FUEL DELIVERY NETWORK OPTIMIZATION FOR GAZPROM NEFT COMPANY
|
INTRODUCTION 5
CHAPTER 1. OVERVIEW OF APPROACHES TO TRANSPORTATION PROBLEM 9
1.1. Transportation problem and its importance 9
1.2. Review of optimization problems 10
1.3 Methods of solving transportation problem 13
1.4. Review of existing Python libraries 17
1.5. Summary of chapter 1 19
CHAPTER 2. METHODOLOGY FOR FUEL DELIVERY NETWORK OPTIMIZATION 21
2.1. Data description 21
2.2. Data exploration 23
2.3. Mathematical model 28
2.4. Definition of optimization problem 35
2.5. Choice of solvers 36
2.6. Overview of overall approach 36
2.7. Technical implementation 37
2.8. Summary of chapter 2 46
CHAPTER 3. ANALYSIS AND EVALUATION OF RESULTS 48 Comparison of results based on developed criteria 48
3.2. Approach to visualization of results 56
3.3. Discussion on user interaction with model 61
3.4. Summary of chapter 3 63
CONCLUSION 65
LIST OF REFERENCES 68
APPENDIX 71
CHAPTER 1. OVERVIEW OF APPROACHES TO TRANSPORTATION PROBLEM 9
1.1. Transportation problem and its importance 9
1.2. Review of optimization problems 10
1.3 Methods of solving transportation problem 13
1.4. Review of existing Python libraries 17
1.5. Summary of chapter 1 19
CHAPTER 2. METHODOLOGY FOR FUEL DELIVERY NETWORK OPTIMIZATION 21
2.1. Data description 21
2.2. Data exploration 23
2.3. Mathematical model 28
2.4. Definition of optimization problem 35
2.5. Choice of solvers 36
2.6. Overview of overall approach 36
2.7. Technical implementation 37
2.8. Summary of chapter 2 46
CHAPTER 3. ANALYSIS AND EVALUATION OF RESULTS 48 Comparison of results based on developed criteria 48
3.2. Approach to visualization of results 56
3.3. Discussion on user interaction with model 61
3.4. Summary of chapter 3 63
CONCLUSION 65
LIST OF REFERENCES 68
APPENDIX 71
In recent years, the topic of effective resources distribution and transportation organization in particular has attracted attention of many corporations and researchers alike. There are more than 10 thousand research papers written and published according to EBSCO on the topics of resource allo¬cation and the transportation problem. А well-built logistics model helps to lower transportation costs, which is especially relevant if these constitute a large proportion of a company’s overall expenses. The latter is true for an oil and gas industry, where operations include fuel delivery on a regular basis.
This thesis paper addresses a fuel delivery network optimization for Gazprom Neft company and involves a development of an automated model of fuel transportations from oil bases to petrol stations. The scope of this work is on the level of FLD (Fuel & Logistics Department), which has a mission of guaranteeing minimal cost of oil products at salespoint by timely fulfilling the needs to the full extent, while preserving the product quality. Therefore, the business problem itself is focused on minimizing operational expenses (OPEX) on logistics. To be more specific, the fuel delivery network includes pairing of oil bases and petrol stations and the related volumes needed to be delivered from first to second. For each pair there are concrete costs calculated based on tariffs; these depend on the distance, volumes, and type of the product. Thus, the model was supposed to match oil bases and petrol stations, so that the overall OPEX is minimal.
Although this task could be performed manually, the business representatives were interested in creating a model which would automate this process and save time and labor costs. At the same time, there were several restrictions that should be taken into account. Firstly, there was concrete demand for each brand at each petrol station that was to be fully satisfied. Secondly, oil bases had a certain volume of each product available that could not be exceeded. Thirdly, there was a so-called «one to one» limitation, which meant that each petrol station must get all brands from one and only one oil base. Finally, management could set a loading threshold for an oil base, so that its supply was used up to a certain degree. As an expected result the model optimally distributes volumes of each product between oil bases and petrol stations, considering all limitations and ideally providing a result in 25 minutes or less; the latter was the requirement set by the business representatives.
Prior to creating a model, a literature review was carried out to specify the main groups of methods related to the solution of a transportation problem; almost all of them refer to linear program¬ming. The methods differ in the initial mathematical meaning, optimization approach, complexity and time needed for calculation of result. Then, based on the business request, the mathematical model was formulated. After that, there was a need to understand how a concrete mathematical model dic¬tates the type of a problem to be solved; the important parameters to consider were a target function, limitations and variable data types. Finally, specific tools (such as python libraries, packages, solvers) were chosen to technically solve the specific problem. Overall, the following research goal was for¬mulated: to determine the most suitable methods and tools given existing business requirements. The business goal of the thesis was to minimize transportation costs (OPEX) for delivery network of Gaz¬prom Neft company....
