Gas steel cylinder multi-stage pressure control method based on self-adaptive threshold value

Abstract

The invention relates to the technical field of gas steel cylinder pressure control, in particular to a gas steel cylinder multi-stage pressure control method based on a self-adaptive threshold value, which can dynamically correct the influence of ambient temperature on a pressure reference value through a constructed pressure-temperature compensation function, eliminate measurement errors caused by temperature drift, and improve the accuracy of pressure control. A pressure fluctuation entropy value is calculated in real time based on a sliding window algorithm, a self-adaptive threshold value adjustment coefficient alpha is generated in combination with historical working condition database matching, a control threshold value is dynamically adjusted along with gas flow velocity fluctuation and equipment aging degree, and the pressure over-limit risk is reduced; according to the method, fuzzy neural network control is introduced in a critical adjustment stage through a multi-stage pressure adjustment strategy, an improved radial basis function dynamic updating mechanism can adapt to a pressure sudden change mode online, the control response speed is increased, and the steady-state error is controlled within + / -1.5%.

Description

Technical Field

[0001] The present invention relates to the technical field of gas cylinder pressure control, and particularly to a multi-stage pressure control method for gas cylinders based on an adaptive threshold. Background Art

[0002] At present, gas cylinders are widely used in various fields, such as medical treatment, chemical industry, welding, etc. To ensure safe use, real-time monitoring of the internal pressure of gas cylinders is crucial. Currently, the following technical problems mainly exist in the existing technologies:

[0003] 1. In the existing technologies, the pressure control of gas cylinders usually relies on fixed thresholds and manually adjusted pressure reducing valves. It cannot dynamically adapt to complex working conditions such as environmental temperature changes, gas flow rate fluctuations, and equipment aging, resulting in low pressure control accuracy and safety risks. Its pressure regulating layer needs to be manually adjusted in position, and the manifold system only switches the gas supply group through fixed pressure thresholds, lacking real-time self-adaptability.

[0004] 2. In the existing technologies, single-stage decompression or simple interlock mechanisms are mostly adopted in the safety pressure relief stage of gas cylinders. The pressure reducing valves of gas cylinders set in the existing technologies only rely on overpressure protection, lacking a hierarchical pressure relief strategy and unable to effectively cope with the chain risks caused by pressure mutations.

[0005] 3. In the existing technologies, the sensor configuration of gas cylinders lacks a cross-check mechanism, and the single-point measurement error is likely to cause control failure.

[0006] Therefore, a multi-stage pressure control method for gas cylinders based on an adaptive threshold that can solve the above problems is needed. Summary of the Invention

[0007] The present invention provides a multi-stage pressure control method for gas cylinders based on an adaptive threshold. Through the constructed pressure-temperature compensation function, the present invention can dynamically correct the influence of environmental temperature on the pressure reference value, eliminate the measurement error caused by temperature drift, and calculate the pressure fluctuation entropy value in real time based on the sliding window algorithm. By combining with the historical working condition database to match and generate an adaptive threshold adjustment coefficient α, the control threshold can be dynamically adjusted with the gas flow rate fluctuation and equipment aging degree, reducing the risk of pressure overrun. Through the multi-stage pressure regulation strategy, the fuzzy neural network control is introduced in the critical adjustment stage, and its improved radial basis function dynamic update mechanism can adapt to the pressure mutation mode online, improving the control response speed.

[0008] The technical solution adopted by the present invention to solve the above technical problems is: a multi-stage pressure control method for gas cylinders based on an adaptive threshold, including the following steps:

[0009] Step S1: Real-time collect the data of the pressure value, ambient temperature and gas flow rate inside the gas cylinder through multi-source sensors;

[0010] Step S2: Construct a pressure stage division model, and determine whether the gas cylinder enters the normal filling stage, the critical adjustment stage or the safe release stage according to the dynamic deviation amount between the current pressure value and the preset reference threshold;

[0011] Step S3: Dynamically calculate the pressure fluctuation characteristic value inside the gas cylinder by using the sliding window algorithm, and generate an adaptive threshold adjustment coefficient α based on the matching degree between the characteristic value and the historical working condition database;

[0012] Step S4: Perform a multiplication operation on the reference threshold and the adjustment coefficient α to generate the dynamic control threshold for each stage;

[0013] Step S5: According to the stage determination result and the dynamic control threshold, activate the corresponding pressure regulation strategy, including: starting the PID proportional valve linear control in the normal filling stage, starting the fuzzy neural network control in the critical adjustment stage, and triggering the multi-stage decompression interlock mechanism in the safe release stage;

[0014] Step S6: Perform a time-domain analysis on the execution effect through the pressure state feedback module. When the pressure change gradient of the gas cylinder exceeds the set value within 3 consecutive sampling periods, trigger the threshold re-optimization instruction.

[0015] Further, the specific implementation steps of the pressure stage division model in step S2 include:

[0016] S21: Establish a pressure-temperature compensation function f(T)=P base ×(1 + 0.023×(T - T0));

[0017] where T represents the actual temperature of the current environment;

[0018] where T0 is the calibration temperature, and f(T) represents the pressure value inside the gas cylinder after temperature compensation, which is a function of temperature T;

[0019] where P base represents the basic pressure value of the gas cylinder measured or set at the calibration temperature T0, which refers to the pressure value that should be in the gas cylinder at room temperature;

[0020] S22: Calculate the normalized deviation amount

[0021] S23: When |δ| ≤ 5%, it is determined as the normal filling stage; when 5% < |δ| ≤ 15%, it is determined as the critical adjustment stage; when |δ| > 15%, it is determined as the safe release stage.

