Artificial intelligence optimization method for natural gas liquefaction system
By constructing a mechanism map and mechanism model of the cooling system, converting performance indicators into generalized energy, and adopting a hybrid optimization strategy, the problems of long optimization modeling time and high solution complexity of the natural gas liquefaction system were solved, efficient and accurate optimization modeling and equipment scheduling recommendations were achieved, and the system's adaptability was improved.
Patent Information
- Application Number
- CN202510807709.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-17
- Publication Date
- 2025-09-23
AI Technical Summary
The optimization modeling process of the existing natural gas liquefaction system relies on digital twin technology and lacks further abstraction, resulting in long modeling time, lack of industry consensus on weight allocation, high solution complexity, and computational links that rely on explicit solution methods that cannot be processed in parallel, resulting in extended operation cycles.
By constructing a mechanism map and mechanism model of the cooling system, using digital twin technology to analyze the equipment unit structure and control logic, converting multi-dimensional performance indicators into generalized energy, and adopting a hybrid optimization strategy (mixed integer programming for the linear part + mechanism simulation for the nonlinear part), an optimal scheduling model is generated and the optimization suggestions are reversely mapped, and the model is dynamically updated to adapt to changes in working conditions.
It achieves efficient and accurate optimization modeling, simplifies multi-objective optimization problems, improves solution efficiency and accuracy, generates actionable equipment scheduling suggestions, ensures that optimization strategies are directly implemented, and has self-evolution capabilities.
Smart Images
Figure CN120688747A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of cooling system optimization, and in particular relates to an artificial intelligence optimization method for a natural gas liquefaction system. Background Art
[0002] Natural gas liquefaction systems are large-scale cooling systems in chemical plants. The core thermodynamic cycle of this type of cooling system follows four basic processes: compression, condensation, expansion, and evaporation. The compression process involves increasing the pressure and temperature of the refrigerant through a compressor. The condensation process transfers the refrigerant's heat to the surrounding environment through a heat exchange medium such as air or water. The expansion or throttling process uses an expansion valve or throttling device to reduce the refrigerant's pressure and temperature, creating conditions for subsequent evaporation and heat absorption. The evaporation process involves the refrigerant absorbing heat and vaporizing in the evaporator, achieving the desired cooling effect.
[0003] The production process of a cooling system involves continuous physical changes and the coupling of multiple variables (such as temperature, pressure, and material flow). Engineers, with system energy consumption or cost optimization as the goal, extracted and analyzed the specific process mechanisms from the cooling system design plan, operating manual, and distributed control system (DCS). They then developed a multi-physics model of the cooling system. By combining the optimization objective with the constraint model, they rationally simplified and approximated the entire cooling system. Finally, through discrete sampling simulation, they arrived at the optimized scheduling decision. The entire decision model was compared with production data or experiments, and iteratively refined until it aligned with factory realities.
[0004] The current optimization modeling process relies on a digital twin cooling system, lacking further abstracted objects. This limits the efficiency of model iterations throughout the modeling-verification-adjustment cycle and prolongs the time to market. Regarding the configuration of optimization objectives, existing technologies focus on the coordinated optimization of multiple objectives (such as energy consumption, cost, and efficiency) and multiple weights. However, there is a lack of industry consensus or a standardized framework for weight allocation. For example, while Pareto optimality theory can generate a non-dominated solution set, different stakeholders (such as engineers, managers, and environmental protection departments) have widely varying preferences for weights, leading to exponential growth in solution complexity (as the number of objectives increases) and increased communication costs.
[0005] Traditional simulation optimization uses explicit solution methods (such as explicitly solving temperature and pressure changes layer by layer). These methods have strong dependencies on computational links and cannot be processed in parallel. For example, without implicit solution or modular design, distributed computing or GPU acceleration technologies cannot be utilized, resulting in extended computation cycles. Summary of the Invention
[0006] In order to solve the above problems existing in the prior art, the present invention provides an artificial intelligence optimization method for a natural gas liquefaction system to solve the problem of long modeling time.