This thesis paper addresses a fuel delivery network optimization for Gazprom Neft company and involves a development of an automated model of fuel transportations from oil bases to petrol stations. The scope of this work is on the level of FLD (Fuel & Logistics Department), which has a mission of guaranteeing minimal cost of oil products at salespoint by timely fulfilling the needs to the full extent, while preserving the product quality. Therefore, the business problem itself is focused on minimizing operational expenses (OPEX) on logistics. To be more specific, the fuel delivery network includes pairing of oil bases and petrol stations and the related volumes needed to be delivered from first to second. For each pair there are concrete costs calculated based on tariffs; these depend on the distance, volumes, and type of the product. Thus, the model was supposed to match oil bases and petrol stations, so that the overall OPEX is minimal.
Although this task could be performed manually, the business representatives were interested in creating a model which would automate this process and save time and labor costs. At the same time, there were several restrictions that should be taken into account. Firstly, there was concrete demand for each brand at each petrol station that was to be fully satisfied. Secondly, oil bases had a certain volume of each product available that could not be exceeded. Thirdly, there was a so-called «one to one» limitation, which meant that each petrol station must get all brands from one and only one oil base. Finally, management could set a loading threshold for an oil base, so that its supply was used up to a certain degree. As an expected result the model optimally distributes volumes of each product between oil bases and petrol stations, considering all limitations and ideally providing a result in 25 minutes or less; the latter was the requirement set by the business representatives.
Prior to creating a model, a literature review was carried out to specify the main groups of methods related to the solution of a transportation problem; almost all of them refer to linear program¬ming. The methods differ in the initial mathematical meaning, optimization approach, complexity and time needed for calculation of result. Then, based on the business request, the mathematical model was formulated. After that, there was a need to understand how a concrete mathematical model dic¬tates the type of a problem to be solved; the important parameters to consider were a target function, limitations and variable data types. Finally, specific tools (such as python libraries, packages, solvers) were chosen to technically solve the specific problem. Overall, the following research goal was for¬mulated: to determine the most suitable methods and tools given existing business requirements. The business goal of the thesis was to minimize transportation costs (OPEX) for delivery network of Gaz¬prom Neft company....
To sum up, the thesis paper was devoted to fuel delivery network optimization for Gazprom Neft company and involved a development of an automated model of fuel transportation from oil bases to petrol stations, considering several business restrictions. The business goal of the study, which was to minimize transportation costs (OPEX) for the Gazprom Neft delivery network, was successfully fulfilled. To achieve this, certain objectives listed below were accomplished. Also, an answer was received to a research question related to the study of which of the existing methods in the linear programming area and related approaches is mostly applicable to the fuel delivery transpor¬tation problem in terms of current business restrictions. Such a combination of solver and library is currently CBC from PuLP library based on the Branch and Cut method (Table 15) — research goal is achieved. All in all, the goal of the thesis — to determine the most suitable methods and tools given existing business requirements for minimization of transportation costs for delivery network of Gaz¬prom Neft company — was successfully completed.
In the first part of the paper, based on the overview of the theoretical background, optimization problems were investigated, to which transportation problems may belong and their taxonomy was compiled (Figure 3). Then the methods for solving them were summarized (Figure 4), and a mind map of libraries and solvers was created (Figure 5), from which the appropriate ones can be chosen based on the current type of problem.
In the second chapter, the description of secondary data provided by Gazprom Neft company was given (Table 2). Results of data exploration and summarization were provided in Tables 3-6 and in Figures 7 and 8. Then, the mathematical model of the current transportation problem with con¬straints was formulated (formulas 15-22). Also, some variations of the mathematical model were pre¬sented in formulas 23-25. It was found out that this optimization problem refers to the Mixed Integer Linear Problem, or MILP, because the model contains continuous and discrete variables (Figure 9). Suitable libraries and solvers were selected for this type of problem based on the approach developed in the theoretical part (Figure 10). As the model is designed to be flexible, in Table 7 the configuration of the model was described: the user was able to choose origins to set the lower bound of supply to, both kind of lower and upper bounds for supply, list of regions and date. A few experiments were conducted: variations of the mathematical model were tested in Table 8; all the solvers were tested on a part of the dataset, with the results present in Table 9; it should be mentioned that not all the solvers from Figure 10 could be successfully implemented; for more details, refer to Table 10. At the end of the second chapter, the final solution was given; the whole dataset was tested on CBC solvers from PuLP and CVXPY libraries; both of these tools provided the result with the costs lower than the baseline specified by the company, and the time of performance close to the required time limit of 25 minutes.
In the last chapter, a set of criteria was developed to evaluate the results yielded with the usage of different libraries and solvers. Libraries were compared in Tables 11-15 based on criteria such as flexibility, interactivity, quality of documentation; and solvers — based on the speed and accuracy of the results. In order to combine different criteria, a set of weights was applied. All the criteria were discussed with the company’s representatives. As a result, the best libraries (CVXPY, PuLP) and solvers for solving the MILP transportation problem (CBC from PuLP, CVXPY) were selected. In addition, a visualization in Python of such important aspects as business value, the solution itself, etc. was made. In Figures 12-18 the possible approach to the visualization of model’s results that one can use is presented. Finally, the discussion was performed, where the details on user interaction with model were described in the form of a flowchart (Figure 19) with possible options for error handling. This flowchart can be used for debugging.
