 | 70.51 | 1,061 | 4 | 1 | 1 | Yes |
2012 | 3 | 4 | 48 | 1,440 | 1 | 2 | 2 | Yes |
2012 | 2 | 4 | 18.99 | 275 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 45.88 | 546 | 1 | 2 | 2 | Yes |
2013 | 4 | 4 | 48 | 720 | 1 | 2 | 2 | Yes |
2013 | 4 | 4 | 45.88 | 1,632 | 4 | 2 | 2 | Yes |
2013 | 4 | 4 | 128.11 | 275 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 48 | 285 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 51 | 795 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 34.5 | 508 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 48 | 365 | 1 | 2 | 2 | Yes |
2013 | 2 | 4 | 64 | 920 | 1 | 2 | 2 | Yes |
2013 | 3 | 4 | 55.29 | 1,198 | 1 | 2 | 2 | Yes |
2013 | 4 | 4 | 51 | 1,298 | 2 | 1 | 1 | Yes |
2013 | 4 | 4 | 49.26 | 1,176 | 1 | 2 | 2 | Yes |
2014 | 2 | 1 | 23.64 | 425 | 3 | 1 | 1 | Yes |
2014 | 4 | 4 | 28.88 | 720 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 53.33 | 275 | 3 | 1 | 1 | Yes |
2014 | 4 | 4 | 24.84 | 0 | 1 | 2 | 2 | Yes |
2014 | 3 | 4 | 24.84 | 1,198 | 1 | 2 | 2 | Yes |
2014 | 2 | 4 | 28.85 | 714 | 1 | 2 | 2 | Yes |
2014 | 4 | 4 | 52.95 | 952 | 1 | 2 | 2 | Yes |
2014 | 4 | 4 | 64 | 275 | 1 | 2 | 2 | Yes |
2014 | 4 | 4 | 21.43 | 125 | 1 | 2 | 2 | Yes |
2014 | 3 | 4 | 51.28 | 1,198 | 1 | 2 | 2 | Yes |
2014 | 3 | 4 | 51.28 | 1,336 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 42.55 | 326 | 1 | 2 | 2 | Yes |
2015 | 2 | 1 | 28.37 | 225 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 40 | 1,142 | 4 | 2 | 2 | Yes |
2015 | 2 | 4 | 34 | 875 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 9.17 | 960 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 28.88 | 500 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 45.88 | 0 | 2 | 2 | 1 | No |
2015 | 2 | 4 | 48 | 400 | 1 | 2 | 2 | Yes |
2015 | 2 | 1 | 21.43 | 285 | 1 | 2 | 2 | Yes |
2015 | 1 | 4 | 39.20 | 2,220 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 32 | 600 | 1 | 2 | 2 | Yes |
2015 | 4 | 4 | 80 | 275 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 32 | 250 | 1 | 2 | 2 | Yes |
2015 | 2 | 4 | 26.78 | 751 | 1 | 2 | 2 | Yes |
2015 | 3 | 4 | 29.45 | 911 | 1 | 2 | 2 | Yes |
2016 | 4 | 4 | 128.11 | 584 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 77.72 | 1,337 | 1 | 2 | 2 | Yes |
2016 | 2 | 1 | 20 | 275 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 0.05 | 784 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 64 | 285 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 34.5 | 304 | 1 | 2 | 2 | Yes |
2016 | 3 | 4 | 30.25 | 1,220 | 1 | 2 | 2 | Yes |
2016 | 4 | 4 | 59.11 | 1,156 | 1 | 2 | 2 | Yes |
2016 | 2 | 1 | 64 | 285 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 28.67 | 1,440 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 27.4 | 275 | 1 | 2 | 2 | Yes |
2016 | 2 | 4 | 34.5 | 305 | 3 | 1 | 1 | Yes |
2016 | 4 | 4 | 128.11 | 541 | 3 | 1 | 1 | Yes |
2017 | 4 | 4 | 21.43 | 750 | 1 | 2 | 2 | Yes |
2017 | 2 | 2 | 11.94 | 0 | 1 | 1 | 1 | Yes |
2017 | 2 | 4 | 48 | 492 | 1 | 2 | 2 | Yes |
2017 | 3 | 4 | 36.53 | 1,440 | 2 | 2 | 2 | Yes |
2017 | 3 | 4 | 56 | 1,440 | 2 | 2 | 2 | Yes |
2017 | 2 | 4 | 62.5 | 285 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 19.23 | 275 | 1 | 2 | 2 | Yes |
2017 | 1 | 4 | 36 | 1,628 | 1 | 1 | 1 | Yes |
2017 | 2 | 4 | 106.76 | 701 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 39.41 | 750 | 2 | 2 | 2 | Yes |
2017 | 3 | 4 | 31.91 | 1,250 | 1 | 2 | 2 | Yes |
2017 | 2 | 4 | 31.91 | 535 | 1 | 2 | 2 | Yes |
2017 | 2 | 1 | 42.67 | 275 | 1 | 2 | 2 | Yes |
2018 | 4 | 4 | 38.46 | 1,191 | 2 | 1 | 1 | Yes |
2018 | 2 | 4 | 28.88 | 640 | 1 | 2 | 2 | Yes |
2018 | 2 | 4 | 32 | 408 | 1 | 2 | 2 | Yes |
2018 | 2 | 4 | 72 | 275 | 1 | 2 | 2 | Yes |
2018 | 2 | 4 | 120.99 | 275 | 1 | 2 | 2 | Yes |
Prediction of Soil Contamination
The model used for predicting soil contamination due to hazardous liquid pipeline accidents can forecast environmental impact with 86% accuracy against testing data. Table 4 reports the input–output set used to test the model along with predicted and actual values.