[0007] The purpose of the present invention can be achieved by the following technical solution: comprising the following steps: S1: Based on the system design drawings and operation manuals, digital twin technology is used to analyze the structural information of the equipment units and the control logic information between the equipment units. Based on the structural information and control logic information of the equipment units, a mechanism map of the cooling system is constructed; S2: extracting the topological connection relationship of each device unit of the system and the mechanism equations between the device units based on the mechanism map, and constructing a mechanism model; S3: convert the multi-dimensional performance indicators of the cooling system into generalized energy through weighted coefficients; S4: The generalized energy and mechanism models are integrated to generate an optimal scheduling model and the model is optimized. The input data of the optimization model is divided into linear and nonlinear using a preset judgment model. The nonlinear part is optimized using the mechanism model simulation, and the linear part is optimized using the mixed integer programming method to generate the corresponding optimization results. S5: Based on the optimization results, the optimization results are reversely mapped to the corresponding nodes of the mechanism map to generate scheduling optimization suggestions, which include the start and stop of valves and the adjustment of control lines; S6: Output scheduling optimization suggestions and generate reports, and dynamically update the mechanism map and optimal scheduling model based on the actual operation data of the liquefaction system.
[0008] Preferably, step S1 includes the following sub-steps: S11: Multi-dimensional data analysis and feature extraction: Extracting physical parameters of equipment units from design drawings. The physical parameters include analyzing the equipment operation process and operation thresholds in the operation manual, extracting the PID operation logic relationship between multiple devices, and extracting the transmission process of control signals. S12: Convert the collected physical model into a virtual model: Using digital twin technology, convert the collected physical information into a virtual model; S13: Construction of mechanism map: Based on the analyzed equipment unit structure information and control logic information, a mechanism map of the cooling system is constructed.
[0009] Preferably, in step S2, generating the mechanism model includes: Extract topological structure information from the mechanism map, extract the connection relationship of the devices from the map, and generate the adjacency matrix A. The elements of the adjacency matrix A represent the static constraint relationship between the generalized energy from the previous device to the next device. The association matrix B is generated based on the control logic between the devices. The elements of the association matrix B represent the causal constraint relationship between the control variable K and the device.
[0010] Preferably, in step S3, converting the multi-dimensional performance indicators of the cooling system into generalized energy through weighted coefficients includes: S31: Define multi-dimensional performance indicators; S32: Data normalization: All performance indicators are standardized to dimensionless intervals through Z-score and Min-Max normalization, and the normalized data are tested by KS test to ensure that they conform to the standard normal distribution; S33: Determine the weight of each indicator based on the expert experience method combined with the dimensional analysis method; S34: multiplying each performance index by a corresponding weighting coefficient, and then summing the results to obtain generalized energy; S35: Verify the validity of the total generalized energy.
[0011] Preferably, calculating each performance indicator includes: , in, is the load of the i-th index in the j-th principal component, is its corresponding eigenvalue.
[0012] Preferably, in step S4, fusing the total generalized energy and the mechanism model to generate an optimal scheduling model includes: The total generalized energy is taken as the optimization objective, and the mechanism tensor is embedded into the optimization model as a constraint.
[0013] Preferably, the data of the mechanism model are stored using sparse form to store the boundary conditions.
[0014] Preferably, in step S4, constructing the judgment model includes: Obtain historical input data for the optimization model and build a training set; Use the training set to train the linear regression model and the polynomial regression model respectively; Calculate the mean square error of the linear regression model and the polynomial regression model using a cross-validation method; The two mean square errors are subtracted to obtain a difference value. If the difference value is greater than a preset threshold value, the data is determined to be nonlinear data.
[0015] Preferably, presetting the threshold comprises: calculating the maximum difference in mean square error of linear data in the linear regression model and the polynomial regression model in historical input data, and setting the maximum difference as the threshold.
