A gas system management method and system based on data analysis

Abstract

This application discloses a gas system management method and system based on data analysis, which belongs to the technical field of gas system management. It includes combining a wax deposition kinetics model with a deep neural network to design a hybrid prediction model architecture; the wax deposition kinetics model is used to describe the physical and chemical processes of wax deposition and provide basic prediction capabilities; the deep neural network is used to capture nonlinear relationships and complex patterns in the data to improve the accuracy and generalization of predictions; based on an improved particle swarm algorithm (PSO-ADE), an adaptive pigging decision engine is designed; the algorithm is initialized to generate an initial particle swarm; the algorithm is run to evaluate and update the particles according to the objective function; the optimal pigging strategy is found through iterative optimization; feedback data is collected and the algorithm is further adjusted and optimized to improve the accuracy and effectiveness of decisions. This application can optimize the pigging strategy of oil pipelines.

Description

Technical Field

[0001] The present application relates to the technical field of gas system management, and in particular to a gas system management method and system based on data analysis. Background Art

[0002] As a vital component of urban infrastructure, gas systems not only play a key role in residents' lives, industrial production, and commercial services, but also have a positive impact on environmental protection and economic development. Therefore, optimizing the management and operational efficiency of gas systems is crucial for improving urban quality of life, promoting industrial development, and fostering commercial prosperity.

[0003] With the acceleration of urbanization and the widespread use of gas, the safe and efficient management of gas systems has become increasingly important. Wax deposition in gas systems is a complex physical and chemical process that is difficult to accurately predict. Wax deposition not only affects pipeline flow capacity but can also lead to blockages, increased pressure loss, and safety hazards. Traditional wax deposition prediction methods, often based on simplified physical models, struggle to fully capture the nonlinear relationships and complex patterns in the wax deposition process, resulting in inaccurate predictions.

[0004] At the same time, developing a pigging strategy is a challenging issue. Pigging operations consume significant resources and time, and frequent pigging can damage pipelines. Therefore, developing an effective pigging strategy based on real-time wax deposition predictions to reduce unnecessary pigging operations, improve pigging efficiency, and reduce operating costs is a pressing issue. Summary of the Invention

[0005] To address these issues, this application provides a gas system management method and system based on data analysis. This method combines a wax deposition dynamics model with a deep neural network to design a hybrid prediction model architecture to improve the accuracy and generalization of wax deposition predictions. Furthermore, an adaptive pigging decision engine, based on an improved particle swarm optimization algorithm (PSO-ADE), is designed to find the optimal pigging strategy through iterative optimization.

[0006] This application provides a gas system management method and system based on data analysis, which adopts the following technical solutions:

[0007] In a first aspect, the present application provides a gas system management method based on data analysis, comprising the following steps:

[0008] A hybrid prediction model architecture was designed by combining a wax deposition kinetics model with a deep neural network. The wax deposition kinetics model describes the physical and chemical processes of wax deposition and provides basic prediction capabilities. The deep neural network captures nonlinear relationships and complex patterns in the data, improving prediction accuracy and generalization.

[0009] Design an adaptive pigging decision engine based on an improved particle swarm optimization algorithm (PSO-ADE);

[0010] Initialize the algorithm and generate the initial particle swarm;

[0011] Run the algorithm to evaluate and update particles according to the objective function;

[0012] Find the optimal pigging strategy through iterative optimization;

[0013] Collect feedback data to further adjust and optimize the algorithm to improve the accuracy and effectiveness of decision-making.

[0014] Furthermore, the step of combining the wax deposition dynamics model with the deep neural network to design a hybrid prediction model architecture also includes:

[0015] A four-field coupled model of heat, fluid, solidification and chemical reaction is constructed based on COMSOL Multiphysics. The equations to be solved include:

[0016]

[0017] Where ρ is the fluid density, which is used to represent the mass of the fluid per unit volume; u is the velocity vector, which is used to represent the movement speed of the fluid particles; t is the time; p is the pressure, which is used to represent the force acting vertically on the unit area inside the fluid; μ is the dynamic viscosity coefficient, which is used to represent the internal friction of the fluid and reflects the viscosity of the fluid; S wax The wax phase change source term is used, and an improved molecular diffusion-gel deposition dual mechanism model is used to represent the effect of wax phase change in the fluid on the flow;

[0018] Introducing dynamic grid adaptation technology to automatically refine the grid to 0.5mm resolution in areas with sudden deposition rate changes;

[0019] Train the XGBoost-GRU hybrid model, with input features including temperature gradient (ΔT), wall shear stress (τ_w), wax crystal volume fraction (φ), and output deposition rate correction factor α;

[0020] A generative adversarial network (GAN) is used to expand limited experimental data and generate a synthetic dataset covering the temperature range of -10 to 50°C to address the problem of insufficient extrapolation capability of traditional models.

[0021] Furthermore, the step of initializing the algorithm further includes:

[0022] Establish the objective function:

[0023]

[0024] Among them, C pigging is the cost of the pigging operation, which is used to represent the cost of each pigging operation; f i is the frequency of the i-th pigging operation, which is used to indicate the number of pigging operations within a certain period of time; λ is the risk penalty coefficient, which is used to indicate the degree of penalty for risk and reflects the cost of risk; δ i is the predicted sediment thickness, which is used to represent the thickness of the sediment in the pipeline predicted during the i-th pigging operation; δ safe Safe sediment thickness is used to indicate the safe thickness of sediment in the pipeline. Exceeding this thickness may cause risks.

[0025] The objective function is to find a balance between the cost of the pigging operation and the risk posed by sediment, so as to minimize the total cost;

[0026] The improved PSO-ADE algorithm is used to introduce an inertia weight adaptive mechanism to enhance the local search capability in the later stage of iteration. The formula is as follows:

[0027] w(t)=w min +(w max -w min )·e -20t / T

[0028] The formula is used to describe the change of w(t) with time t. The specific parameters have the following meanings: w min represents the minimum or stable value of w(t). When time t approaches infinity, w(t) will approach this value. max Indicates the initial maximum value of w(t). When time t = 0, the value of w(t) is w max ; t represents time; T represents the time constant, the unit of which is consistent with the time t, reflecting the speed at which w(t) decays to a stable value. The larger the time constant T, the slower the decay process.