[0022] Further, the sliding window algorithm uses dynamic window size determination, and its specific steps include:

[0023] S31: Calculate the minimum analysis period according to the volume V of the gas cylinder and the current speed v of the gas flowing out of or into the gas cylinder

[0024] S32: Set the window width where Δt is the sampling interval time;

[0025] S33: Calculate the pressure fluctuation entropy value H=-∑(p i ·log2(p i )) within the window, where H is the pressure fluctuation entropy value used to quantify the uncertainty of the pressure fluctuation, and p i is the probability of a certain pressure value appearing within the window.

[0026] Further, starting the fuzzy neural network control in the critical adjustment stage includes constructing a five-layer network structure, including an input layer, a membership degree generation layer, a rule reasoning layer, a feature fusion layer, and an output defuzzification layer; the input layer includes the pressure deviation, the deviation change rate, and the temperature compensation factor, and the feature fusion layer uses an improved radial basis function, and its center vector is dynamically updated through an online clustering algorithm; the output layer generates the valve opening correction amount.

[0027] Further, the dynamic update mechanism of the improved radial basis function includes: automatically merging similar feature clustering centers according to the real-time collected pressure change trend data, and when it is detected that the pressure fluctuation pattern changes suddenly, creating a new clustering node and freezing the working state of the historical low-frequency nodes.

[0028] Further, triggering the multi-stage decompression interlock mechanism in the safety release stage includes:

[0029] When the gas cylinder enters the safety release stage, perform emergency decompression operations in sequence:

[0030] a. First, open the main exhaust valve for primary decompression, and the duration is calculated according to a preset ratio based on the current pressure overlimit value;

[0031] b. If the pressure value is still higher than the dynamic control threshold after primary decompression, activate the auxiliary pressure relief channel to perform secondary pulse decompression;

[0032] c. When it is detected that the pressure drops back to the preset safe range, automatically close all pressure relief valves and generate an event log.

[0033] Further, the method for calculating the matching degree of the historical working condition database includes: extracting the pressure fluctuation spectrum characteristics within the current sliding window, comparing the similarity with the typical working condition feature vectors stored in the database, and dynamically adjusting the learning rate and update frequency of the control parameters according to the comparison results, where the historical working condition data includes three typical modes: normal filling, equipment aging, and sudden leakage.

[0034] Further, the configuration method of the multi-source sensors includes: deploying three pressure sensing nodes equidistantly in the circumferential direction on the inner wall of the gas cylinder, installing a temperature compensation sensor group on the top of the gas cylinder, and setting a bidirectional flow velocity detection device at the outlet pipeline. A cross-check mechanism is adopted among the sensor nodes to eliminate the single-point measurement error.

[0035] Further, the execution logic of the threshold re-optimization instruction includes:

[0036] When the re-optimization instruction is continuously triggered more than the preset number of times, the system self-check program is automatically started:

[0037] a. Calibrate the zero drift of the pressure sensing unit;

[0038] b. Recalculate the reference pressure calibration value at the current ambient temperature;

[0039] c. Reset the control weight matrix of the fuzzy neural network to the initial state.

[0040] The advantages of the present invention are as follows:

[0041] 1. The present invention can dynamically correct the influence of the ambient temperature on the pressure reference value through the constructed pressure-temperature compensation function, eliminate the measurement error caused by temperature drift, calculate the pressure fluctuation entropy value in real time based on the sliding window algorithm, and generate an adaptive threshold adjustment coefficient α by combining with the historical working condition database, so that the control threshold is dynamically adjusted with the gas flow rate fluctuation and the equipment aging degree, reducing the risk of pressure overlimit; through the multi-stage pressure regulation strategy, the fuzzy neural network control is introduced in the critical adjustment stage, and its improved radial basis function dynamic update mechanism can adapt to the pressure mutation mode online, improving the control response speed, and the steady-state error is controlled within ±1.5%.

[0042] 2. Aiming at the problem that the single - stage pressure relief strategy cannot cope with cascading risks, the present invention designs a multi - stage pressure relief interlock mechanism. In the safety release stage, a cooperative strategy of primary pressure relief by the main exhaust valve and secondary pulse - type pressure relief through the auxiliary pressure relief channel is adopted. The pressure relief duration is calculated by the ratio of the pressure over - limit value, and the pressure relief efficiency is increased by 50%, avoiding the pressure rebound caused by single - stage pressure relief. The secondary pulse - type pressure relief adopts intermittent pressure relief control, and dynamically adjusts the pulse frequency through the feedback of the pressure drop gradient, which can reduce the impact force on the pipeline during the pressure relief process and prevent mechanical structure fatigue damage. After the pressure relief is completed, an event log is automatically generated to support post - event traceability analysis and provide data support for system optimization.