Prediction of Water Contamination
The model used for predicting water contamination in case of hazardous liquid pipeline accidents can forecast environmental impact with 94% accuracy against testing data. Table 5 reports the input–output set used to test the model along with predicted and actual values.
Prediction of Adverse Impact on Wildlife
The model used for predicting adverse effects on wildlife in case of hazardous liquid pipeline accidents can forecast environmental impact with 97% accuracy against testing data. Table 6 reports the input–output set used to test the model along with predicted and actual values.
Conclusion and Discussion
A predictive model was developed to evaluate whether an environmental impact is likely to occur in case of hazardous liquid pipeline failures. Explanatory variables, including commodity transported, accident location, diameter-to-wall-thickness ratio, maximum operating pressure, and material release mode were used to train and validate an ANFIS model. Three different outcomes were investigated: soil and water contamination and adverse effects on wildlife.
Results show that the model is capable of accurately predicting outcomes for the three impact types, as follows:
1.
Soil contamination is predicted with 86% accuracy (14 AIP). The model incorrectly predicted soil contamination for 11% of the total number of failures, and did not predict soil contamination 3% of the time when contamination did occur.
2.
Water contamination is predicted with 94% accuracy (6 AIP). The model predicted water contamination 5% of the time when no actual contamination was recorded, and did not predict water contamination 1% of the time when contamination did occur.
3.
Adverse impact on wildlife is predicted with 97% accuracy (3 AIP). The model predicted adverse effects on wildlife 3% of the time when no actual adverse effect was recorded. The model was capable of predicting with 100% accuracy those cases where no impact on wildlife occurred.
Data used to test the model show that approximately 65% of hazardous liquid pipeline accidents resulted in soil contamination, 32% resulted in water contamination, and 15% resulted in adverse impacts on wildlife. In addition, 57% of incorrectly predicted outcomes for soil contamination occurred on pipelines transporting crude oil and located underground. This combination of the two parameter values is also the most recurrent in the training database (52% of accidents). Similarly, the results of the model for soil contamination show that 79% of incorrectly predicted cases include the combination of being located underground and with leak as the release mode. These instances are the most commonly recurring combinations of this pair of parameters in the training data set (73% of the training data set). This means that the model has a poor capacity to predict rare outcomes because, when there is a set of explanatory variables that rarely leads to an outcome, the model cannot make an accurate prediction for that rare event. This limitation could be further investigated by focusing the analysis on those sets of explanatory variables that rarely lead to soil contamination. The results for the water contamination model do not show any particular trend or commonalities similar to the soil contamination and wildlife impact models. In fact, the sets of variables that lead the model to the incorrect outcome are always diverse. Finally, the low number of accidents in the training data set that had adverse effects on wildlife (4.8%) could explain why the model for predicting adverse impacts on wildlife was 100% accurate. This should be further investigated by using a different set of accident data to test the model, although PHMSA reports few cases of impact on wildlife during the past 10 years.
The selected set of explanatory variables used in this study represents the optimal combination to obtain the most accurate results. However, accident location and release mode are less informative variables than others because more than 80% of accidents involved pipelines that failed underground, and more than 50% of the pipelines failed by leaking. The low variance shown by these variables would lead to the assumption that they do not provide substantial information to the model, although, by including them in the analysis, the model performs more accurately.