[0016] The beneficial effects of the present invention are: The present invention constructs mechanism maps and mechanism models through digital twin technology, realizes high-precision digital mapping of the natural gas liquefaction system structure and control logic, and provides a reliable theoretical basis for optimization; by uniformly quantifying multi-dimensional performance indicators into total generalized energy, it simplifies the solution of complex multi-objective optimization problems, taking into account multiple requirements such as energy efficiency, safety and environmental protection; adopts a hybrid optimization strategy (mixed integer programming for the linear part + mechanism simulation for the nonlinear part) to significantly improve the optimization efficiency and accuracy; generates actionable equipment scheduling recommendations based on the reverse mapping of the optimization results to ensure that the optimization strategy can be directly implemented; finally, the model is dynamically updated through real-time data, so that the system has the ability to self-evolve and continuously adapt to changes in working conditions and equipment aging. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] To facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0018] Figure 1 It is a flowchart of the method steps of the present invention. DETAILED DESCRIPTION
[0019] In order to further illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.
[0020] See also Figure 1 This embodiment provides an artificial intelligence optimization method for a natural gas liquefaction system, comprising the following steps: S1: Construction of cooling system mechanism map: S11: Multi-dimensional data analysis and feature extraction: Extracting physical parameters of equipment units from design drawings. The physical parameters include analyzing the equipment operation process and operation thresholds in the operation manual, extracting the PID operation logic relationship between multiple devices, and extracting the transmission process of control signals. At the same time, the control logic information between equipment units is extracted, such as the start and stop conditions of the compressor, and the interlocking control relationship between the compressor and other equipment; S12: Convert the collected physical models into virtual models: Digital twin technology can be used to transform collected physical information into a virtual model. For example, a heat exchanger can be used to accurately simulate the fluid flow and heat exchange process within it. Existing digital twin platforms, such as Siemens' MindSphere or PTC's ThingWorx, can be used to digitally represent the physical characteristics and control logic of the equipment unit. S13: Construction of mechanism map: Based on the parsed device unit structure and control logic information, a cooling system mechanism diagram is constructed. In this diagram, nodes represent device units, and edges represent the control logic relationships between them. For example, a compressor node and a heat exchanger node are connected by an edge, and the attributes of the edge can represent the control parameters of the compressor on the heat exchanger, such as flow regulation.
[0021] You can use a graph database (such as Neo4j) to store and manage the model data of this mechanism graph; S2: Constructing a mechanism model: S21: Extraction of topological connection relationships: Extract the topological connection relationship of each device unit in the system from the mechanism map, and clarify the physical connection and information transmission path between devices; Use algorithms such as Node2Vec and GraphSAGE to map device nodes into low-dimensional vectors and automatically identify similar connection patterns; S22: Extracting operation mechanism information: Extract the operating mechanism information between equipment units, including the operating principle of the equipment, energy conversion process, etc.; S23: Constructing a mechanism model: Extract topological structure information from the mechanism map, extract the connection relationship of the devices from the map, and generate the adjacency matrix A. The elements of the adjacency matrix A represent the static constraint relationship between the generalized energy from the previous device to the next device. The association matrix B is generated based on the control logic between the devices. The elements of the association matrix B represent the causal constraint relationship between the control variable K and the device.
[0022] The mechanism model can use mathematical equations, physical models or simulation models to describe the operation process of the system. For example, for a simple compression-heat exchange system, a corresponding mathematical model can be established to describe its operating state; S3: Convert the multi-dimensional performance indicators of the cooling system into total generalized energy through weighted coefficients: S31: Define multi-dimensional performance indicators: Determine the multi-dimensional performance indicators of the cooling system, such as energy consumption, cooling capacity, equipment life, maintenance cost, etc.
[0023] The energy consumption of the equipment can be obtained by testing the power consumption and fuel consumption of the equipment; The cooling capacity can be calculated based on the cooling effect of the cooling system; the equipment life and maintenance cost can be evaluated based on the equipment's usage time and maintenance records; S32: Normalize the data by standardizing each performance indicator to a dimensionless range through Z-score and Min-Max normalization; The KS test was used to verify whether the normalized data conformed to the standard normal distribution; S33: Determine the weight of each performance indicator based on expert experience combined with dimensional analysis, including: Identify the fundamental dimensions and convert them to generalized energy: Identify the dimensions of each performance indicator data, conduct unified analysis, and convert it into energy form. The generalized energy of thermodynamic nodes is uniformly calculated to the corresponding enthalpy value, and the electricity nodes are uniformly calculated to the corresponding kilowatt-hour electricity. Contribution rate extraction: Estimate the contribution rate by converting the unit purchase price of materials or energy to broad energy.