[0043] 3. The present invention deploys three pressure - sensing nodes equidistantly in the circumferential direction on the inner wall of the gas cylinder. Abnormal data is eliminated through the variance analysis algorithm. Combining with the top - mounted temperature compensation sensor group and the bidirectional flow velocity detection device, multi - dimensional data fusion is achieved, and the measurement error is reduced to ±0.2 MPa. The cross - verification mechanism of the present invention compares the multi - sensor data in real - time. When the deviation of a single node exceeds 5%, redundant switching is automatically triggered, improving the system fault tolerance rate. Through the threshold re - optimization instruction combined with the time - domain analysis result, the zero - point drift calibration and the reset of the reference pressure calibration value are carried out to ensure the data stability under long - term operation. Detailed implementation manners

[0044] The technical solutions of the present invention will be described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0045] Embodiment 1:

[0046] The present invention provides a multi-stage pressure control method for gas cylinders based on an adaptive threshold, which includes the following steps: Step S1: Real-time collect the pressure value, ambient temperature, and gas flow rate data inside the gas cylinder through multi-source sensors; Step S2: Construct a pressure stage division model, and determine whether the gas cylinder enters the normal filling stage, the critical adjustment stage, or the safe release stage according to the dynamic deviation between the current pressure value and the preset reference threshold; Step S3: Dynamically calculate the pressure fluctuation characteristic value inside the gas cylinder by using a sliding window algorithm, and generate an adaptive threshold adjustment coefficient α based on the matching degree between this characteristic value and the historical working condition database, where 0.8 ≤ α ≤ 1.5; Step S4: Perform a multiplication operation on the reference threshold and the adjustment coefficient α to generate the dynamic control threshold for each stage; Step S5: Activate the corresponding pressure adjustment strategy according to the stage determination result and the dynamic control threshold, including: starting the PID proportional valve linear control in the normal filling stage, starting the fuzzy neural network control in the critical adjustment stage, and triggering the multi-stage decompression interlock mechanism in the safe release stage; Step S6: Perform a time-domain analysis on the execution effect through the pressure state feedback module, and trigger a threshold re-optimization instruction when the pressure change gradient inside the gas cylinder exceeds the set value within 3 consecutive sampling periods. Among them, the multi-source sensors refer to multiple sensors used to collect the pressure value, ambient temperature, and gas flow rate data inside the gas cylinder, and specifically can be implemented by using a pressure sensor, a temperature sensor, and a flow rate sensor. Among them, the pressure stage division model refers to a model for determining the stage of the gas cylinder according to the dynamic deviation between the current pressure value and the preset reference threshold, and specifically can be implemented by using a mathematical algorithm or a logical judgment. Among them, the sliding window algorithm refers to an algorithm for dynamically calculating the pressure fluctuation characteristic value inside the gas cylinder within a certain time window, and specifically can be implemented by using a statistical method or a signal processing technology. Among them, the adaptive threshold adjustment coefficient α refers to a coefficient used to adjust the reference threshold, and specifically can be generated by using machine learning or pattern recognition technology. Among them, the pressure adjustment strategy refers to the pressure control method adopted in different stages, and specifically can be implemented by using PID control, fuzzy neural network control, and multi-stage decompression interlock mechanism.

[0047] The present invention realizes the generation of dynamic control thresholds by collecting multi-source data in real time, constructing a pressure stage division model, and using a sliding window algorithm and a historical working condition database to generate an adaptive threshold adjustment coefficient. According to different stages, corresponding pressure regulation strategies are activated, and real-time monitoring and optimization are carried out through a pressure status feedback module, thereby realizing the precise control and safety management of the pressure of gas cylinders. When the present invention is applied, a pressure sensor is installed inside the cylinder to monitor the internal pressure in real time; a temperature sensor is installed outside the cylinder to collect the ambient temperature; a flow rate sensor is installed at the gas outlet of the gas cylinder to measure the gas flow rate. These sensors transmit the collected data to the control system in real time through a data acquisition module. Next, the control system makes a judgment based on the collected data using a pre-set pressure stage division model. This model divides the state of the gas cylinder into a normal filling stage, a critical adjustment stage, or a safe release stage by calculating the dynamic deviation amount between the current pressure value and the preset reference threshold. This division method can flexibly adjust the control strategy according to the real-time pressure situation.

[0048] Then the system uses a sliding window algorithm to analyze the pressure data within a certain time period. This algorithm calculates the characteristic values of pressure fluctuations, such as the fluctuation amplitude, frequency, etc., within a set time window. These characteristic values are then matched with the data in the historical working condition database to generate an adaptive threshold adjustment coefficient α. The historical working condition database contains the pressure change patterns under various working conditions. By matching with the current situation, the pressure change trend can be predicted more accurately. Based on the generated adaptive threshold adjustment coefficient α, the system multiplies it with the reference threshold to obtain the dynamic control thresholds for each stage. This method enables the control thresholds to be dynamically adjusted according to the actual working conditions, improving the adaptability and control accuracy of the system. According to the determined pressure stage and the calculated dynamic control thresholds, the system activates the corresponding pressure regulation strategies. In the normal filling stage, PID proportional valve linear control is adopted to stabilize the pressure by adjusting the valve opening; in the critical adjustment stage, fuzzy neural network control is started, which can handle situations with strong nonlinearity and uncertainty; in the safe release stage, a multi-stage decompression interlock mechanism is triggered to ensure safety by releasing the pressure step by step and in an orderly manner.

