A method and system for controlling the operation of a gas holder

Abstract

This invention discloses a gas holder operation control method and system, relating to the field of gas holder control technology. The method includes: acquiring real-time pressure data inside the gas holder and processing the real-time pressure data to extract pressure characteristic data; using an anomaly identification algorithm to perform anomaly analysis on the pressure characteristic data to identify key factors affecting the pressure stability of the gas holder; establishing a pressure regulation model based on the key factors; dynamically adjusting the real-time operating parameters of the gas holder using the pressure regulation model; and optimizing the control strategy. This invention not only improves the automation and intelligence level of gas holder operation but also effectively optimizes the control strategy, ensuring the safe and stable operation of the gas holder under high load and complex operating conditions, reducing the risk of failure, and thus improving the overall reliability and efficiency of the production system.

Description

Technical Field

[0001] This invention relates to the field of gas holder control technology, and more specifically, to a gas holder operation control method and system. Background Technology

[0002] Gas holders are critical pressure regulation and gas storage facilities in industrial enterprises, widely used in important aspects such as gas transmission and distribution, energy storage balancing, and process stabilization. In actual production processes, gas holders not only buffer load fluctuations and stabilize supply pressure, but also directly affect the safety and operational continuity of the entire industrial system. Especially under high-frequency regulation or dynamic load conditions, the operating status of gas holders is extremely sensitive to pressure fluctuations; even slight abnormalities can affect system stability. To ensure their safe and reliable operation, real-time monitoring and control of operating parameters such as gas holder pressure are necessary. Failure to adjust in a timely manner can lead to seal failure, causing hazardous gas leaks and posing serious safety hazards.

[0003] However, existing technologies lack real-time data analysis and dynamic adjustment capabilities, making it difficult to detect and handle anomalies caused by pressure fluctuations in a timely manner. They also make it difficult to accurately identify key factors affecting pressure stability, which can lead to pressure runaway or safety hazards in gas holder systems under high loads or special operating conditions, affecting production efficiency and equipment safety.

[0004] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a gas holder operation control method and system, which solves the problems mentioned in the background art, namely, the lack of real-time data analysis and dynamic adjustment capabilities, the difficulty in timely detection and handling of anomalies caused by pressure fluctuations, and the inconvenience in accurately identifying key factors affecting pressure stability. These problems lead to pressure runaway or safety hazards in gas holder systems under high load or special operating conditions, affecting production efficiency and equipment safety.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] According to one aspect of the present invention, a gas holder operation control method is provided, comprising:

[0008] Acquire real-time pressure data inside the gas holder, process the real-time pressure data, and extract pressure feature data;

[0009] Anomaly detection algorithms are used to analyze pressure feature data and construct an initial pressure feature library. A new set of candidate pressure feature combination codes is formed through perturbation and mutation operations. The new candidate pressure feature combination codes are screened based on a screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power. The projection results of all pressure feature combination codes are globally evaluated through an optimization algorithm, and the best candidate pressure feature combination code is determined based on the discriminative power, thus identifying the key factors affecting the stability of gas holder pressure.

[0010] Based on key factors, a pressure control model is established, which is used to dynamically adjust the real-time operating parameters of the gas holder and optimize the control strategy.

[0011] Furthermore, before filtering the new candidate pressure feature combination codes based on the screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power, the following steps are also included:

[0012] Randomly generate several combinations of pressure feature codes and calculate the corresponding evaluation values ​​to construct an initial pressure feature library;

[0013] Pressure feature combinations are randomly selected from the initial pressure feature library according to a preset probability;

[0014] The selected pressure feature combinations are perturbed and mutated to form a new set of candidate pressure feature combination codes.

[0015] After filtering the new candidate pressure feature combination codes based on the screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power, the following steps are also taken:

[0016] Calculate the evaluation value of the candidate stress feature combination codes after screening and compare it with the evaluation value of the worst stress feature combination code in the initial stress feature library. If the evaluation value of the candidate stress feature combination code is greater than the evaluation value of the worst stress feature combination code, then replace the worst stress feature combination code; otherwise, do not replace it.

[0017] Determine if the maximum number of iterations has been reached. If it has, output the optimal pressure feature combination code as the key factor affecting the stability of the gas holder pressure; otherwise, continue iterating.

[0018] Furthermore, the selected pressure feature combinations are perturbed and mutated to form a new set of candidate pressure feature combination codes, including:

[0019] Initialize the parameters of the perturbation mutation operation, and set the maximum number of iterations and the tabu search control strategy;

[0020] Randomly select a pressure feature combination code as the initial solution, and clear the taboo table and frequency record table at the same time;

[0021] Starting with the current pressure feature combination encoding, construct a neighborhood combination with dynamic variation of disturbance, and calculate its corresponding objective function value;

[0022] The optimal stress feature combination code that meets the criteria is selected from the candidate stress feature combination codes generated by the mutation and used as the current updated solution;

[0023] If no candidate pressure feature combination code satisfies the criterion, then the optimal variant candidate pressure feature combination code is selected from the untaboo and frequency-compliant combinations and the current solution is updated.

[0024] Record the current candidate pressure feature combination encoding result in the taboo table and frequency table, and determine whether the maximum number of iterations has been reached. If not, continue to perturb and generate new candidate pressure feature combination encodings. Otherwise, output the optimal candidate pressure feature combination encoding in all iterations as a new set of candidate pressure feature combination encodings.