Predicting adverse environmental impacts of hazardous pipeline accidents is challenging in the absence of an adequate, complete, and informative data set that gathers information from past accidents. Because the model presented in this study is data driven, a major limitation is presented in the training/validating data set; missing data and incorrect reports introduce conflicting scenarios into the model that, as a consequence, it is not able to accurately predict, especially in the case of rare events. Currently, comprehensive design information must be collected by regulators after pipeline accidents, including pipeline material; despite this, a large number of missing data points makes it challenging to include certain design variables as inputs for the model. Belvederesi et al. (2018) show that, over time, regulators are becoming stricter with regard to reporting and the quality and quantity of information collected after an accident. For this reason, future iterations of this model should include more explanatory variables as they become available, such as pipeline material, and the model should be updated with newly collected data to ensure that predictions will be accurate in addition to the model being computationally efficient.
During pipeline planning and design, this model could provide an alternative method to predict the hazards and, consequently, the level of risk that a pipeline would pose to the surrounding environment. In those instances where the failure of hazardous liquid pipelines would be likely to lead to adverse environmental outcomes, additional protective actions should be adopted for both planned and existing pipelines. Protective measures can include installing casings on water-crossing pipelines and improving or reinforcing leak detection systems. Additionally, in the case of planned pipelines, an alternative combination of design characteristics should be evaluated, especially in environmentally sensitive areas.
Data Availability Statement
All data, models, or code generated or used during the study are available in a PHMSA repository online in accordance with funder data retention policies. They are available at https://www.phmsa.dot.gov/data-and-statistics/pipeline/gas-distribution-gas-gathering-gas-transmission-hazardous-liquids.
Acknowledgments
The authors acknowledge the contribution of Dr. Petr E. Komers, president of MSES, Inc., for his ongoing emotional, professional, and financial support.
References
Alizadeh, M. J., M. R. Kavianpour, M. Danesh, J. Adolf, S. Shamshirband, and K. W. Chau. 2018. “Effect of river flow on the quality of estuarine and coastal waters using machine learning models.” Eng. Appl. Comput. Fluid Mech. 12 (1): 810–823. https://doi.org/10.1080/19942060.2018.1528480.
Badde, D. S., A. K. Gupta, and V. K. Patki. 2015. “Comparison of fuzzy logic and ANFIS for prediction of compressive strength of RMC.” IOSR J. Mech. Civ. Eng. 3: 7–15.
Baghban, A., J. Sasanipour, P. Haratipour, M. Alizad, and M. Vafaee Ayouri. 2017. “ANFIS modeling of rhamnolipid breakthrough curves on activated carbon.” Chem. Eng. Res. Des. 126 (Oct): 67–75. https://doi.org/10.1016/j.cherd.2017.08.007.
Baghmolaei, M. M., M. Mahmoudy, D. Jafari, R. M. Baghmolaei, and F. Tabkhi. 2014. “Assessing and optimization of pipeline system performance using intelligent systems.” J. Nat. Gas Sci. Eng. 18 (May): 64–76. https://doi.org/10.1016/j.jngse.2014.01.017.
Belvederesi, C. 2017. Statistical analysis of oil and gas pipeline accidents with a focus on the relationship between pipeline design and accident consequences. Calgary, AB: Univ. of Calgary.
Belvederesi, C., M. S. Thompson, and P. E. Komers. 2017. “Canada’s federal database is inadequate for the assessment of environmental consequences of oil and gas pipeline failures.” Environ. Rev. 25 (4): 415–422. https://doi.org/10.1139/er-2017-0003.
Belvederesi, C., M. S. Thompson, and P. E. Komers. 2018. “Statistical analysis of environmental consequences of hazardous liquid pipeline accidents.” Heliyon 4 (11): e00901. https://doi.org/10.1016/j.heliyon.2018.e00901.
Dastorani, M. T., A. Moghadamnia, J. Piri, and M. Rico-Ramirez. 2010. “Application of ANN and ANFIS models for reconstructing missing flow data.” Environ. Monit. Assess. 166 (1–4): 421–434. https://doi.org/10.1007/s10661-009-1012-8.
EPA. 2015. “ExxonMobil Mayflower clean water settlement.” Accessed November 22, 2018. https://www.epa.gov/enforcement/exxonmobil-mayflower-clean-water-settlement.