[0029] Furthermore, it also includes:

[0030] Select typical gas system accidents as analysis objects to clarify the severity of the accident consequences;

[0031] A typical accident of the gas system is regarded as a top event, and the top event is decomposed into multiple sub-top events, wherein the sub-top events are the possible causes of the top event;

[0032] Based on the sub-top event and the top event, a fault tree logic model is constructed; the causes are decomposed layer by layer using "AND gate" (all conditions are met at the same time) and "OR gate" (any condition is met) until the basic events corresponding to the sub-top event are obtained;

[0033] Calculating the minimum cut set and structural importance coefficient, wherein the minimum cut set is the minimum combination of basic events that lead to the top event, and is used to identify key risk factors;

[0034] Structural importance coefficient

[0035] Among them, I i Indicates the value of the i-th indicator, which is a comprehensive value obtained by summation; ∑ represents the summation operation, which accumulates all items that meet the conditions; n j The number of events representing the jth minimum cut set; 2 nj-1 : Indicates that the base is 2 and the exponent is n j -1 is raised to the power of the denominator of each term.

[0036] Combine historical data to calculate the probability of occurrence of basic events and derive the probability of top events.

[0037] Furthermore, after the step of deriving the probability of the top event, the method further includes:

[0038] Obtain the ranking of key risk factors from a pre-set database;

[0039] Generate a safety checklist based on the probability of the top events and the ranking of key risk factors;

[0040] The safety checklist is pushed to the user's smart terminal.

[0041] Furthermore, the TOPSIS method is used to calculate the structural risk closeness of basic events, including:

[0042] According to the structural importance and risk probability of basic events, a matrix is ​​established and processed with forward, standardization and normalization;

[0043] The entropy weight method is used to calculate the weights of the evaluation indicators and obtain the weighted normalized indicator matrix;

[0044] Calculate the structural risk proximity of basic events, sort them according to the degree of proximity, and determine the criticality of each basic event to the top event.

[0045] Furthermore, the steps of establishing a matrix and performing forward, standardization, and normalization processing based on the structural importance and risk probability of the basic events further include:

[0046] First, establish an initial matrix A=(bij ) m×n Since the structural importance I and risk probability index have been positive, the Min-Max normalization method is used to normalize the initial matrix A=(b ij ) m×n To standardize the indicators, the formula is as follows:

[0047] y=(bA min ) / (A max -A min )

[0048] Among them, y is the normalized value; b is the original value of a certain attribute; A min is the minimum value of a certain attribute; A max The maximum value of a certain attribute;

[0049] After normalization, we get the indicator matrix B=(x ij ) m×n , the normalized calculation formula is as follows:

[0050]

[0051] Among them, x ij are normalized values.

[0052] Furthermore, after normalization, the indicator matrix B is obtained as follows: ij ) m×n After the steps, it also includes:

[0053] Based on the obtained indicator matrix B, the weights Wj of each indicator are calculated to obtain the weighted normalized indicator matrix C. The expression of the weighted normalized indicator matrix C is as follows:

[0054]

[0055] The entropy weight method is used to calculate the indicator weights to ensure the scientificity and rationality of the indicator weights. The specific calculation formula is as follows:

[0056]

[0057] Among them, W j is the entropy weight of the calculation index; H j is the entropy value of the calculation indicator; k is the Boltzmann constant; x ij is the i-th standardized and normalized value under the j-th indicator;

[0058] When calculating the closeness, take the maximum value of a certain indicator as the positive ideal point v j + , the minimum value is the negative ideal point v j -; Calculate the relative distance S between a certain item and all positive and negative ideal points i + and S i - , and then calculate the closeness S i The calculation formula is as follows:

[0059]

[0060] In the formula

[0061]

[0062] Among them, v ij is the weighted and standardized value of the jth parameter of the i-th item; S i The closer the degree of closeness is, the closer it is to the ideal state; S i Assess the criticality of basic events to top events as a benchmark.

[0063] In a second aspect, the present application provides a gas system management system based on data analysis, comprising:

[0064] A model architecture design module is used to design a hybrid prediction model architecture by combining a wax deposition kinetics model that describes the physical and chemical processes of wax deposition and provides basic prediction capabilities with a deep neural network that captures nonlinear relationships and complex patterns in the data, improving prediction accuracy and generalization capabilities.

[0065] Decision engine design module, used to design an adaptive pigging decision engine based on an improved particle swarm optimization algorithm (PSO-ADE);

[0066] Initialization module, used to initialize the algorithm and generate the initial particle swarm;

[0067] The running module is used to run the algorithm and evaluate and update the particles according to the objective function;

[0068] Iterative optimization module, used to find the optimal pigging strategy through iterative optimization;

[0069] The feedback data collection module is used to collect feedback data and further adjust and optimize the algorithm to improve the accuracy and effectiveness of decision-making.

[0070] In a third aspect, the present application provides an intelligent terminal comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the above-mentioned gas system management method based on data analysis.

[0071] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program that can be loaded by a processor and execute the above-mentioned gas system management method based on data analysis.

[0072] In summary, compared with the prior art, the above technical solution has the following beneficial effects:

[0073] The data analysis-based gas system management method described in this application, combined with a wax deposition kinetics model and a deep neural network, can more accurately predict wax deposition in gas pipelines. The wax deposition kinetics model provides a prediction basis based on physical and chemical processes, while the deep neural network can capture nonlinear relationships and complex patterns in the data, thereby improving the accuracy and generalization ability of the prediction.

[0074] Based on an improved particle swarm optimization algorithm (PSO-ADE), the solution designed an adaptive pigging decision engine. This engine can find the optimal pigging strategy through iterative optimization based on real-time data and prediction results. This helps reduce unnecessary pigging operations, improve pigging efficiency, and reduce operating costs. By collecting feedback data, the algorithm can be further adjusted and optimized, which means that the algorithm can adapt to different pipeline conditions, operating environments, and wax deposition characteristics, thereby improving the accuracy and effectiveness of decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a flow chart of a gas system management method based on data analysis in an embodiment of the present application. DETAILED DESCRIPTION

[0076] The present application is further described in detail below in conjunction with all the accompanying drawings.

[0077] The present application discloses a gas system management method and system based on data analysis, referring to Figure 1 , a gas system management method based on data analysis includes:

[0078] S101. Combine the wax deposition dynamics model with a deep neural network to design a hybrid prediction model architecture.