[0025] Furthermore, if no candidate stress feature combination encoding satisfies the criterion, the optimal variant candidate stress feature combination encoding is selected from the untabulated and frequency-compliant combinations, and the current solution is updated, including:

[0026] Iterate through all candidate stress feature combinations that are not taboo and meet the frequency limit, and calculate their objective function values;

[0027] Select candidate pressure feature combinations with the optimal objective function value and ensure that they are within the limits of frequency and taboo list;

[0028] The selected optimal candidate pressure feature combination is encoded as the current solution, and the taboo table and frequency table are updated to record the solution.

[0029] Furthermore, before globally evaluating the projection results of all stress feature combination codes through optimization algorithms and determining the best candidate stress feature combination code based on discriminative power, the following steps are also included:

[0030] Initialize the parameters of the sieving algorithm and set the maximum number of iterations;

[0031] Randomly select a pressure feature combination code from the candidate pressure feature combination code set as the initial solution, and define the unselected pressure feature combination code set;

[0032] Calculate the projection value between the remaining pressure feature combination code and the currently selected pressure feature combination code;

[0033] After globally evaluating the projection results of all stress feature combination codes through optimization algorithms, and determining the best candidate stress feature combination codes based on discriminative power, the following steps are also included:

[0034] Determine whether the maximum number of iterations has been reached. If not, continue calculating the projection value. Otherwise, select the best candidate pressure feature combination code as the candidate pressure feature combination code with the highest discriminative power throughout all iterations.

[0035] Furthermore, the formula for calculating the projection value of the residual pressure feature combination code and the currently selected pressure feature combination code is as follows:

[0036] ;

[0037] In the formula, P i，c Represents the encoding of residual pressure feature combination s i In the currently selected pressure feature combination encoding vector s c Weighted normalized projection values ​​in the direction; s c This represents the currently selected combination of pressure features encoding vector; s i Indicates the first i A combined encoding vector of residual pressure features; G Represents a feature-diagonal matrix weighted by its feature dimensions; T Indicates the transpose operation; ε This represents a very small constant.

[0038] Furthermore, the projection results of all stress feature combination codes are globally evaluated through an optimization algorithm, and the best candidate stress feature combination codes are determined based on the discriminative power, including:

[0039] Randomly generate an initial pressure feature combination code, set it as the current optimal pressure feature combination code, and initialize the restriction level and the number of cycles;

[0040] If the restriction level does not exceed the threshold, then select some combinations from the current pressure feature combination coding neighborhood to form a projection candidate pressure feature combination coding set;

[0041] Calculate the projection distance between the candidate pressure feature combination code and the current optimal pressure feature combination code; if the projection distance of the candidate pressure feature combination code is less than the limit range, accept the candidate pressure feature combination code and update it to a new candidate pressure feature combination code;

[0042] If the discriminant of the new candidate stress feature combination code is greater than that of the current best stress feature combination code, then replace the current best stress feature combination code and reset the restriction level and the number of iterations; if the discriminant is not improved, then increase the restriction level and record the status of the current best stress feature combination code.

[0043] Once the maximum limit or number of cycles is reached, the projection results of all pressure feature combination codes are combined, and the best candidate pressure feature combination code is determined based on the discriminative power.

[0044] Furthermore, based on key factors, a pressure control model is established. This model is used to dynamically adjust the real-time operating parameters of the gas holder and optimize the control strategy, including:

[0045] The key factors affecting the stability of the gas holder pressure are identified, and the real-time state parameters of the key factors are extracted to construct the original feature sample set.

[0046] The original feature sample set was processed using a decomposition algorithm to extract key regulatory features that reflect the pattern of pressure changes.

[0047] Reconstruct feature sequences related to pressure change patterns, generate high-quality trainable samples, and divide the high-quality samples into training and testing sets;

[0048] A pressure control model is constructed, a training set is input, and the real-time operating parameters of the gas holder are dynamically adjusted using the pressure control model, and the control strategy is optimized.

[0049] Furthermore, by using a decomposition algorithm to process the original feature sample set, key regulatory features reflecting the pressure change pattern are extracted, including:

[0050] An initial feature subset is generated using a diversified sampling method, and the feature combination of each subset is optimized through a local search strategy.

[0051] A reference set is constructed by selecting several of the most discriminative and widely distributed feature subsets from the initial feature subset to ensure spatial coverage;

[0052] The reference set is subjected to feature subset cross-combination to generate new feature combinations, followed by secondary optimization.

[0053] Merge the optimized new feature subset with the reference set, reselect several optimal feature combinations, and update the reference set;

[0054] Determine if the maximum number of iterations has been reached. If it has, output the optimal feature combination as the key regulatory feature reflecting the pressure change pattern. Otherwise, continue the combination optimization loop.

[0055] According to another aspect of the present invention, a gas holder operation control system is also provided, the system comprising:

[0056] The data acquisition module is used to acquire real-time pressure data inside the gas holder, process the real-time pressure data, and extract pressure feature data.