As for the managerial implications of the results of this study, the resulting model of fuel transportation from oil bases to petrol stations, which is more advanced than the previously existing one, can be applied at the FLD (Fuel and Logistics Department) level of Gazprom Neft company. This model meets business requirements and can automate the process, as well as save time and labor costs. Thus, the provided solution, which is focused on minimizing operational costs (OPEX) for logistics, may help to guarantee the mission of the department to minimize the cost of oil products at salespoint by timely fulfilling the needs to the full extent, while maintaining the product quality. It should be noted that different solutions can be made based on the model’s assumptions; moreover, the model can highlight some areas for development inside the company. For instance, if no lower bound is specified and the model underloads a particular oil base, it might be the sign of managerial issue (high tariffs). In addition, exploratory analysis performed in this paper revealed that some oil bases have more supply when they can potentially deliver to petrol stations (see Table 6); this might be the sign that for such oil bases the network of transportations could be expanded. Finally, the overview of tools and methods provided in this paper can be used by those who have to solve similar tasks (for example, in the retail and transport sector) and have various professional backgrounds...
In the first part of the paper, based on the overview of the theoretical background, optimization problems were investigated, to which transportation problems may belong and their taxonomy was compiled (Figure 3). Then the methods for solving them were summarized (Figure 4), and a mind map of libraries and solvers was created (Figure 5), from which the appropriate ones can be chosen based on the current type of problem.
In the second chapter, the description of secondary data provided by Gazprom Neft company was given (Table 2). Results of data exploration and summarization were provided in Tables 3-6 and in Figures 7 and 8. Then, the mathematical model of the current transportation problem with con¬straints was formulated (formulas 15-22). Also, some variations of the mathematical model were pre¬sented in formulas 23-25. It was found out that this optimization problem refers to the Mixed Integer Linear Problem, or MILP, because the model contains continuous and discrete variables (Figure 9). Suitable libraries and solvers were selected for this type of problem based on the approach developed in the theoretical part (Figure 10). As the model is designed to be flexible, in Table 7 the configuration of the model was described: the user was able to choose origins to set the lower bound of supply to, both kind of lower and upper bounds for supply, list of regions and date. A few experiments were conducted: variations of the mathematical model were tested in Table 8; all the solvers were tested on a part of the dataset, with the results present in Table 9; it should be mentioned that not all the solvers from Figure 10 could be successfully implemented; for more details, refer to Table 10. At the end of the second chapter, the final solution was given; the whole dataset was tested on CBC solvers from PuLP and CVXPY libraries; both of these tools provided the result with the costs lower than the baseline specified by the company, and the time of performance close to the required time limit of 25 minutes.
In the last chapter, a set of criteria was developed to evaluate the results yielded with the usage of different libraries and solvers. Libraries were compared in Tables 11-15 based on criteria such as flexibility, interactivity, quality of documentation; and solvers — based on the speed and accuracy of the results. In order to combine different criteria, a set of weights was applied. All the criteria were discussed with the company’s representatives. As a result, the best libraries (CVXPY, PuLP) and solvers for solving the MILP transportation problem (CBC from PuLP, CVXPY) were selected. In addition, a visualization in Python of such important aspects as business value, the solution itself, etc. was made. In Figures 12-18 the possible approach to the visualization of model’s results that one can use is presented. Finally, the discussion was performed, where the details on user interaction with model were described in the form of a flowchart (Figure 19) with possible options for error handling. This flowchart can be used for debugging.
As for the managerial implications of the results of this study, the resulting model of fuel transportation from oil bases to petrol stations, which is more advanced than the previously existing one, can be applied at the FLD (Fuel and Logistics Department) level of Gazprom Neft company. This model meets business requirements and can automate the process, as well as save time and labor costs. Thus, the provided solution, which is focused on minimizing operational costs (OPEX) for logistics, may help to guarantee the mission of the department to minimize the cost of oil products at salespoint by timely fulfilling the needs to the full extent, while maintaining the product quality. It should be noted that different solutions can be made based on the model’s assumptions; moreover, the model can highlight some areas for development inside the company. For instance, if no lower bound is specified and the model underloads a particular oil base, it might be the sign of managerial issue (high tariffs). In addition, exploratory analysis performed in this paper revealed that some oil bases have more supply when they can potentially deliver to petrol stations (see Table 6); this might be the sign that for such oil bases the network of transportations could be expanded. Finally, the overview of tools and methods provided in this paper can be used by those who have to solve similar tasks (for example, in the retail and transport sector) and have various professional backgrounds...