[0049] The present invention continuously monitors the operating status of the system through a pressure status feedback module. When it is detected that the pressure change gradient of the gas cylinder exceeds the set value within three consecutive sampling periods, the system will trigger a threshold re-optimization instruction to adjust the control parameters to adapt to the new working conditions, fully considering the complexity and dynamics of the gas cylinder pressure control. Through multi-level control strategies and adaptive mechanisms, precise, safe, and efficient pressure management is achieved. As a preferred implementation, the present invention can be applied in the following specific scenarios: In an industrial gas supply system, a high-pressure gas cylinder with a volume of 50 L is installed for storing and supplying nitrogen. The system is equipped with the following sensors: a pressure sensor with a range of 0 - 30 MPa, installed inside the gas cylinder; a temperature sensor with a measurement range of -20°C to 80°C, installed on the outer wall of the gas cylinder; a flow rate sensor with a maximum measurement flow rate of 100 L / min, installed on the outlet pipeline. The pressure stage division model is set as follows: When the pressure deviation is within ±5%, it is the normal filling stage; from ±5% to ±15%, it is the critical adjustment stage; when it exceeds ±15%, it is the safety release stage. The sliding window algorithm uses a time window of 60 seconds and samples data once per second. The historical operating condition database stores the operating data of the past six months, including the pressure change patterns at different temperatures and flow rates. The system generates an adaptive threshold adjustment coefficient α by comparing the current pressure fluctuation characteristics with the patterns in the database. For example, when it is detected that the ambient temperature rises and the gas flow rate increases, the α value is adjusted to 1.2 to adapt to the situation of rapid pressure rise. In the normal filling stage, the parameters of the PID controller are set as Kp = 0.5, Ki = 0.1, Kd = 0.05. In the critical adjustment stage, the fuzzy neural network controller uses a 5x5x5 rule base, and the input variables are pressure deviation, deviation change rate, and temperature compensation factor. In the safety release stage, the multi-stage decompression interlock mechanism is set as follows: First, the main exhaust valve is opened for 30 seconds. If the pressure is still too high, the auxiliary pressure relief channel is activated, and secondary decompression is carried out in a pulse mode of opening for 5 seconds and closing for 5 seconds. The pressure status feedback module is configured to trigger a threshold re-optimization instruction when the pressure change gradient exceeds 0.5 MPa / s within three consecutive 10-second sampling periods. The re-optimization process includes recalibrating the sensor zero point, updating the reference pressure value, and resetting the weight matrix of the fuzzy neural network. Through this implementation, the system can dynamically adjust the control strategy according to the real-time operating conditions, effectively cope with the impacts brought by factors such as temperature changes and flow rate fluctuations, and achieve precise and safe gas cylinder pressure control.

[0050] The present invention further proposes a multi-stage pressure control method for gas cylinders based on an adaptive threshold, including the following steps: First, establish a pressure-temperature compensation function f(T) = P base×(1 + 0.023×(T - T0)), where T represents the actual temperature of the current environment; where T0 is the calibration temperature, and where f(T) represents the pressure value inside the gas cylinder after temperature compensation, which is a function of temperature T; where P base represents the base pressure value of the gas cylinder measured or set at the calibration temperature T0, which refers to the pressure value that should be in the cylinder at room temperature; this function takes into account the difference between the actual temperature T of the current environment and the calibration temperature T0, and the pressure value inside the gas cylinder after temperature compensation is represented by f(T). Among them, Pbase represents the base pressure value of the gas cylinder measured or set at the calibration temperature T0, that is, the pressure value that should be in the cylinder at room temperature.

[0051] Next, the normalized deviation is calculated by the formula to quantify the difference between the actual pressure and the reference pressure, providing a basis for subsequent stage determination. The pressure stage is divided according to the absolute value |δ| of the normalized deviation. When |δ| ≤ 5%, it is determined as the normal filling stage; when 5% < |δ| ≤ 15%, it is determined as the critical adjustment stage; when |δ| > 15%, it is determined as the safety release stage. By introducing a pressure-temperature compensation function, the present invention effectively eliminates the influence of ambient temperature changes on pressure measurement. The calculation of the normalized deviation takes into account the relative difference between the actual pressure and the reference pressure, rather than simply the absolute difference. This method can better adapt to gas cylinders of different specifications and improve the universality of pressure control. Combining this multi-stage division method with the adaptive threshold control strategy of the present invention can achieve more precise pressure control. In the normal filling stage, conventional methods such as PID control can be adopted by the system; after entering the critical adjustment stage, the fuzzy neural network control can be started to improve the response speed and adaptability of the system; while in the safety release stage, a multi-stage decompression interlock mechanism is triggered to quickly reduce the system pressure.

[0052] During specific implementation, a timer can be set to perform pressure stage determination every certain period (such as 1 second). First, the current temperature T and pressure P are read, and then f(T) is calculated through the temperature compensation function. Then the normalized deviation δ is calculated, and the current pressure stage is judged according to the value of |δ|. The judgment result can be displayed through a status indicator light or sent to the control system for corresponding pressure adjustment. Compared with the traditional fixed threshold method, the pressure stage division model of the present invention has the following advantages:

[0053] 1. Temperature compensation: By introducing a temperature compensation function, the influence of ambient temperature changes on pressure judgment is effectively eliminated, improving the adaptability of the system under different temperature conditions.