[0024] Calculate the weight of each performance indicator, ,in, is the load of the i-th index in the j-th principal component, is its corresponding eigenvalue; S34: Constructing total generalized energy: multiplying each performance indicator by the corresponding weighting coefficient, and then summing the linear indicators to obtain the total generalized energy; A nonlinear function is introduced to deal with the coupling effect between indicators, and a product term is introduced for strongly coupled indicators (such as energy consumption and temperature difference). The weight coefficient β of the nonlinear coefficient is determined by the partial least squares method. S35: Verify the validity of generalized energy.
[0025] S4: Generate an optimal scheduling model and optimize it S41: Model Fusion: The generalized energy and mechanism models are integrated to generate an optimal scheduling model. The generalized energy is used as the optimization target, and the mechanism tensor is embedded in the optimization model as a constraint. S42: Data classification: Use the preset judgment model to divide the optimization model data into linear and nonlinear parts. For example, the simple linear relationship between the power and flow of the equipment can be classified as the linear part; while the complex nonlinear relationship between the efficiency and working conditions of the equipment can be classified as the nonlinear part. S43: Optimize the linear part The linear part is optimized using the mixed integer programming method. Mixed integer programming is an optimization algorithm used to solve linear programming problems, which can handle mixed situations of integer variables and continuous variables. S44: Optimizing nonlinear components For the nonlinear part, simulation is performed through the mechanism model. Professional simulation software can be used to simulate and analyze the system in detail to find the optimal operating parameters. For example, in the process of simulating the operation of the compressor, by changing the compressor speed, intake pressure and other parameters, the effect on the generalized energy is observed, so as to find the parameter combination that minimizes the generalized energy. S5: Based on the optimization results, the optimization results are reversely mapped to the corresponding nodes of the mechanism map to generate scheduling optimization suggestions. The scheduling optimization suggestions include the start and stop of valves and the adjustment of control lines. S6: Output scheduling optimization suggestions and generate reports, and dynamically update the mechanism map and optimal scheduling model based on the actual operation data of the liquefaction system.
[0026] The execution process of this optimization method includes: Step 1: Input: Through the mechanism diagram in Neo4j, including nodes such as compressor, heat exchanger, valve, etc. Use Cypher to query the series / parallel relationship and generate a weighted adjacency matrix A, where the weights are the flow conductivity of the pipeline; Step 2: Mechanism extraction and improvement: Input: compressor setting parameters, call the Modelica compressor template, automatically generate equations, couple the compressor and heat exchanger models in COMSOL using the adjacency matrix A, correlation matrix B, and device equations, export the ROM, and deploy it to the embedded controller for real-time optimization.
[0027] Furthermore, in step S4, constructing a judgment model for judging whether the data in the data optimization model belongs to linear data or nonlinear data includes the following steps: Data preparation and parameter optimization: Collect raw data (such as power load records, temperature, equipment parameters, etc.), handle missing values (interpolation or deletion), and handle outliers (based on the 3σ principle or boxplot identification); Specify input characteristics (e.g., temperature, humidity) and output target variables (e.g., power load); The input data is divided into training set and test set in a ratio of 7:3, and the test set is reserved for subsequent final verification; Select an appropriate polynomial order (e.g., 2-5) based on the specific data type and dimension, and use cross-validation to evaluate the performance of different orders to avoid overfitting caused by high-order polynomials. Obtain historical input data for the optimization model and build a training set; Use the training set to train the linear regression model and the polynomial regression model respectively; Calculate the mean square error of the linear regression model and the polynomial regression model using a cross-validation method; The two mean square errors are subtracted to obtain a difference. If the difference is greater than a preset threshold, the data is judged to be nonlinear data.
[0028] The preset threshold value includes: calculating the maximum difference in mean square error of linear data in the linear regression model and the polynomial regression model in the historical input data, and setting the calculated maximum difference as the threshold value.