[0079] Specifically, the wax deposition kinetics model describes the physical and chemical processes of wax deposition, providing basic predictive capabilities. The deep neural network captures nonlinear relationships and complex patterns in the data, improving predictive accuracy and generalization. The hybrid architecture combining the wax deposition kinetics model (physical mechanism) with the deep neural network (data-driven) requires bidirectional coupling rather than simple superposition.

[0080] Wax deposition dynamics equations (such as the molecular diffusion coefficient formula and the shear peeling effect formula in the Matzain model) are used as physical constraints in the neural network, and hard constraints are implemented through residual connections or loss function design. For example, a loss term based on Fick's diffusion law is added to the output layer of the neural network to ensure that the prediction results conform to the diffusion mechanism.

[0081] Then, a parallel-serial hybrid structure is adopted to dynamically weightedly fuse the dynamic model output (such as wax deposition rate and temperature gradient) with the original sensor data (flow rate and pressure) through the attention mechanism. Multi-scale feature extraction is introduced, and a convolutional neural network (CNN) is used to process the temperature field image of the pipeline cross section. The recurrent neural network (RNN) captures the temporal dynamics and splices them with the scalar features output by the dynamic model.

[0082] A transfer learning framework is constructed for different oil characteristics (such as wax content and asphaltene ratio). Domain-Adversarial Training is used to eliminate the impact of oil quality differences on model generalization. An online learning module is deployed, and the measured wax layer thickness data after pipe cleaning operations is used to dynamically update the model parameters, forming a closed-loop optimization system.

[0083] S102. Design an adaptive pigging decision engine based on the improved particle swarm optimization algorithm (PSO-ADE).

[0084] Specifically, the particle swarm optimization algorithm (PSO) is an optimization algorithm based on swarm intelligence that simulates the foraging behavior of bird flocks and finds the optimal solution through cooperation and competition between individuals and groups. In order to improve the performance of PSO, an adaptive dynamic adjustment mechanism (ADE) is introduced to dynamically adjust parameters such as inertia weight and learning factor to balance global search capability and local search accuracy, avoid premature convergence and fall into local optimality. The adaptive pipeline cleaning decision engine is based on PSO-ADE. By combining pipeline operation data and sediment monitoring data, it dynamically adjusts the pipeline cleaning strategy to achieve automation, intelligence and efficiency of pipeline cleaning operations.

[0085] In this embodiment, the inputs to the decision engine are pipeline operation data (such as flow rate, pressure, temperature, etc.), sediment monitoring data (such as sediment thickness, composition, etc.), and historical records of pipe cleaning operations (such as pipe cleaning frequency, cost, effect, etc.); the output of the decision engine is pipe cleaning operation recommendations (such as whether to perform pipe cleaning, pipe cleaning method, pipe cleaning time, etc.). The objective function is to minimize pipe cleaning costs and pipeline operation risks, which requires combining factors such as pipe cleaning costs, sediment risks, and operational complexity; the constraints are the safety, effectiveness, and feasibility of pipe cleaning operations, such as the pipe cleaning frequency cannot be too high to avoid affecting pipeline operation, and the pipe cleaning effect must meet the requirements. Each particle represents a pipe cleaning strategy, which includes parameters such as pipe cleaning frequency, pipe cleaning method, and pipe cleaning time. The management system calculates the fitness of each particle based on the objective function, evaluates the advantages and disadvantages of the pipe cleaning strategy, and uses the PSO-ADE update rule to dynamically adjust the particle speed and position to find the optimal pipe cleaning strategy, terminating when the maximum number of iterations is reached or the fitness meets the requirements.

[0086] Specifically, the management system collects transportation information from the pipeline operation system and sediment monitoring system, including flow, pressure, temperature, sediment thickness, and composition data. It removes outliers and fills in missing values ​​to ensure data accuracy and completeness. It also normalizes the data to a certain range to improve the algorithm's convergence speed and stability. The particle swarm is initialized, and parameters such as the particle number and dimension are determined. The particle positions and velocities are initialized, and then inertia weights and learning factors are set. Based on the ADE mechanism, these inertia weights and learning factors are dynamically adjusted to balance global and local search capabilities. Based on actual conditions, the objective function and constraints are defined to ensure the algorithm's feasibility and effectiveness.

[0087] Use programming languages ​​(such as Python and MATLAB) to implement the PSO-ADE algorithm, including functions such as particle updating, fitness calculation, and optimal solution selection. Test the algorithm on simulated or historical data, and adjust parameters and optimize the algorithm based on the results to improve performance and effectiveness. Evaluate the performance of the decision engine based on actual pigging results, such as whether pigging costs have been reduced and pipeline operational risks have been mitigated. Feedback the evaluation results into the algorithm to further optimize and improve the decision engine, forming a closed-loop optimization.

[0088] S103: Initialize the algorithm and generate an initial particle swarm.

[0089] Specifically, the management system sets the maximum number of iterations or termination conditions for the algorithm to ensure convergence within a reasonable time. It also randomly assigns an initial position and velocity to each particle. Positions can be generated uniformly between the upper and lower bounds of the solution space, and velocities can be set to zero or a small random number. For example, the number of particles N in the swarm is determined to be 50, and the maximum number of iterations T of the algorithm is set to be 100.

[0090] S104: Run the algorithm to evaluate and update the particles according to the objective function.

[0091] Specifically, the management system defines the objective function, which reflects the optimization goals of the pigging strategy, such as minimizing pigging costs and maximizing pipeline flow capacity. Appropriate weights and constraints are then set for the objective function based on actual needs and pipeline operating conditions. A particle evaluation mechanism is designed to assess the fitness of particles based on the objective function. Update strategies, such as inertia weights and learning factors, are used to update the particle positions and velocities, guiding them toward more optimal solutions.

[0092] Based on actual needs, an objective function f(x) is defined, where x represents the particle's position (i.e., a parameter of the pigging strategy). For example, the objective function can be a weighted sum of pigging cost and pipeline flow capacity, with the weights set based on actual needs. In each iteration, the management system calculates the fitness value of each particle, i.e., the value of the objective function f(x), and sorts the particles according to their fitness values, identifying the current optimal particle (i.e., the particle with the smallest fitness value) and the global optimal particle (i.e., the particle with the smallest fitness value in the previous iteration).