[0057] The anomaly analysis module is used to perform anomaly analysis on pressure feature data using anomaly identification algorithms and to build an initial pressure feature library. It generates a new set of candidate pressure feature combination codes through perturbation and mutation operations. The new candidate pressure feature combination codes are screened based on a screening algorithm to obtain the candidate pressure feature combination codes with the highest discrimination. The projection results of all pressure feature combination codes are globally evaluated through an optimization algorithm, and the best candidate pressure feature combination code is determined based on the discrimination, thus identifying the key factors affecting the stability of the gas holder pressure.

[0058] The model building and optimization control module is used to build a pressure control model based on key factors, dynamically adjust the real-time operating parameters of the gas holder using the pressure control model, and optimize the control strategy.

[0059] The beneficial effects of this invention are as follows:

[0060] 1. This invention, through real-time monitoring and data processing, can promptly identify key factors affecting the stability of gas holder pressure, thereby effectively preventing abnormal situations during operation. By using anomaly identification algorithms to analyze pressure data, it helps to accurately locate the root cause of the problem and establish a pressure control model based on these key factors to achieve dynamic adjustment. This not only improves the automation and intelligence level of gas holder operation but also effectively optimizes control strategies, ensuring the safe and stable operation of the gas holder under high load and complex operating conditions, reducing the risk of failure, and thus improving the overall reliability and efficiency of the production system.

[0061] 2. This invention improves the accuracy and diversity of feature extraction by combining anomaly identification algorithms with optimization strategies such as perturbation mutation, tabu search, and screening algorithms. It effectively avoids the local optimum dilemma, enhances the generalization ability of the model, thereby improving the safety and stability of the gas holder under complex operating conditions. At the same time, it reduces control delay and ensures continuous and stable production operation. Furthermore, it can effectively identify key feature combinations that affect the pressure stability of the gas holder, and realize refined dynamic pressure control.

[0062] 3. This invention extracts key features of pressure regulation through decomposition algorithms and trains a pressure regulation model with high-quality samples to achieve dynamic adjustment of gas holder operating parameters and optimization of control strategies. This effectively improves the accuracy of pressure response and regulation efficiency, enhances the stability and safety of the system under complex operating conditions, and reduces the operating risk and energy loss of the gas holder. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a flowchart of a gas holder operation control method according to an embodiment of the present invention;

[0065] Figure 2 This is a schematic block diagram of a gas holder operation control system according to an embodiment of the present invention.

[0066] In the picture:

[0067] 1. Data acquisition module; 2. Anomaly analysis module; 3. Model building and optimization control module. Detailed Implementation

[0068] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0069] In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Furthermore, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0070] According to an embodiment of the present invention, a gas holder operation control method and system are provided.

[0071] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the gas holder operation control method according to an embodiment of the present invention includes:

[0072] S1. Obtain real-time pressure data inside the gas holder, process the real-time pressure data, and extract pressure feature data;

[0073] Specifically, real-time pressure data of the gas holder can be obtained through pressure sensors installed in different parts of the gas holder (such as cavity pressure sensors, pipeline pressure transmitters, piston lower chamber pressure sensors, etc.).

[0074] Specifically, pressure data includes chamber pressure, inlet / outlet pressure, pressure difference between the upper and lower chambers of the piston, ambient temperature correction value, peak value, valley value, average pressure, and pressure change rate (slope).

[0075] Specifically, pressure characteristic data include the first derivative of pressure, rate of change of pressure, average growth rate, peak frequency, number of instantaneous mutations, stability coefficient, dominant frequency, frequency energy concentration, pressure difference between sensing points, and delay response characteristics.

[0076] S2. Anomaly detection algorithms are used to analyze pressure feature data and construct an initial pressure feature library. A new set of candidate pressure feature combination codes is formed through perturbation and mutation operations. The new candidate pressure feature combination codes are screened based on a sieving algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power. The projection results of all pressure feature combination codes are globally evaluated through an optimization algorithm, and the best candidate pressure feature combination code is determined based on the discriminative power, thus identifying the key factors affecting the stability of the gas holder pressure.

[0077] Specifically, key factors include fluctuations in gas source operating conditions, abnormal structural component conditions, control response deviations, and environmental disturbances.

[0078] S3. Based on key factors, establish a pressure control model, use the pressure control model to dynamically adjust the real-time operating parameters of the gas holder, and optimize the control strategy.

[0079] Specifically, after identifying the key factors affecting the stability of the gas holder pressure, the pressure control model senses the changes in the state of these factors in real time, and then dynamically adjusts the operating parameters of the gas holder. The model can automatically optimize the control strategy based on the current characteristic changes, so as to achieve precise regulation of the gas holder pressure and ensure safe operation.

[0080] In this optional embodiment, an anomaly detection algorithm is used to perform anomaly analysis on the pressure characteristic data to identify key factors affecting the stability of the gas holder pressure, including:

[0081] Randomly generate several combinations of pressure feature codes and calculate the corresponding evaluation values ​​to construct an initial pressure feature library;

[0082] Pressure feature combinations are randomly selected from the initial pressure feature library according to a preset probability;

[0083] The selected pressure feature combinations are perturbed and mutated to form a new set of candidate pressure feature combination codes.

[0084] The new candidate pressure feature combination codes are screened based on the screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power.

[0085] Calculate the evaluation value of the candidate stress feature combination codes after screening and compare it with the evaluation value of the worst stress feature combination code in the initial stress feature library. If the evaluation value of the candidate stress feature combination code is greater than the evaluation value of the worst stress feature combination code, then replace the worst stress feature combination code; otherwise, do not replace it.