[0054] 2. Dynamic Threshold: The use of the normalized deviation allows the system to dynamically adjust the judgment criteria according to the current reference pressure, which is applicable to gas cylinders of different specifications and filling pressures.

[0055] 3. Fine-grained Control: The three-level pressure stage division provides a finer control granularity, enabling the system to adopt different control strategies according to the degree of pressure deviation, ensuring both safety and improving control accuracy.

[0056] 4. Scalability: This model is easy to integrate with other control algorithms (such as PID, fuzzy neural network, etc.), providing a basis for implementing complex multi-stage pressure control.

[0057] Through this improved pressure stage division method, the present invention can more accurately identify the working state of gas cylinders, providing a reliable decision-making basis for subsequent pressure control, thereby improving the safety and efficiency of the entire system.

[0058] The present invention further proposes a technical solution for determining the dynamic window size of the sliding window algorithm. This technical solution dynamically adjusts the size of the sliding window to adapt to the pressure fluctuation characteristics under different working conditions, improving the calculation accuracy and flexibility of the pressure fluctuation characteristic values. Specifically, the implementation steps of the sliding window algorithm include: First, calculate the minimum analysis period according to the cylinder volume V and the current gas flow rate v of the gas flowing in or out of the cylinder. Secondly, set the window width where Δt is the sampling interval time, and then calculate the pressure fluctuation entropy value H=-∑(p i ·log2(p i )) within the window. Where H is the pressure fluctuation entropy value, which is used to quantify the uncertainty of pressure fluctuations, and p i is the probability of a certain pressure value appearing within the window. The calculation of the entropy value can effectively characterize the complexity of pressure fluctuations. The design of this dynamic window size is associated with the cylinder volume and gas flow rate, enabling the sliding window algorithm to better adapt to different working conditions. For example, when the gas flow rate is fast, the window size can be reduced to capture the rapidly changing pressure characteristics; when the flow rate is slow, the window size can be appropriately increased to obtain more stable pressure fluctuation characteristics. By adopting a dynamic window size, the sliding window algorithm of the present invention can calculate the pressure fluctuation characteristic values more accurately. This method not only improves the accuracy of pressure control but also enhances the adaptability of the system to different working conditions. For example, during the rapid filling or release of gas, a smaller window size can promptly capture the sharp changes in pressure, enabling the control system to respond quickly; while in the stable state, a larger window size can smooth short-term fluctuations and provide more stable pressure characteristic values.

[0059] In specific implementation, an initial window width W and 100 sampling points are set. Then, according to the current gas flow rate v and the cylinder volume V, the window width is dynamically adjusted. For example, when the v / V ratio increases, the window width can be reduced to 50 sampling points; when the v / V ratio decreases, the window width can be increased to 150 sampling points. This can ensure that the most suitable pressure fluctuation characteristic values can be obtained under different working conditions. When calculating the pressure fluctuation entropy value, the discretization method can be used to divide the continuous pressure values into several intervals. For example, the pressure range is equally divided into 10 intervals, and then the frequency of the pressure values appearing in each interval is counted as the probability p i . In this way, an entropy value H reflecting the pressure distribution characteristics can be obtained. Compared with the prior art, the dynamic window size sliding window algorithm of the present invention has obvious advantages. The traditional method usually uses a fixed window size and is difficult to adapt to the pressure change characteristics under different working conditions. However, the method of the present invention can better capture the characteristics of pressure fluctuation by dynamically adjusting the window size, improving the accuracy of pressure control and the adaptability of the system. This improvement is of great significance for industrial applications that require precise pressure control, such as precision chemical production or medical gas supply systems.