[0093] The management system updates the speed and position of each particle based on parameters such as inertia weight and learning factor. The inertia weight can control the exploration and development capabilities of the particle, and the learning factor can affect the degree to which the particle learns towards the individual optimum and the global optimum until the maximum number of iterations T is reached or other termination conditions (such as fitness value convergence) are met. After each iteration, the position and fitness value of the current optimal particle and the global optimal particle are recorded.

[0094] S105. Find the optimal pipe cleaning strategy through iterative optimization.

[0095] Specifically, the management system outputs the position of the global optimal particle, that is, the optimal pipe cleaning strategy parameters, and outputs the fitness value of the global optimal particle, that is, the objective function value corresponding to the optimal pipe cleaning strategy. According to actual needs and changes in pipeline operating conditions, the algorithm parameters (such as the number of particle swarms, maximum number of iterations, inertia weight, learning factor, etc.) are adjusted and optimized, thereby effectively initializing the algorithm, generating the initial particle swarm, and running the algorithm to find the optimal pipe cleaning strategy.

[0096] S106. Collect feedback data and further adjust and optimize the algorithm.

[0097] Specifically, after the pigging strategy is implemented, the management system collects pipeline operation data, wax deposition monitoring data, and pigging operation data in real time or periodically. The system analyzes the performance of the current algorithm and any existing issues, such as slow convergence and insufficient search capabilities. Based on the analysis results, the algorithm parameters are adjusted, such as increasing the number of particle swarms, increasing the number of iterations, adjusting the learning factor and inertia weight, and revising and improving the objective function to more accurately reflect the optimization goals of the pigging strategy. The algorithm is then rerun for a new round of iterative optimization. This effectively finds the optimal pigging strategy through iterative optimization, and feedback data is collected to further adjust and optimize the algorithm to improve the accuracy and effectiveness of decision-making. At the same time, through continuous monitoring and optimization, we can ensure that the algorithm always meets the needs of actual gas system management.

[0098] In another embodiment, S101 specifically includes the following sub-steps:

[0099] A four-field coupled model of heat, fluid, solidification and chemical reaction is constructed based on COMSOL Multiphysics. The equations to be solved include:

[0100]

[0101] Where ρ is the fluid density, which is used to represent the mass of the fluid per unit volume; u is the velocity vector, which is used to represent the movement speed of the fluid particles; t is the time; p is the pressure, which is used to represent the force acting vertically on the unit area inside the fluid; μ is the dynamic viscosity coefficient, which is used to represent the internal friction of the fluid and reflects the viscosity of the fluid; S wax The wax phase change source term is used, and an improved molecular diffusion-gel deposition dual mechanism model is used to represent the effect of wax phase change in the fluid on the flow;

[0102] Introducing dynamic grid adaptation technology to automatically refine the grid to 0.5mm resolution in areas with sudden deposition rate changes;

[0103] Train the XGBoost-GRU hybrid model, with input features including temperature gradient (ΔT), wall shear stress (τ_w), wax crystal volume fraction (φ), and output deposition rate correction factor α;

[0104] A generative adversarial network (GAN) is used to expand limited experimental data and generate a synthetic dataset covering the temperature range of -10 to 50°C to address the problem of insufficient extrapolation capability of traditional models.

[0105] Specifically, the management system uses COMSOL Multiphysics software to create a model, defining the geometry and dimensions based on the actual problem, ensuring that the geometry accurately describes the simulation area. Physical parameters related to heat conduction, fluid flow, solid mechanics, and chemical phase change are selected and set as follows:

[0106] Heat conduction module: set parameters such as thermal conductivity and heat capacity.

[0107] Fluid flow module: Set parameters such as fluid density ρ, velocity vector u, dynamic viscosity coefficient μ, etc.

[0108] Solid Mechanics Module: Set material properties such as elastic modulus, Poisson's ratio, etc.

[0109] Chemical Phase Change Module: Set the wax phase change source term S wax , an improved molecular diffusion-gel deposition dual mechanism model was adopted.

[0110] Define the physical fields related to wax phase change, such as phase field variables, chemical potential, etc.

[0111] The management system meshes the geometric model to ensure mesh quality meets simulation requirements. Dynamic mesh adaptation technology is introduced to automatically refine the mesh to 0.5mm resolution in areas with sudden changes in deposition rate to capture detailed changes during wax deposition. Boundary and initial conditions, including those for temperature, pressure, velocity, and initial conditions for wax phase transitions, are applied to ensure they accurately reflect the physical context of the problem.

[0112] Collect experimental data, including temperature gradient (ΔT), wall shear stress (τ_w), wax crystal volume fraction The data is preprocessed based on features such as normalization and missing value processing, and an XGBoost-GRU hybrid model is constructed. The feature data is input for training, and the deposition rate correction factor α is output to correct the deposition rate in the simulation results. In addition, a generative adversarial network (GAN) is used to expand the experimental data to generate a synthetic data set covering a temperature range of -10 to 50°C. The synthetic data set is combined with the experimental data to improve the generalization and extrapolation capabilities of the model. The simulation results are analyzed, including changes in the temperature field, flow field, solid mechanics field, and chemical phase change field, to verify the accuracy of the simulation results, such as by comparing with the experimental results and performing grid independence verification. The model is optimized and adjusted based on the simulation results and the output of the machine learning model.

[0113] By introducing dynamic mesh adaptation technology, the mesh is automatically refined in areas where the deposition rate changes suddenly, improving the accuracy and resolution of the simulation results. The accuracy of the simulation results is further improved by training a machine learning model and applying a deposition rate correction factor. A generative adversarial network (GAN) is used to expand the experimental data to generate a synthetic dataset covering a wider temperature range, addressing the insufficient extrapolation capability of traditional models and enabling the model to better adapt to the simulation of wax deposition processes under different temperatures and conditions. Based on the simulation results and the output of the machine learning model, the design of the coupled heat-fluid-solidification-chemical system is optimized, providing decision support and solutions to wax deposition problems in industrial processes, such as optimizing operating parameters and improving equipment structure.