[0086] Determine if the maximum number of iterations has been reached. If it has, output the optimal pressure feature combination code as the key factor affecting the stability of the gas holder pressure; otherwise, continue iterating.

[0087] Specifically, firstly, several stress feature combinations are randomly generated and their evaluation values ​​are calculated to construct an initial feature library. Then, combinations are randomly selected based on preset probabilities for perturbation and mutation, generating a new set of candidate combinations. Next, a screening algorithm is used to select the candidate combination with the highest discriminative power, and its evaluation value is calculated and compared with the worst combination in the initial library; if it is better, it is replaced. Finally, the number of iterations is determined; if the required number is not reached, optimization continues, and once the required number is reached, the optimal combination is output as the key factor. This accurately identifies key features, improving anomaly identification efficiency and control reliability.

[0088] Specifically, the anomaly detection algorithm is a harmony search algorithm, an intelligent optimization algorithm that simulates the process of finding harmonious notes in musical improvisation. In this invention, the harmony search algorithm is used to find the optimal combination from pressure feature combinations: first, an initial feature library is generated; then, candidate combinations are generated through perturbation mutation and screening mechanisms; the worst solution is continuously updated iteratively; and finally, the optimal combination is output as the key factor. Therefore, it has strong global search capabilities and can efficiently identify the core features affecting the pressure of the gas holder.

[0089] In this optional embodiment, perturbation and mutation operations (based on tabu search algorithm) are performed on the selected stress feature combinations to form a new candidate stress feature combination encoding set, including:

[0090] Initialize the parameters of the perturbation mutation operation, and set the maximum number of iterations and the tabu search control strategy;

[0091] Randomly select a pressure feature combination code as the initial solution, and clear the taboo table and frequency record table at the same time;

[0092] Starting with the current pressure feature combination encoding, construct a neighborhood combination with dynamic variation of disturbance, and calculate its corresponding objective function value;

[0093] The optimal stress feature combination code that meets the criteria is selected from the candidate stress feature combination codes generated by the mutation and used as the current updated solution;

[0094] If no candidate pressure feature combination code satisfies the criterion, then the optimal variant candidate pressure feature combination code is selected from the untaboo and frequency-compliant combinations and the current solution is updated.

[0095] Record the current candidate pressure feature combination encoding result in the taboo table and frequency table, and determine whether the maximum number of iterations has been reached. If not, continue to perturb and generate new candidate pressure feature combination encodings. Otherwise, output the optimal candidate pressure feature combination encoding in all iterations as a new set of candidate pressure feature combination encodings.

[0096] Specifically, the perturbation mutation parameters are first initialized, the maximum number of iterations is set, and a tabu search strategy is configured. Then, a stress feature combination is randomly selected as the initial solution, and the tabu table and frequency table are cleared. Multiple perturbation combinations are generated starting from the initial solution, and the objective function value is calculated. The optimal combination that meets the criteria is then selected for updating. If no combination meets the criteria, the optimal solution is selected from the combinations that have never been tabu and whose frequency complies with the rules. In each round, the current combination is recorded in the tabu and frequency tables. If the iteration is not completed, generation continues, and finally, the optimal combination set is output. This avoids the local optimum trap and improves the global search capability and convergence efficiency of feature combinations.

[0097] Specifically, the selected stress feature combinations are perturbed and mutated using a tabu search algorithm. Tabu search is an intelligent optimization method based on local search, which avoids getting trapped in local optima by introducing a "taboo table." In this invention, tabu search is used to perturb and mutate stress feature combinations: after initializing parameters, a neighborhood solution is generated based on the current combination, and the target value is evaluated; if no optimal candidate satisfies the criteria, the optimal solution is updated from combinations that have never been taboo and whose frequency complies with the rules; after iterating to the upper limit, the optimal solution set is output. This enhances the global nature of the search and improves the efficiency and diversity of feature optimization.

[0098] In this optional embodiment, if no candidate stress feature combination encoding satisfies the criterion, the optimal variant candidate stress feature combination encoding is selected from the untabulated and frequency-compliant combinations, and the current solution is updated, including:

[0099] Iterate through all candidate stress feature combinations that are not taboo and meet the frequency limit, and calculate their objective function values;

[0100] Select candidate pressure feature combinations with the optimal objective function value and ensure that they are within the limits of frequency and taboo list;

[0101] The selected optimal candidate pressure feature combination is encoded as the current solution, and the taboo table and frequency table are updated to record the solution.

[0102] Specifically, firstly, if none of the current perturbation candidate combinations meet the selection criteria, then all candidate pressure feature combinations that are not taboo and meet the frequency requirements are traversed, and their objective function values ​​are calculated. Next, the combination with the optimal objective function value is selected, ensuring it is not taboo and meets the frequency limit. This combination is then used as the current solution for updating, and is simultaneously recorded in the taboo table and frequency table to ensure combination diversity and non-repetitive search paths. If the iteration limit is not reached, the next round of perturbation optimization continues. This enhances global search coverage and effectively avoids getting trapped in local optima.