[0060] The present invention further proposes a technical solution for starting fuzzy neural network control in the critical adjustment stage. Specifically, a five-layer network structure is constructed, including an input layer, a membership degree generation layer, a rule inference layer, a feature fusion layer, and an output defuzzification layer. The input layer includes the pressure deviation, the deviation change rate, and the temperature compensation factor. The feature fusion layer adopts an improved radial basis function, and its center vector is dynamically updated through an online clustering algorithm. The output layer generates the valve opening correction amount. By constructing a multi-layer neural network structure, the present invention realizes the adaptive control of a complex nonlinear system. The input layer introduces the pressure deviation, the deviation change rate, and the temperature compensation factor, comprehensively considering the dynamic characteristics of the system and the environmental impact. The membership degree generation layer and the rule inference layer integrate the advantages of fuzzy logic and can handle the uncertainties in the system. The feature fusion layer adopts an improved radial basis function, and its center vector is dynamically updated through an online clustering algorithm, improving the adaptability of the network to system changes. The output layer generates the valve opening correction amount, which directly acts on the actuator to achieve precise control. Specifically, in implementation, the input layer receives the data of the pressure deviation, the deviation change rate, and the temperature compensation factor collected in real time. The membership degree generation layer performs fuzzy processing on the input variables, converting the accurate numerical values into fuzzy sets. The rule inference layer performs inference based on a preset fuzzy rule base to generate a preliminary control decision. The feature fusion layer adopts an improved radial basis function for feature extraction and fusion, and its center vector is dynamically updated through an online clustering algorithm to adapt to the changes in the system state. The output defuzzification layer converts the fuzzy inference result into an accurate valve opening correction amount. The application of this fuzzy neural network control method in the critical adjustment stage significantly improves the accuracy and stability of pressure control. By introducing the temperature compensation factor, the system can better adapt to the changes in the environmental temperature. The dynamic update mechanism of the improved radial basis function enables the control strategy to be adjusted in real time to adapt to complex working conditions such as gas flow rate fluctuations and equipment aging. Compared with the traditional PID control, this method shows stronger robustness and self-adaptability in nonlinear systems. As a preferred implementation manner, the number of neurons in the input layer can be set to 3, corresponding to the pressure deviation, the deviation change rate, and the temperature compensation factor. The membership degree generation layer can adopt a Gaussian membership degree function, and each input variable is divided into 5 fuzzy subsets. The rule inference layer constructs 125 fuzzy rules based on expert knowledge. The feature fusion layer adopts 20 radial basis function neurons, and the initial center vector is determined through the K-means clustering algorithm. The output layer adopts a single neuron to directly output the valve opening correction amount. In practical applications, this fuzzy neural network control system can be realized through the following steps: 1. System initialization: Load the pre-trained network parameters, including the membership degree function parameters, the fuzzy rule weights, and the radial basis function center vector. 2. Data acquisition: Obtain the pressure deviation, the deviation change rate, and the temperature data in real time. 3. Input layer processing: Normalize the collected data. 4. Membership degree calculation: Calculate the membership degree of the input variables using the Gaussian membership degree function.5. Fuzzy inference: Perform inference calculations based on the activated fuzzy rules. 6. Feature fusion: Use the improved radial basis function for feature extraction and fusion. 7. Output generation: Calculate and output the valve opening correction amount. 8. Online learning: According to the control effect, use the backpropagation algorithm to adjust the network parameters. 9. Center vector update: Regularly execute the online clustering algorithm to update the center vector of the radial basis function. Compared with the traditional PID control, this method can still maintain stability when facing parameter perturbations of ±20%, demonstrating excellent robustness. In addition, through the online learning and dynamic update mechanism, the performance decay of the system during long-term operation is reduced by 60%, greatly improving the reliability and lifespan of the control system.

[0061] The present invention further proposes a dynamic update mechanism for the improved radial basis function, including automatically merging similar feature clustering centers according to the pressure change trend data collected in real time, and when detecting a sudden change in the pressure fluctuation pattern, creating a new clustering node and freezing the working state of the historical low-frequency nodes.

[0062] The dynamic update mechanism of the improved radial basis function is mainly applied to the fuzzy neural network control in the critical adjustment stage. Specifically, the improved radial basis function is adopted in the feature fusion layer of the fuzzy neural network, and its center vector is dynamically updated through the online clustering algorithm. This dynamic update mechanism can make the network structure adjust in real time according to the pressure change trend, improving the response speed and adaptability of the system to sudden pressure changes.

[0063] The dynamic update mechanism of the improved radial basis function includes the following key steps: First, perform similarity analysis on the existing feature clustering centers according to the pressure change trend data collected in real time. If it is found that the pressure characteristics represented by multiple clustering centers are highly similar, these clustering centers will be automatically merged. The merging process can adopt the weighted average method to retain the main feature information. This merging operation can streamline the network structure and improve the calculation efficiency. Second, monitor the pressure fluctuation pattern in real time. When detecting a sudden change in the pressure fluctuation pattern, such as the pressure change rate or amplitude exceeding the normal range, the system will automatically create a new clustering node to represent this new pressure feature. The initial parameters of the new node can be set based on the current pressure data and then continuously optimized through online learning. At the same time, the system will freeze the clustering nodes that are used less frequently in history. Specifically, a usage frequency threshold can be set, and when the activation times of a certain clustering node are lower than this threshold within a certain time window, it will be temporarily frozen. In the frozen state, this node does not participate in the regular calculation but still retains its parameter information. This mechanism can reduce redundant calculations while retaining historical experience.

[0064] Through the above dynamic update mechanism, the improved radial basis function can achieve the adaptive adjustment of the clustering center. When the pressure change is stable, the network structure is simplified by merging similar features; when the pressure changes suddenly, new nodes are quickly created to capture new features; at the same time, the network scale is balanced by freezing low-frequency nodes. This dynamic balance mechanism enables the fuzzy neural network to quickly adapt to various pressure change patterns while maintaining computational efficiency. Further, this dynamic update mechanism closely cooperates with other levels of the fuzzy neural network. For example, the pressure deviation, deviation change rate, and temperature compensation factor in the input layer provide basic data for dynamic update. The rule inference layer can dynamically adjust the fuzzy rules according to the updated feature distribution. The output defuzzification layer generates a more accurate valve opening correction amount based on the optimized network structure. During specific implementation, the trigger conditions and frequencies of dynamic update can be set. For example, a similarity analysis and merging operation of the clustering center are performed every 100 sampling periods; when the pressure change rate exceeds the preset threshold for 10 consecutive sampling periods, the creation of new clustering nodes is triggered; nodes with activation times lower than 1% of the total sample number within 1000 consecutive sampling periods are frozen. These parameters can be flexibly adjusted according to the actual application scenario. As a preferred implementation method, historical data analysis can be combined to set differentiated dynamic update strategies for different pressure intervals. For example, when approaching the critical pressure interval, the sensitivity of new node creation is increased to more precisely depict the pressure change characteristics; while in the normal pressure interval, the frequency of clustering center merging can be appropriately increased to optimize the network structure. By adopting the dynamic update mechanism of the improved radial basis function, the fuzzy neural network control system of the present invention can better adapt to complex and changeable pressure environments. Compared with traditional neural networks with fixed structures, this mechanism can adjust the network structure in real time according to the pressure change trend, significantly improving the response speed and adaptability of the system to sudden pressure changes.