[0029] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Although the present invention has been disclosed as above in terms of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art can, without departing from the scope of the technical solution of the present invention, make some changes or modifications to equivalent embodiments using the technical contents disclosed above. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.
Claims
1. An artificial intelligence optimization method for a natural gas liquefaction system, characterized by: The following steps are involved: S1: Based on the system design drawings and operation manuals, digital twin technology is used to analyze the structural information of the equipment units and the control logic information between the equipment units. Based on the structural information and control logic information of the equipment units, a mechanism map of the cooling system is constructed; S2: extracting the topological connection relationship of each device unit of the system and the mechanism relationship between the device units based on the mechanism map, and constructing a mechanism model; S3: convert the multi-dimensional performance indicators of the cooling system into generalized energy through weighted coefficients; S4: The total generalized energy and the mechanism model are integrated to generate an optimal scheduling model and the model is optimized. The input data of the optimization model is divided into linear and nonlinear using a preset judgment model. The nonlinear part is optimized using the mechanism model simulation, and the linear part is optimized using the mixed integer programming method to generate the corresponding optimization results. S5: Based on the optimization results, the optimization results are reversely mapped to the corresponding nodes of the mechanism map to generate scheduling optimization suggestions, which include the start and stop of valves and the adjustment of control lines; S6: Output scheduling optimization suggestions and generate reports, and dynamically update the mechanism map and optimal scheduling model based on the actual operation data of the liquefaction system.
2. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 1, characterized in that: Step S1 includes the following sub-steps: S11: Multi-dimensional data analysis and feature extraction: Extracting physical parameters of equipment units from design drawings. The physical parameters include analyzing the equipment operation process and operation thresholds in the operation manual, extracting the PID operation logic relationship between multiple devices, and extracting the transmission process of control signals. S12: Convert the collected physical model into a virtual model: Using digital twin technology, convert the collected physical information into a virtual model; S13: Construction of mechanism map: Based on the analyzed equipment unit structure information and control logic information, a mechanism map of the cooling system is constructed.
3. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 1, characterized in that: In step S2, generating the mechanism model includes: Extract topological structure information from the mechanism map, extract the connection relationship of the devices from the map, and generate the adjacency matrix A. The elements of the adjacency matrix A represent the static constraint relationship between the generalized energy from the previous device to the next device. The association matrix B is generated based on the control logic between the devices. The elements of the association matrix B represent the causal constraint relationship between the control variable K and the device.
4. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 1, characterized in that: In step S3, the multi-dimensional performance indicators of the cooling system are converted into total generalized energy using weighted coefficients, including: S31: Define multi-dimensional performance indicators; S32: Data normalization: All performance indicators are standardized to dimensionless intervals through Z-score and Min-Max normalization, and the normalized data are tested by KS to ensure that they conform to the standard normal distribution. S33: Determine the weight of each indicator based on the expert experience method combined with the dimensional analysis method; S34: multiplying each performance index by a corresponding weighting coefficient, and then summing the results to obtain a total generalized energy; S35: Verify the validity of the total generalized energy.
5. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 4, characterized in that: Calculation of various performance indicators includes: ， in, is the load of the i-th index in the j-th principal component, is its corresponding eigenvalue.
6. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 1, characterized in that: In step S4, the total generalized energy and the mechanism model are integrated to generate an optimal scheduling model, including: The total generalized energy is taken as the optimization objective, and the mechanism tensor is embedded into the optimization model as a constraint.
7. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 1, characterized in that: The data of the mechanism model is stored using sparse form to store boundary conditions.
8. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 1, characterized in that: In step S4, constructing the judgment model includes: Obtain historical input data for the optimization model and build a training set; Use the training set to train the linear regression model and the polynomial regression model respectively; The mean square error of the linear regression model and the polynomial regression model was calculated using a cross-validation method; The two mean square errors are subtracted to obtain a difference value. If the difference value is greater than a preset threshold value, the data is determined to be nonlinear data.
9. The artificial intelligence optimization method for a natural gas liquefaction system according to claim 8, characterized in that: Presetting the threshold includes: calculating a maximum difference in mean square errors of linear data in the linear regression model and the polynomial regression model in historical input data, and setting the maximum difference as the threshold.