[0103] In this optional embodiment, the new candidate pressure feature combination codes are screened based on a screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power, including:

[0104] Initialize the parameters of the sieving algorithm and set the maximum number of iterations;

[0105] Randomly select a pressure feature combination code from the candidate pressure feature combination code set as the initial solution, and define the unselected pressure feature combination code set;

[0106] Calculate the projection value between the remaining pressure feature combination code and the currently selected pressure feature combination code;

[0107] The projection results of all stress feature combination codes are globally evaluated by optimizing the algorithm, and the best candidate stress feature combination codes are determined based on the discriminative power.

[0108] Determine whether the maximum number of iterations has been reached. If not, continue calculating the projection value. Otherwise, select the best candidate pressure feature combination code as the candidate pressure feature combination code with the highest discriminative power throughout all iterations.

[0109] Specifically, the screening algorithm parameters are first initialized and the maximum number of iterations is set. Then, an initial combination is randomly selected from the candidate pressure feature combination set, and an unselected combination set is defined. Next, the projection values ​​of the remaining combinations and the current combination are calculated, and the discriminative power of all combinations is evaluated through an optimization algorithm to determine the current optimal combination. If the maximum number of iterations has not been reached, a new round of projection and evaluation is performed, and finally, the combination code with the highest discriminative power is output as the optimal candidate. This enhances the feature discrimination capability and improves the effectiveness and accuracy of abnormal feature identification.

[0110] Specifically, the screening algorithm is a continuous projection algorithm, an optimization method for feature selection and dimensionality reduction, which improves feature discriminativeness through multiple rounds of projection evaluation. In this invention, after algorithm initialization, an initial solution is randomly selected from candidate feature combinations, and its projection value with the remaining combinations is calculated. Then, the combination with the best discriminativeness is selected through global evaluation. If the iteration limit is not reached, the projection and selection continue, and finally the best candidate combination is output. This strengthens feature discriminativeness and helps identify the most representative abnormal feature combinations.

[0111] In this optional embodiment, the formula for calculating the projection value of the remaining pressure feature combination code and the currently selected pressure feature combination code is as follows:

[0112] ;

[0113] In the formula, P i，c Represents the encoding of residual pressure feature combination s i In the currently selected pressure feature combination encoding vector s cWeighted normalized projection values ​​in the direction; s c This represents the currently selected combination of pressure features encoding vector; s i Indicates the first i A combined encoding vector of residual pressure features; G Represents a feature-diagonal matrix weighted by its feature dimensions; T Indicates the transpose operation; ε This represents a very small constant.

[0114] In this optional embodiment, the projection results of all pressure feature combination codes are globally evaluated using an optimization algorithm, and the best candidate pressure feature combination code is determined based on the discriminative power, including:

[0115] Randomly generate an initial pressure feature combination code, set it as the current optimal pressure feature combination code, and initialize the restriction level and the number of cycles;

[0116] If the restriction level does not exceed the threshold, then select some combinations from the current pressure feature combination coding neighborhood to form a projection candidate pressure feature combination coding set;

[0117] Calculate the projection distance between the candidate pressure feature combination code and the current optimal pressure feature combination code; if the projection distance of the candidate pressure feature combination code is less than the limit range, accept the candidate pressure feature combination code and update it to a new candidate pressure feature combination code;

[0118] If the discriminant of the new candidate stress feature combination code is greater than that of the current best stress feature combination code, then replace the current best stress feature combination code and reset the restriction level and the number of iterations; if the discriminant is not improved, then increase the restriction level and record the status of the current best stress feature combination code.

[0119] Once the maximum limit or number of cycles is reached, the projection results of all pressure feature combination codes are combined, and the best candidate pressure feature combination code is determined based on the discriminative power.

[0120] Specifically, firstly, an initial stress feature combination code is randomly generated and set as the current optimal combination, while simultaneously initializing the restriction level and the number of iterations. Then, a subset of candidate combinations are selected from the neighborhood of the current combination, and their projected distances to the optimal combination are calculated. If the distance is less than the restriction range, it is updated to a new candidate combination. If the new combination has higher discriminative power than the optimal combination, it is updated to the optimal combination, and the restriction level and number of iterations are reset. If the discriminative power is not improved, the restriction level is increased, and the current combination state is recorded. When the maximum restriction or number of iterations is reached, the best candidate combination is selected based on a comprehensive evaluation of the discriminative power. This optimizes feature combination selection and improves system performance and stability.

[0121] Specifically, the optimization algorithm is a predator-prey search algorithm, a biomimetic intelligent optimization method that simulates the behavior of predators searching for prey in a dynamic environment. In this invention, the algorithm randomly initializes the optimal combination of pressure features and filters candidate solutions in the neighborhood; it determines whether to update the current solution based on the projection distance and discriminative power, and avoids premature convergence through a constraint level mechanism; finally, after iteration, it comprehensively evaluates the results to determine the optimal combination. This enhances the flexibility of the search strategy and improves the global optimality of feature selection.

[0122] In this optional embodiment, a pressure regulation model (i.e., a neural network model) is established based on key factors. This model is used to dynamically adjust the real-time operating parameters of the gas holder and optimize the control strategy, including:

[0123] The key factors affecting the stability of the gas holder pressure are identified, and the real-time state parameters of the key factors are extracted to construct the original feature sample set.

[0124] The original feature sample set was processed using a decomposition algorithm to extract key regulatory features that reflect the pattern of pressure changes.