Green, K. P., and T. Jackson. 2015. “Safety in the transportation of oil and gas: Pipelines or rail?” Accessed November 22, 2018. https://www.fraserinstitute.org/research/safety-transportation-oil-and-gas-pipelines-or-rail.
He, S., Y. Zou, D. Quan, and H. Wang. 2012. “Application of RBF neural network and ANFIS on the prediction of corrosion rate of pipeline steel in soil.” In Vol. 124 of Recent advances in computer science and information engineering. Lecture notes in electrical engineering. Edited by Z. Qian, L. Cao, W. Su, T. Wang, and H. Yang. Berlin: Springer.
Jang, J. S. 1991. “Fuzzy modeling using generalized neural networks and Kalman filter algorithm.” In Vol. 2 of Proc., 9th National Conf. on Artificial Intelligence (AAAI’91), 762–767. Palo Alto, CA: AAAI Press.
Jang, J. S. 1993. “ANFIS: Adaptive-network-based fuzzy inference system.” IEEE Trans. Syst. Man Cybern. 23 (3): 665–685. https://doi.org/10.1109/21.256541.
Liu, Y., and Y. Zhang. 2015. “Iterative local ANFIS-based human welder intelligence modeling and control in pipe GTAW process: A data-driven approach.” IEEE/ASME Trans. Mechatron. 20 (3): 1079–1088. https://doi.org/10.1109/TMECH.2014.2363050.
Nabavi-Pelesaraei, A., S. Rafiee, S. S. Mohtasebi, H. Hosseinzadeh-Bandbafha, and K. Chau. 2019. “Comprehensive model of energy, environmental impacts and economic in rice milling factories by coupling adaptive neuro-fuzzy inference system and life cycle assessment.” J. Cleaner Prod. 217 (Apr): 742–756. https://doi.org/10.1016/j.jclepro.2019.01.228.
PHMSA (Pipeline and Hazardous Materials Safety Administration). 2018a. “Pipeline mileage and facilities.” Accessed November 22, 2018. https://www.phmsa.dot.gov/data-and-statistics/pipeline/pipeline-mileage-and-facilities.
PHMSA (Pipeline and Hazardous Materials Safety Administration). 2018b. “Updates on ExxonMobil pipeline incident in Mayflower, Arkansas.” Accessed November 22, 2018. https://www.phmsa.dot.gov/foia/updates-exxonmobil-pipeline-incident-mayflower-arkansas.
PreventionWeb. 2018. “Deterministic and probabilistic risk.” Accessed November 22, 2018. https://www.preventionweb.net/risk/deterministic-probabilistic-risk.
Shahrak, M. N., M. Esfandyari, and M. Karimi. 2018. “Efficient prediction of water vapor adsorption capacity in porous metal–organic framework materials: ANN and ANFIS modeling.” J. Iran. Chem. Soc. 16 (1): 11–20. https://doi.org/10.1007/s13738-018-1476-y.
Sharkawy, A., M. El-Sharief, and M. S. Soliman. 2014. “Surface roughness prediction in end milling process using intelligent systems.” Int. J. Mach. Learn. Cybern. 5 (1): 135–150. https://doi.org/10.1007/s13042-013-0155-7.
Singleton, M. 2013. “What’s the safest way to transport oil? US transportation and state departments won’t say.” Accessed November 22, 2018. http://www.ibtimes.com/whats-safest-way-transport-oil-us-transportation-state-departments-wont-say-1172847.
Suparta, W., and K. M. Alhasa. 2016. Modeling of tropospheric delays using ANFIS. New York: Springer.
TransCanada. 2017. “TransCanada responds to oil leak in Amherst, South Dakota.” Accessed November 22, 2018. https://www.transcanada.com/en/announcements/2017-11-16transcanada-responds-to-oil-leak-in-amherst-south-dakota/.
Wijayasekara, D., and M. Manic. 2014. “Data driven fuzzy membership function generation for increased understandability.” In Proc., 2014 IEEE Int. Conf. on Fuzzy Systems (FUZZ-IEEE). Piscataway, NJ: IEEE.
Information & Authors
Information
Published In
Copyright
This work is made available under the terms of the Creative Commons Attribution 4.0 International license, http://creativecommons.org/licenses/by/4.0/.
History
Received: Mar 13, 2019
Accepted: Jun 4, 2019
Published online: Nov 26, 2019
Published in print: Feb 1, 2020
Discussion open until: Apr 26, 2020
Authors
Metrics & Citations
Metrics
Citations
Download citation
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.