[0114] In another embodiment, S103 specifically includes the following sub-steps:

[0115] Establish the objective function:

[0116]

[0117] Among them, C pigging is the cost of the pigging operation, which is used to represent the cost of each pigging operation; f i is the frequency of the i-th pigging operation, which is used to indicate the number of pigging operations within a certain period of time; λ is the risk penalty coefficient, which is used to indicate the degree of penalty for risk and reflects the cost of risk; δ i is the predicted sediment thickness, which is used to represent the thickness of the sediment in the pipeline predicted during the i-th pigging operation; δ safe Safe sediment thickness is used to indicate the safe thickness of sediment in the pipeline. Exceeding this thickness may cause risks.

[0118] The objective function is to find a balance between the cost of the pigging operation and the risk posed by sediment, so as to minimize the total cost;

[0119] The improved PSO-ADE algorithm is used to introduce an inertia weight adaptive mechanism to enhance the local search capability in the later stage of iteration. The formula is as follows:

[0120] w(t)=w min +(w max -w min )·e -20t / T

[0121] The formula is used to describe the change of w(t) with time t. The specific parameters have the following meanings: w min represents the minimum or stable value of w(t). When time t approaches infinity, w(t) will approach this value. max Indicates the initial maximum value of w(t). When time t = 0, the value of w(t) is wmax ; t represents time; T represents the time constant, the unit of which is consistent with the time t, reflecting the speed at which w(t) decays to a stable value. The larger the time constant T, the slower the decay process.

[0122] Specifically, the management system determines the form of the objective function, which should include the cost of the pipe cleaning operation and the risk cost. The cost of the pipe cleaning operation is represented by the product of the cost of each pipe cleaning and the frequency of pipe cleaning. The risk cost is represented by the functional relationship between the risk penalty coefficient, the difference between the predicted sediment thickness and the safe sediment thickness (if the difference is positive, it means that the sediment thickness exceeds the safe value). The goal is to minimize the total cost (pipe cleaning operation cost + risk cost). The management system selects the improved PSO-ADE algorithm for optimization, sets the algorithm parameters, and adjusts the size of the particle swarm, the number of iterations, etc. according to the scale and complexity of the problem. The objective function is converted into a fitness function to evaluate the quality of the particles. The smaller the fitness function value, the better the solution corresponding to the particle.

[0123] The management system initializes the particle swarm, including its positions and velocities, evaluates the quality of the particles based on a fitness function, updates their positions and velocities, introduces an adaptive inertia weight mechanism to enhance local search capabilities, and iterates these steps until the preset number of iterations is reached or convergence conditions are met. The management system analyzes the optimization results, including optimal and suboptimal solutions, to verify their rationality and effectiveness, ensuring they meet actual needs.

[0124] By optimizing pigging frequency and sediment management strategies, total costs are reduced, balancing pigging operational costs and risk costs to maximize economic benefits. Sediment thickness is predicted, enabling timely pigging measures to prevent sediment from exceeding safe limits, mitigating risks such as pipeline blockages and leaks caused by excessive sediment. This provides a scientific basis for decision-making, helping managers develop rational pigging plans and risk management strategies, improving decision-making efficiency and accuracy while mitigating risk. The introduction of an inertia weight adaptive mechanism enhances the algorithm's local search capabilities, improving convergence speed and solution quality, making it suitable for solving complex optimization problems.

[0125] Furthermore, it also includes:

[0126] Select typical gas system accidents as analysis objects to clarify the severity of the accident consequences;

[0127] A typical accident of the gas system is regarded as a top event, and the top event is decomposed into multiple sub-top events, wherein the sub-top events are the possible causes of the top event;

[0128] Based on the sub-top event and the top event, a fault tree logic model is constructed; the causes are decomposed layer by layer using "AND gate" (all conditions are met at the same time) and "OR gate" (any condition is met) until the basic events corresponding to the sub-top event are obtained;

[0129] Calculating the minimum cut set and structural importance coefficient, wherein the minimum cut set is the minimum combination of basic events that lead to the top event, and is used to identify key risk factors;

[0130] Structural importance coefficient

[0131] Among them, I i Indicates the value of the i-th indicator, which is a comprehensive value obtained by summation; ∑ represents the summation operation, which accumulates all items that meet the conditions; n j The number of events representing the jth minimum cut set; 2 nj-1 : Indicates that the base is 2 and the exponent is n j -1 is raised to the power of the denominator of each term.

[0132] Combine historical data to calculate the probability of occurrence of basic events and derive the probability of top events.

[0133] Specifically, the management system selects typical gas system accidents as analysis targets, such as liquefied petroleum gas (LPG) or natural gas (NG) leaks and explosions, and identifies the severity of the accident consequences, including casualties, property damage, and environmental impact. Typical gas system accidents are considered top events, such as "gas system explosions," and are broken down into multiple sub-top events, representing the possible causes of the top event, such as "equipment failure," "human error," and "environmental factors."

[0134] Based on the secondary top events and top events, an accident tree logic model is constructed, and "AND gate" (all conditions are met at the same time) and "OR gate" (any condition is met) are used to connect events to express the logical relationship between events. Then, logical reasoning is continued to be used to decompose the secondary top events into more specific causes until the basic events are obtained. Among them, the basic events are the most direct causes of the secondary top events, usually specific factors such as equipment failure, human error, and environmental factors.

[0135] The management system uses Boolean algebra to simplify the accident tree and find the minimum cut set. The minimum cut set is the minimum combination of basic events that leads to the occurrence of the top event. It is used to identify key risk factors and calculate the structural importance coefficient to assess the impact of each basic event on the occurrence of the top event. The structural importance coefficient can be obtained by summing the number of events in the minimum cut set and the corresponding power operation, and collecting relevant historical data, including the frequency of occurrence of basic events and accident records. Based on the historical data, the probability of occurrence of basic events is calculated, and then the probability of occurrence of the top event is derived. This can be achieved through methods such as statistical frequency, trend analysis, and time series analysis.

[0136] By constructing an accident tree logic model, the key factors and paths leading to typical gas system accidents can be clearly identified. The identification of minimal cut sets helps determine key risk factors, allowing targeted preventive measures to reduce the likelihood of accidents. Calculating the structural importance coefficient provides decision makers with information on which basic events have the greatest impact on the occurrence of top events, helping them prioritize factors to achieve the greatest risk reduction with limited resources. Understanding the probability of occurrence of basic events and top events helps develop more effective emergency plans and response measures, ensuring a swift and effective response when an accident occurs, thereby mitigating the losses caused by the accident.