[0125] Reconstruct feature sequences related to pressure change patterns, generate high-quality trainable samples, and divide the high-quality samples into training and testing sets;

[0126] A pressure control model is constructed, a training set is input, and the real-time operating parameters of the gas holder are dynamically adjusted using the pressure control model, and the control strategy is optimized.

[0127] Specifically, the process begins by identifying key factors affecting the stability of the gas holder pressure, such as environmental disturbances and gas source fluctuations, and extracting corresponding state parameters in real time to construct an initial feature sample set. Then, a decomposition algorithm is used to extract control features that accurately reflect pressure change patterns, reconstructing the feature sequence to generate high-quality samples. Next, the samples are divided into training and testing sets and input into the constructed pressure control model for training. Finally, the model can dynamically sense state changes, adjust operating parameters in real time, and optimize control strategies. This improves the accuracy of control response and achieves intelligent management and control of pressure stability.

[0128] In this optional embodiment, the original feature sample set is processed using a decomposition algorithm to extract key regulatory features reflecting the pressure change pattern, including:

[0129] An initial feature subset is generated using a diversified sampling method, and the feature combination of each subset is optimized through a local search strategy.

[0130] A reference set is constructed by selecting several of the most discriminative and widely distributed feature subsets from the initial feature subset to ensure spatial coverage;

[0131] The reference set is subjected to feature subset cross-combination to generate new feature combinations, followed by secondary optimization.

[0132] Merge the optimized new feature subset with the reference set, reselect several optimal feature combinations, and update the reference set;

[0133] Determine if the maximum number of iterations has been reached. If it has, output the optimal feature combination as the key regulatory feature reflecting the pressure change pattern. Otherwise, continue the combination optimization loop.

[0134] Specifically, firstly, a diversified sampling method is used to generate several initial feature subsets, and the combination of each subset is optimized through local search. Then, several subsets with high discriminative power and dispersed distribution are selected to form a reference set, improving spatial coverage. Next, the reference set is cross-combined and optimized a second time to obtain new feature combinations. The optimization result is then merged with the reference set to update the new optimal subset set. If the maximum number of iterations has not been reached, the combination optimization continues cyclically until the key regulatory features are finally output. This enhances the feature representation ability and accurately extracts key elements highly correlated with pressure changes.

[0135] Specifically, the decomposition algorithm is a distributed search algorithm, an optimization method based on swarm search, which simulates diverse search behavior to avoid prematurely getting trapped in local optima. In this invention, the algorithm generates an initial feature subset through diversified sampling and optimizes each subset using local search; then, it selects the most discriminative subset to form a reference set to ensure spatial coverage; new combinations are generated through feature cross-validation and secondary optimization to update the reference set; after iterating to the maximum number of times, the optimal feature is output. This improves the diversity of feature search and ensures the comprehensiveness and accuracy of key regulatory features.

[0136] According to another embodiment of the invention, such as Figure 2 As shown, a gas holder operation control system is also provided, the system comprising:

[0137] Data acquisition module 1 is used to acquire real-time pressure data inside the gas holder, process the real-time pressure data, and extract pressure feature data.

[0138] Anomaly analysis module 2 is used to perform anomaly analysis on pressure feature data using anomaly identification algorithms and to build an initial pressure feature library. It forms a new set of candidate pressure feature combination codes through perturbation and mutation operations. Based on the screening algorithm, the new candidate pressure feature combination codes are screened to obtain the candidate pressure feature combination codes with the highest discrimination. The projection results of all pressure feature combination codes are globally evaluated through optimization algorithms, and the best candidate pressure feature combination code is determined according to the discrimination, thus identifying the key factors affecting the stability of the gas holder pressure.

[0139] The model building and optimization control module 3 is used to build a pressure control model based on key factors, dynamically adjust the real-time operating parameters of the gas holder using the pressure control model, and optimize the control strategy.

[0140] The data acquisition module 1 is connected to the anomaly analysis module 2 and the model building and optimization control module 3.

[0141] In summary, by utilizing the above-mentioned technical solutions of this invention, the present invention improves the accuracy and diversity of feature extraction through anomaly identification algorithms combined with optimization strategies such as perturbation mutation, tabu search, and screening algorithms. This effectively avoids local optima and enhances the generalization ability of the model, thereby improving the safety and stability of the gas holder under complex operating conditions. Simultaneously, it reduces control delays, ensures continuous and stable production operation, and effectively identifies key feature combinations affecting gas holder pressure stability, achieving refined dynamic pressure control. This invention extracts key features for pressure control through decomposition algorithms and trains the pressure control model using high-quality samples, enabling dynamic adjustment of gas holder operating parameters and optimization of control strategies. This effectively improves the accuracy of pressure response and control efficiency, enhances the stability and safety of the system under complex operating conditions, and ultimately reduces the operational risks and energy losses of the gas holder.

[0142] To facilitate understanding of the above technical solutions of the present invention, the following provides a detailed description of the operation control of the gas holder in actual operation.

[0143] I. A certain industrial gas holder system is equipped with a high-precision cavity pressure sensor to monitor the internal pressure of the cavity, with the goal of maintaining the pressure within a stable range of 45±2kPa.