[0065] The present invention further proposes a technical solution for triggering a multi-stage pressure reduction interlock mechanism during the safety release stage. Specifically, when the gas cylinder enters the safety release stage, an emergency pressure reduction operation is sequentially performed: First, the main exhaust valve is preferentially opened for primary pressure reduction, and the duration is calculated according to a preset ratio based on the current pressure overlimit value; Second, if the pressure value after primary pressure reduction is still higher than the dynamic control threshold, the auxiliary pressure relief channel is activated to perform secondary pulsed pressure reduction; Finally, when it is detected that the pressure drops back to the preset safe range, all pressure relief valves are automatically closed and an event log is generated. The multi-stage pressure reduction interlock mechanism proposed by the present invention mainly solves the problem in the prior art that the single-stage pressure reduction strategy cannot effectively cope with the chain risk caused by sudden pressure changes. By adopting the cooperative strategy of primary pressure reduction by the main exhaust valve and secondary pulsed pressure reduction by the auxiliary pressure relief channel, the pressure reduction process of the gas cylinder can be controlled more flexibly and precisely.

[0066] Specifically, the multi-stage pressure reduction interlock mechanism of the present invention includes the following key features: First, the primary pressure reduction is carried out by the main exhaust valve. The main exhaust valve usually has a relatively large diameter and can rapidly reduce the pressure inside the cylinder within a short period of time. The calculation method of the duration can adopt a linear or non-linear function. For example, it can be set as: Duration = k * (Current Pressure - Dynamic Control Threshold), where k is a preset coefficient that can be adjusted according to the actual situation. Secondly, the secondary pressure reduction adopts a pulsed pressure reduction method. This method can more precisely control the pressure drop rate and avoid the impact on the system caused by a sudden pressure drop. The pulse frequency can be dynamically adjusted according to the pressure drop gradient. For example, it can be set as: Pulse Frequency = f0 * exp(-β * (Current Pressure - Target Pressure)), where f0 is the initial frequency and β is the attenuation coefficient. The system will continuously monitor the pressure value. When the pressure drops back to the preset safe range, all pressure relief valves will be automatically closed. This safe range can be set to 80% - 90% of the dynamic control threshold to leave a certain buffer space. At the same time, the system will generate a detailed event log to record the changes of key parameters during the entire pressure reduction process. These features are closely related to form a complete pressure reduction control process. The primary pressure reduction provides the ability to rapidly reduce pressure, and the secondary pressure reduction provides the ability for fine adjustment. The combination of the two can achieve efficient pressure reduction while ensuring safety. The real-time monitoring and automatic closing mechanism ensure the controllability and safety of the entire process.

[0067] In practical applications, the multi-stage pressure reduction interlock mechanism of the present invention can be implemented as follows: When the system detects that the pressure of the gas cylinder exceeds the safety threshold, the main exhaust valve is first activated for primary pressure reduction. Assume the current pressure is 25 MPa, the dynamic control threshold is 20 MPa, and the preset coefficient k is 0.5 s / MPa. Then the duration of the primary pressure reduction is: 0.5 * (25 - 20) = 2.5 s. If the pressure is still higher than the dynamic control threshold after 2.5 s, the system will activate the secondary pulsed pressure reduction. Set the initial pulse frequency f0 as 2 Hz, the attenuation coefficient β as 0.1, and the target pressure as 18 MPa. As the pressure drops, the pulse frequency will gradually decrease. For example, when the pressure drops to 22 MPa, the pulse frequency becomes: 2 * exp(-0.1 * (22 - 18)) ≈ 1.6 Hz. During the entire pressure reduction process, the system will continuously monitor the pressure change. When the pressure drops to 18 MPa (assuming the set safe range is 90% of the dynamic control threshold), the system automatically closes all pressure relief valves and generates a detailed event log containing information such as the pressure change curve and the valve opening and closing status.

[0068] By adopting this multi-stage pressure reduction interlock mechanism, the present invention can significantly improve the control accuracy and efficiency of gas cylinders during the safe release stage. Compared with the traditional single-stage pressure reduction method, the solution of the present invention has the following advantages:

[0069] 1. Improved pressure relief efficiency: Through the cooperation of rapid pressure reduction by the main exhaust valve and fine adjustment of the auxiliary pressure relief channel, the pressure can be more quickly reduced to the target range while ensuring safety.

[0070] 2. Reduced mechanical shock: The two-stage pulsed pressure relief can effectively reduce the impact of sudden pressure drop on pipelines and valves, extending the service life of the equipment.