[0137] Furthermore, after the step of deriving the probability of the top event, as another implementation manner, the embodiment of the present application may further include the following steps:

[0138] S201. Obtain a ranking of key risk factors from a preset database.

[0139] Specifically, the management system retrieves key risk factors and their rankings from a preset database. The key risk factor rankings are pre-set and generated by the user based on historical accident analysis, expert evaluation, industry standards and other factors to ensure that the risk factor rankings are up-to-date and match the specific conditions of the current gas system.

[0140] S202: Generate a safety checklist based on the top event probabilities and the ranking of key risk factors.

[0141] Specifically, the management system combines the derived probability of top events with the ranking of key risk factors obtained from the database, and associates the probability value with the ranking of risk factors, so that when generating a safety checklist, it can clearly reflect which factors have a higher risk level. Based on the integrated information, a safety checklist is designed, which contains key risk factors, corresponding risk levels (based on probability and ranking), recommended inspection frequency, inspection methods, preventive measures and other information. Ensure that the checklist is easy to understand and convenient for users to execute in actual operations. By combining the ranking of key risk factors with the probability of top events, users can more clearly understand which factors require priority attention and management. The generation and push of safety checklists makes the risk management process more systematic and standardized, and improves management efficiency.

[0142] S203: Push the safety checklist to the user's smart terminal.

[0143] Specifically, the management system will push the generated safety checklist to the user's smart terminal through relevant channels (such as email, text messages, mobile applications, etc.), ensure that the push process is safe and reliable, and consider user privacy protection, provide necessary instructions and guidance, and help users understand and use the safety checklist.

[0144] This example uses the TOPSIS method to calculate the structural risk closeness of basic events, which specifically includes the following steps:

[0145] S301, establish a matrix and perform forward transformation, standardization and normalization processing.

[0146] Specifically, the management system establishes a matrix based on the structural importance and risk probability of basic events and performs forward processing, standardization, and normalization. Data on the structural importance and risk probability of basic events is collected and cleaned and preprocessed to ensure data accuracy and consistency. An initial matrix is ​​established based on the structural importance and risk probability of basic events. This matrix is ​​forward processed, and all indicator types are uniformly converted to extremely large indicators (if extremely small, intermediate, or interval indicators exist, corresponding conversions are required). The forward-processed matrix is ​​then standardized to eliminate the influence of the dimensions of each indicator. The standardized matrix is ​​then normalized to obtain a normalized matrix.

[0147] S302: Calculate the weights of evaluation indicators using the entropy weight method to obtain a weighted normalized indicator matrix.

[0148] Specifically, the management system calculates the information entropy of each indicator based on the normalized matrix, calculates the difference coefficient based on the information entropy to reflect the degree of dispersion of each indicator data, obtains the weight of each indicator based on the normalization of the difference coefficient, and multiplies the normalized matrix with the weight vector to obtain a weighted normalized indicator matrix.

[0149] S303. Calculate the structural risk proximity of basic events.

[0150] Specifically, the management system calculates the structural risk proximity of basic events, sorts them by proximity, and determines the criticality of each basic event to the top event. In the weighted normalized indicator matrix, the optimal solution (maximum value of each indicator) and the worst solution (minimum value of each indicator) are determined. The Euclidean distance between each basic event and the optimal and worst solutions is calculated. Based on the Euclidean distance, the relative proximity of each basic event (i.e., structural risk proximity) is calculated. Basic events are sorted by structural risk proximity to determine their criticality to the top event. Basic events with greater proximity have a greater impact on the top event.

[0151] In another embodiment, S301 specifically includes the following sub-steps:

[0152] Furthermore, the steps of establishing a matrix and performing forward, standardization, and normalization processing based on the structural importance and risk probability of the basic events further include:

[0153] First, establish an initial matrix A=(b ij ) m×n Since the structural importance I and risk probability index have been positive, the Min-Max normalization method is used to normalize the initial matrix A=(b ij ) m×n To standardize the indicators, the formula is as follows:

[0154] y=(bA min ) / (A max -A min )

[0155] Among them, y is the normalized value; b is the original value of a certain attribute; A min is the minimum value of a certain attribute; A max The maximum value of a certain attribute;

[0156] After normalization, we get the indicator matrix B=(x ij ) m×n , the normalized calculation formula is as follows:

[0157]

[0158] Among them, x ij are normalized values.

[0159] Specifically, the management system constructs an m×n initial matrix A=(b ij ) m×n Among them, bij Denote the raw value of the i-th item on the j-th indicator. Calculate the minimum and maximum values ​​for each indicator. Use the Min-Max normalization formula to normalize each indicator. Fill the normalized values ​​into a new matrix to obtain a normalized matrix. Based on the normalized matrix, perform normalization. The goal of normalization is to convert each element in the matrix to a value in the range [0, 1] to eliminate dimensional differences between different indicators.

[0160] It should be noted that the specific formula for normalization needs to be adjusted according to the actual situation. For example, if you want to convert each element in the matrix into a unit vector (i.e., the vector has a modulus of 1), you can use the L2 normalization method. This embodiment uses a method similar to Min-Max normalization, that is, maintaining the proportional relationship between elements and mapping them to the range of [0, 1].

[0161] After Min-Max normalization, each element in the matrix is ​​converted to a ratio between the minimum and maximum values ​​of its corresponding indicator. This helps eliminate dimensional differences between different indicators, making subsequent analysis and comparison more accurate and fair. The standardized matrix is ​​more comparable and interpretable, providing strong support for subsequent data analysis and decision-making.

[0162] Normalization further converts each element in the matrix into a value in the range of [0,1], which helps to simplify subsequent calculations and analysis. The normalized matrix is ​​more intuitive and easy to understand, and can clearly show the performance of each project on different indicators. Normalization also helps to improve the stability and accuracy of the model.