[0144] Acquire and process real-time cavity pressure data: Pressure sampling frequency: once per minute. The raw data collected during a certain period is shown in Table 1.

[0145] Table 1. Raw data collected during a certain period

[0146]

[0147] Calculate the rate of change of pressure:

[0148] (rate of change) 12:03 =46.1-45.3=+0.8kPa / min.

[0149] (rate of change) 12:04 =47.0-46.1=+0.9kPa / min.

[0150] Specifically, the rate of change in the above equation 12:03 and rate of change 12:04These represent the rate of change of cavity pressure at time 12:03 and the rate of change of cavity pressure at time 12:04, respectively.

[0151] II. Anomaly Identification and Key Factor Extraction:

[0152] Normal rate of change range set: ±0.3 kPa / min (empirical threshold).

[0153] The measured pressure change rate exceeded the threshold for two consecutive minutes, which the system determined to be an abnormal fluctuation. The environmental data collected at the same time is shown in Table 2.

[0154] Table 2. Collected Environmental Data

[0155]

[0156] After comparison with the anomaly identification model, it was determined that "a sharp increase in ambient temperature + a sudden increase in wind speed" was the key factor causing the sudden increase in the pressure change rate.

[0157] III. Constructing a neural network model to achieve regulation:

[0158] Three-layer fully connected neural network:

[0159] Input layer: ambient temperature, wind speed, and pressure change rate.

[0160] Hidden layer: Units 64-32, ReLU activated.

[0161] Output layer: Adjustment value ΔP, used to guide the controller to change the valve opening.

[0162] Example input:

[0163] Temperature: 33.2℃.

[0164] Wind speed: 5.6 m / s.

[0165] Current pressure change rate: +0.9 kPa / min.

[0166] Neural network output: Predictive control output ΔP = -1.1 kPa.

[0167] System control response: Real-time control logic calculations require increasing the opening of the gas holder outlet valve by 12% to reduce the internal cavity pressure.

[0168] Control result feedback: After the control action was executed, the pressure dropped to 45.8 kPa in the next minute, and the rate of change dropped to -0.6 kPa / min; after further fine-tuning, the cavity pressure returned to the target range (45.1 kPa); the model used this environmental disturbance + pressure response as training samples for feedback learning, and continuously optimized the control accuracy.

[0169] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A gas cabinet operation control method characterized by, include: Acquire real-time pressure data inside the gas holder, process the real-time pressure data, and extract pressure feature data; Anomaly detection algorithms are used to analyze pressure feature data and construct an initial pressure feature library. A new set of candidate pressure feature combination codes is formed through perturbation and mutation operations. The new candidate pressure feature combination codes are screened based on a screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power. The projection results of all pressure feature combination codes are globally evaluated through an optimization algorithm, and the best candidate pressure feature combination code is determined based on the discriminative power, thus identifying the key factors affecting the stability of gas holder pressure. Before performing a global evaluation of the projection results of all pressure feature combination codes using an optimization algorithm, and determining the best candidate pressure feature combination code based on the discriminative power, the process also includes: Initialize the parameters of the sieving algorithm and set the maximum number of iterations; Randomly select a pressure feature combination code from the candidate pressure feature combination code set as the initial solution, and define the unselected pressure feature combination code set; Calculate the projection value between the remaining pressure feature combination code and the currently selected pressure feature combination code; The process of globally evaluating the projection results of all pressure feature combination codes using an optimization algorithm and determining the best candidate pressure feature combination code based on the discriminative power also includes: Determine whether the maximum number of iterations has been reached. If not, continue to calculate the projection value. Otherwise, select the best candidate pressure feature combination code as the candidate pressure feature combination code with the highest discriminative power during all iterations. The formula for calculating the projection value of the residual pressure feature combination code and the currently selected pressure feature combination code is as follows: ； wherein P i，c representing a residual pressure feature combination encoding s i in the currently selected pressure feature combination encoding vector s c a weighted normalized projection value in the direction of s c representing the currently selected pressure feature combination encoding vector; s i representing the i-th i residual pressure feature combination encoding vector; G representing a feature dimension weighted diagonal matrix; T representing a transpose operation; ε representing a minimum constant; Based on key factors, a pressure control model is established, which is used to dynamically adjust the real-time operating parameters of the gas holder and optimize the control strategy.

2. The gas holder operation control method according to claim 1, characterized in that, Before the step of filtering new candidate pressure feature combination codes based on the screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power, the following steps are also included: Randomly generate several combinations of pressure feature codes and calculate the corresponding evaluation values ​​to construct an initial pressure feature library; Pressure feature combinations are randomly selected from the initial pressure feature library according to a preset probability; The selected pressure feature combinations are perturbed and mutated to form a new set of candidate pressure feature combination codes. The step of filtering new candidate pressure feature combination codes based on the screening algorithm to obtain the candidate pressure feature combination codes with the highest discriminative power also includes: Calculate the evaluation value of the candidate stress feature combination codes after screening and compare it with the evaluation value of the worst stress feature combination code in the initial stress feature library. If the evaluation value of the candidate stress feature combination code is greater than the evaluation value of the worst stress feature combination code, then replace the worst stress feature combination code; otherwise, do not replace it. Determine if the maximum number of iterations has been reached. If it has, output the optimal pressure feature combination code as the key factor affecting the stability of the gas holder pressure; otherwise, continue iterating.