[0163] Furthermore, after normalization, we get the indicator matrix B = (x ij ) m×n After the step, as another implementation manner, the embodiment of the present application may further include the following steps:

[0164] Based on the obtained indicator matrix B, the weights Wj of each indicator are calculated to obtain the weighted normalized indicator matrix C. The expression of the weighted normalized indicator matrix C is as follows:

[0165]

[0166] The entropy weight method is used to calculate the indicator weights to ensure the scientificity and rationality of the indicator weights. The specific calculation formula is as follows:

[0167]

[0168] Among them, W j is the entropy weight of the calculation index; Hj is the entropy value of the calculation indicator; k is the Boltzmann constant; x ij is the i-th standardized and normalized value under the j-th indicator;

[0169] When calculating the closeness, take the maximum value of a certain indicator as the positive ideal point v j + , the minimum value is the negative ideal point v j - ; Calculate the relative distance S between a certain item and all positive and negative ideal points i + and S i - , and then calculate the closeness S i The calculation formula is as follows:

[0170]

[0171] In the formula

[0172]

[0173] Among them, v ij is the weighted and standardized value of the jth parameter of the i-th item; S i The closer the degree of closeness is, the closer it is to the ideal state; S i Assess the criticality of basic events to top events as a benchmark.

[0174] Specifically, after normalization, the indicator matrix is ​​obtained, and the entropy value of each indicator is calculated using the entropy weight method, where p ij =x ij / ∑x ij , which represents the probability of the occurrence of the i-th item (or basic event) under the j-th indicator (that is, the proportion of the normalized value in all normalized values); k is a proportional coefficient of the Boltzmann constant, which is used to adjust the range of the entropy value, usually k = 1 / (lnm), m is the number of items (or basic events), and the k value here can be adjusted according to actual conditions.

[0175] According to the entropy value H j Calculate the entropy weight of each indicator, entropy weight W j It reflects the importance of each indicator in the evaluation. The smaller the entropy value, the more certain the information is, and the greater the weight of the indicator in the evaluation should be. According to the entropy weight W of each indicator j And indicator matrix B, construct a weighted normalized indicator matrix, which reflects the comprehensive performance of each project (or basic event) after considering the indicator weights.

[0176] Identify positive and negative ideal points. The positive ideal point is the maximum value of each metric, while the negative ideal point is the minimum value of each metric. Calculate the relative distance between a project (or basic event) and all positive and negative ideal points. Relative distance can be calculated using distance metrics such as Euclidean distance and Manhattan distance. Based on the relative distance, calculate the closeness (Si). The larger the Si, the closer the project (or basic event) is to the ideal state.

[0177] Calculating indicator weights using the entropy weight method avoids the arbitrariness of subjective weighting and improves the scientific nature and rationality of the weights. The entropy weight method can objectively reflect the information uncertainty of each indicator, thereby determining the weights and making the evaluation results more objective and accurate. The construction of the weighted normalized indicator matrix and the calculation of proximity are both based on objective data, improving the accuracy and objectivity of the evaluation. The calculation of proximity takes into account the relative distance of the project (or basic event) from the positive and negative ideal points, and can more accurately reflect the actual situation of the project and its proximity to the ideal state. The steps of this technical solution are clear and easy to understand, making it easy to operate and apply in actual evaluations. By calculating proximity, the criticality of basic events to top events can be intuitively assessed, providing strong support for decision-making. At the same time, this technical solution can be applied to the evaluation of different fields and different projects, showing strong versatility and practicality.

[0178] Based on the above method, the embodiment of the present application further discloses a gas system management system based on data analysis. A gas system management system based on data analysis includes:

[0179] A model architecture design module is used to design a hybrid prediction model architecture by combining a wax deposition kinetics model that describes the physical and chemical processes of wax deposition and provides basic prediction capabilities with a deep neural network that captures nonlinear relationships and complex patterns in the data, improving prediction accuracy and generalization capabilities.

[0180] Decision engine design module, used to design an adaptive pigging decision engine based on an improved particle swarm optimization algorithm (PSO-ADE);

[0181] Initialization module, used to initialize the algorithm and generate the initial particle swarm;

[0182] The running module is used to run the algorithm and evaluate and update the particles according to the objective function;

[0183] Iterative optimization module, used to find the optimal pigging strategy through iterative optimization;

[0184] The feedback data collection module is used to collect feedback data and further adjust and optimize the algorithm to improve the accuracy and effectiveness of decision-making.

[0185] An embodiment of the present application further discloses an intelligent terminal, which includes a memory and a processor, wherein the memory stores a computer program that can be loaded and executed by the processor for a gas system management method based on data analysis as described above.

[0186] The present application also discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program capable of being loaded by a processor and executed by a gas system management method based on data analysis, such as the above-described method. The computer-readable storage medium includes, for example, a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, among other media capable of storing program code.

[0187] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the scope of protection of the invention. Obviously, the embodiments described are only some embodiments of the present invention, rather than all embodiments. Based on these embodiments, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in this field can still combine, add, delete or make other adjustments to the features in the various embodiments of the present invention according to the circumstances without conflict, without making creative work, so as to obtain different other technical solutions that do not deviate from the concept of the present invention in essence, and these technical solutions also fall within the scope of protection of the present invention.