3. The gas holder operation control method according to claim 2, characterized in that, The step of performing a perturbation and mutation operation on the selected pressure feature combinations to form a new candidate pressure feature combination encoding set includes: Initialize the parameters of the perturbation mutation operation, and set the maximum number of iterations and the tabu search control strategy; Randomly select a pressure feature combination code as the initial solution, and clear the taboo table and frequency record table at the same time; Starting with the current pressure feature combination encoding, construct a neighborhood combination with dynamic variation of disturbance, and calculate its corresponding objective function value; The optimal stress feature combination code that meets the criteria is selected from the candidate stress feature combination codes generated by the mutation and used as the current updated solution; If no candidate pressure feature combination code satisfies the criterion, then the optimal variant candidate pressure feature combination code is selected from the untaboo and frequency-compliant combinations and the current solution is updated. Record the current candidate pressure feature combination encoding result in the taboo table and frequency table, and determine whether the maximum number of iterations has been reached. If not, continue to perturb and generate new candidate pressure feature combination encodings. Otherwise, output the optimal candidate pressure feature combination encoding in all iterations as a new set of candidate pressure feature combination encodings.

4. The gas holder operation control method according to claim 3, characterized in that, If no candidate pressure feature combination encoding satisfies the criterion, then selecting the optimal variant candidate pressure feature combination encoding from untabled and frequency-compliant combinations and updating the current solution includes: Iterate through all candidate stress feature combinations that are not taboo and meet the frequency limit, and calculate their objective function values; Select candidate pressure feature combinations with the optimal objective function value and ensure that they are within the limits of frequency and taboo list; The selected optimal candidate pressure feature combination is encoded as the current solution, and the taboo table and frequency table are updated to record the solution.

5. The gas holder operation control method according to claim 1, characterized in that, The step of globally evaluating the projection results of all stress feature combination codes using an optimization algorithm and determining the best candidate stress feature combination code based on the discriminative power includes: Randomly generate an initial pressure feature combination code, set it as the current optimal pressure feature combination code, and initialize the restriction level and the number of cycles; If the restriction level does not exceed the threshold, then select some combinations from the current pressure feature combination coding neighborhood to form a projection candidate pressure feature combination coding set; Calculate the projection distance between the candidate pressure feature combination code and the current optimal pressure feature combination code; if the projection distance of the candidate pressure feature combination code is less than the limit range, accept the candidate pressure feature combination code and update it to a new candidate pressure feature combination code; If the discriminant of the new candidate stress feature combination code is greater than that of the current best stress feature combination code, then replace the current best stress feature combination code and reset the restriction level and the number of iterations; if the discriminant is not improved, then increase the restriction level and record the status of the current best stress feature combination code. Once the maximum limit or number of cycles is reached, the projection results of all pressure feature combination codes are combined, and the best candidate pressure feature combination code is determined based on the discriminative power.

6. The gas holder operation control method according to claim 1, characterized in that, The process of establishing a pressure control model based on key factors, dynamically adjusting the real-time operating parameters of the gas holder using the pressure control model, and optimizing the control strategy includes: The key factors affecting the stability of the gas holder pressure are identified, and the real-time state parameters of the key factors are extracted to construct the original feature sample set. The original feature sample set was processed using a decomposition algorithm to extract key regulatory features that reflect the pattern of pressure changes. Reconstruct feature sequences related to pressure change patterns, generate high-quality trainable samples, and divide the high-quality samples into training and testing sets; A pressure control model is constructed, a training set is input, and the real-time operating parameters of the gas holder are dynamically adjusted using the pressure control model, and the control strategy is optimized.

7. The gas holder operation control method according to claim 6, characterized in that, The process of using a decomposition algorithm to process the original feature sample set and extract key regulatory features reflecting the pressure change pattern includes: An initial feature subset is generated using a diversified sampling method, and the feature combination of each subset is optimized through a local search strategy. A reference set is constructed by selecting several of the most discriminative and widely distributed feature subsets from the initial feature subset to ensure spatial coverage; The reference set is subjected to feature subset cross-combination to generate new feature combinations, followed by secondary optimization. Merge the optimized new feature subset with the reference set, reselect several optimal feature combinations, and update the reference set; Determine if the maximum number of iterations has been reached. If it has, output the optimal feature combination as the key regulatory feature reflecting the pressure change pattern. Otherwise, continue the combination optimization loop.

8. A gas holder operation control system, used to implement the gas holder operation control method according to any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to acquire real-time pressure data inside the gas holder, process the real-time pressure data, and extract pressure feature data. The anomaly analysis module is used to perform anomaly analysis on pressure feature data using anomaly identification algorithms and to build an initial pressure feature library. It generates a new set of candidate pressure feature combination codes through perturbation and mutation operations. The new candidate pressure feature combination codes are screened based on a screening algorithm to obtain the candidate pressure feature combination codes with the highest discrimination. The projection results of all pressure feature combination codes are globally evaluated through an optimization algorithm, and the best candidate pressure feature combination code is determined based on the discrimination, thus identifying the key factors affecting the stability of the gas holder pressure. The model building and optimization control module is used to build a pressure control model based on key factors, dynamically adjust the real-time operating parameters of the gas holder using the pressure control model, and optimize the control strategy.

Images