Claims

1. A gas system management method based on data analysis, characterized in that: The following steps are involved: A hybrid prediction model architecture was designed by combining a wax deposition kinetics model with a deep neural network. The wax deposition kinetics model describes the physical and chemical processes of wax deposition, providing basic prediction capabilities. The deep neural network captures nonlinear relationships and complex patterns in the data, improving prediction accuracy and generalization. The step of combining the wax deposition dynamics model with the deep neural network to design a hybrid prediction model architecture also includes: A four-field coupled model of heat, fluid, solidification and chemical reaction is constructed based on COMSOL Multiphysics. The equations to be solved include: ； Where ρ is the fluid density, which is used to represent the mass of the fluid per unit volume; u is the velocity vector, which is used to represent the movement speed of the fluid particles; t is the time; p is the pressure, which is used to represent the force acting vertically on the unit area inside the fluid; μ is the dynamic viscosity coefficient, which is used to represent the internal friction of the fluid and reflects the viscosity of the fluid; S wax The wax phase change source term is used, and an improved molecular diffusion-gel deposition dual mechanism model is used to represent the effect of wax phase change in the fluid on the flow; Introducing dynamic grid adaptation technology to automatically refine the grid to 0.5mm resolution in areas with sudden deposition rate changes; Train the XGBoost-GRU hybrid model, with input features including temperature gradient ΔT, wall shear stress τ_w, and wax crystal volume fraction φ, and output a deposition rate correction factor α; The Generative Adversarial Network (GAN) was used to expand limited experimental data to generate a synthetic dataset covering the temperature range of -10 to 50°C, which was used to address the problem of insufficient extrapolation ability of traditional models. Design an adaptive pigging decision engine based on the improved particle swarm optimization algorithm PSO-ADE; Initialize the algorithm and generate the initial particle swarm; The step of initializing the algorithm further includes: Establish the objective function: ； Among them, C pigging is the cost of the pigging operation, which is used to represent the cost of each pigging operation; f i is the frequency of the i-th pigging operation, which is used to indicate the number of pigging operations within a certain period of time; λ is the risk penalty coefficient, which is used to indicate the degree of penalty for risk and reflects the cost of risk; δ i is the predicted sediment thickness, which is used to represent the thickness of the sediment in the pipeline predicted during the i-th pigging operation; δ safe Safe sediment thickness is used to indicate the safe thickness of sediment in the pipeline. Exceeding this thickness may cause risks. The objective function is to find a balance between the cost of the pigging operation and the risk posed by sediment, so as to minimize the total cost; The improved PSO-ADE algorithm is used to introduce an inertia weight adaptive mechanism to enhance the local search capability in the later stage of iteration. The formula is as follows: ； The formula is used to describe the change of w(t) with time t. The specific parameters have the following meanings: w min represents the minimum or stable value of w(t). When time t approaches infinity, w(t) will approach this value. max Indicates the initial maximum value of w(t). When time t=0, the value of w(t) is w max ; t represents time; T represents the time constant, the unit of which is consistent with the time t, reflecting the speed at which w(t) decays to a stable value. The larger the time constant T, the slower the decay process. Run the algorithm to evaluate and update particles according to the objective function; Through iterative optimization, the optimal pigging strategy is found.

2. A gas system management method based on data analysis according to claim 1, characterized in that: Also includes: Select typical gas system accidents as analysis objects to clarify the severity of the accident consequences; A typical accident of the gas system is regarded as a top event, and the top event is decomposed into multiple sub-top events, wherein the sub-top events are the possible causes of the top event; Based on the sub-top event and the top event, a fault tree logic model is constructed; the cause is decomposed layer by layer using "AND gate" and "OR gate", where "AND gate" means all conditions are met at the same time, and "OR gate" means any condition is met, until the basic event corresponding to the sub-top event is obtained; Calculating the minimum cut set and structural importance coefficient, wherein the minimum cut set is the minimum combination of basic events that lead to the top event, and is used to identify key risk factors; Structural importance coefficient , Among them, I i Indicates the value of the i-th indicator, which is a comprehensive value obtained by summation; ∑ represents the summation operation, which accumulates all items that meet the conditions; n j represents the number of events in the jth minimum cut set; Indicates base 2, exponent n j -1 is raised to the power of the denominator of each term. Combine historical data to calculate the probability of occurrence of basic events and derive the probability of top events.

3. A gas system management method based on data analysis according to claim 2, characterized in that: After the step of deriving the probability of the top event, the method further includes: Obtain the ranking of key risk factors from a pre-set database; Generate a safety checklist based on the probability of the top events and the ranking of key risk factors; The safety checklist is pushed to the user's smart terminal.

4. A gas system management method based on data analysis according to claim 2, characterized in that: The TOPSIS method is used to calculate the structural risk closeness of basic events, including: According to the structural importance and risk probability of basic events, a matrix is ​​established and processed with forward, standardization and normalization; The entropy weight method is used to calculate the weights of the evaluation indicators and obtain the weighted normalized indicator matrix; Calculate the structural risk proximity of basic events, sort them according to the degree of proximity, and determine the criticality of each basic event to the top event.

5. A gas system management method based on data analysis according to claim 4, characterized in that: The steps of establishing a matrix and performing forward, standardization, and normalization processing based on the structural importance and risk probability of the basic events also include: First, establish an initial matrix A=(b ij ) m×n Since the structural importance I and risk probability index have been positive, the Min-Max normalization method is used to normalize the initial matrix A = (b ij ) m×n To standardize the indicators, the formula is as follows: ； Among them, y is the normalized value; b is the original value of a certain attribute; A min is the minimum value of a certain attribute; A max The maximum value of a certain attribute; After normalization, we get the indicator matrix B=(x ij ) m×n , the normalized calculation formula is as follows: ； Among them, x ij are normalized values.

6. A gas system management method based on data analysis according to claim 5, characterized in that: After normalization, the indicator matrix B=(x ij ) m×n After the steps, it also includes: Calculate the weight W of each indicator based on the obtained indicator matrix B n , thus obtaining the weighted normalized indicator matrix C. The expression of the weighted normalized indicator matrix C is as follows: ； The entropy weight method is used to calculate the indicator weights to ensure the scientificity and rationality of the indicator weights. The specific calculation formula is as follows: ； Among them, W j is the entropy weight of the calculation index; H j is the entropy value of the calculation indicator; k is the Boltzmann constant; x ij is the i-th standardized and normalized value under the j-th indicator; When calculating the closeness, take the maximum value of a certain indicator as the positive ideal point v j + , the minimum value is the negative ideal point v j - ; Calculate the relative distance S between a certain item and all positive and negative ideal points i + and S i - , and then calculate the closeness S i , the calculation formula is as follows: ； In the formula ； Among them, v ij is the weighted and standardized value of the jth parameter of the i-th item; S i The closer the degree of closeness is, the closer it is to the ideal state; S i Assess the criticality of basic events to top events as a benchmark.

7. The gas system management system corresponding to the gas system management method based on data analysis according to claim 1 is characterized in that: include: A model architecture design module is used to combine the wax deposition dynamics model with a deep neural network to design a hybrid prediction model architecture; The wax deposition kinetics model is used to describe the physical and chemical processes of wax deposition and provide basic predictive capabilities. The deep neural network is used to capture nonlinear relationships and complex patterns in the data and improve the accuracy and generalization of predictions. Decision engine design module, used to design an adaptive pigging decision engine based on the improved particle swarm optimization algorithm PSO-ADE; Initialization module, used to initialize the algorithm and generate the initial particle swarm; The running module is used to run the algorithm and evaluate and update the particles according to the objective function; Iterative optimization module, used to find the optimal pigging strategy through iterative optimization; The feedback data collection module is used to collect feedback data and further adjust and optimize the algorithm to improve the accuracy and effectiveness of decision-making.

Images