[0071] 3. Improved control accuracy: By real-time monitoring and dynamically adjusting the pulse frequency, the pressure drop rate can be more precisely controlled, avoiding the phenomenon of pressure rebound.

[0072] 4. Enhanced system reliability: The multi-stage pressure relief strategy provides redundant protection. Even if a single pressure relief channel fails, the safety of the system can still be guaranteed. Compared with the prior art, the multi-stage pressure relief interlock mechanism of the present invention shows obvious advantages in dealing with pressure mutation situations. Traditional single-stage pressure relief or simple interlock mechanisms often have difficulty coping with complex pressure change situations, easily leading to problems such as insufficient pressure relief or over-pressure relief. The present invention can better adapt to different pressure mutation situations by combining rapid pressure relief and fine adjustment, improving the adaptability and reliability of the system.

[0073] The matching degree calculation method of the historical working condition database in the present invention includes: extracting the pressure fluctuation frequency spectrum characteristics within the current sliding window, comparing the similarity with the typical working condition feature vectors stored in the database, and dynamically adjusting the learning rate and update frequency of the control parameters according to the comparison result. The historical working condition data includes three typical modes: normal filling, equipment aging, and sudden leakage. By extracting the pressure fluctuation frequency spectrum characteristics and comparing the similarity with the typical working condition feature vectors, the current working condition can be more accurately identified. Dynamically adjusting the control parameters according to the comparison result can achieve more refined adaptive control. The historical working condition database containing three typical modes can cover the main scenarios during the use of gas cylinders, improving the matching accuracy. Specifically, the matching degree calculation method of the historical working condition database of the present invention includes the following steps: First, use the fast Fourier transform (FFT) algorithm to extract the pressure fluctuation frequency spectrum characteristics within the current sliding window. By performing FFT transformation on the pressure time-domain data, the frequency-domain feature vector can be obtained.

[0074] Secondly, the cosine similarity algorithm is used to calculate the similarity between the extracted spectral features and the characteristic vectors of typical working conditions stored in the database. According to the similarity comparison results, the learning rate and update frequency of the control parameters are dynamically adjusted. When the similarity is high, the learning rate and update frequency are increased to quickly adapt to the current working condition; when the similarity is low, the learning rate and update frequency are decreased to maintain control stability. The historical working condition database contains characteristic vectors of three typical modes: normal filling, equipment aging, and sudden leakage. In the normal filling mode, the pressure changes smoothly; in the equipment aging mode, the pressure fluctuation frequency increases; in the sudden leakage mode, the pressure drops sharply. By matching with these typical modes, the current working condition can be accurately identified and corresponding control strategies can be adopted. For example, in the conventional filling stage, the parameters of the PID controller can be adjusted according to the matching degree with the normal filling mode; in the critical adjustment stage, the learning strategy of the fuzzy neural network can be adjusted according to the matching degree with the equipment aging mode; in the safety release stage, the triggering conditions of the multi-stage decompression interlock mechanism can be optimized according to the matching degree with the sudden leakage mode.

[0075] In practical applications, the method for calculating the matching degree of the historical working condition database of the present invention can be integrated into the embedded processor of the gas cylinder pressure control system. The processor performs feature extraction and matching calculations every certain time interval (such as 1 second) and transmits the results to the control algorithm module to achieve real-time adaptive control. The present invention can more accurately identify the current working condition of the gas cylinder and dynamically adjust the control strategy according to the working condition characteristics. Compared with the traditional fixed-threshold control, this method can improve the pressure control accuracy by about 30% and shorten the identification time of abnormal working conditions by more than 50%. It not only improves the safety of gas cylinder use but also optimizes the filling efficiency and gas utilization rate.

[0076] The present invention further proposes an execution logic for the threshold re-optimization instruction, including automatically starting the system self-check program when the continuous triggering of the re-optimization instruction exceeds a preset number of times. The system self-check program includes zero-drift calibration of the pressure sensing unit, recalculating the reference pressure calibration value at the current ambient temperature, and resetting the control weight matrix of the fuzzy neural network to its initial state. The execution logic of the threshold re-optimization instruction proposed by the present invention aims to solve the problems of system drift and performance degradation that may occur during long-term operation. By setting the triggering threshold, frequent execution of the self-check program is avoided, which may affect the normal operation of the system. The three steps of the self-check program are respectively for calibrating and resetting the hardware, environmental factors, and control algorithm to ensure the long-term stability and accuracy of the system. Specifically, the execution logic of the threshold re-optimization instruction includes the following steps: First, the system continuously monitors the triggering frequency of the re-optimization instruction. When it is detected that the number of consecutive triggering of the re-optimization instruction exceeds the preset threshold, the system self-check program is automatically started. The preset threshold can be adjusted according to the actual application scenario, for example, it can be set to 3 times or 5 times. The system self-check program first performs zero-drift calibration on the pressure sensing unit. The calibration process can adopt the multi-point calibration method, measuring and correcting at different pressure points to improve the accuracy of calibration. Then the system recalculates the reference pressure calibration value at the current ambient temperature, dynamically updating the reference pressure value through real-time temperature data and the pressure-temperature compensation function. Finally, the system resets the control weight matrix of the fuzzy neural network to its initial state, enabling the control algorithm to adapt to the current working conditions again. The reset process includes restoring the membership function parameters, rule weights, etc. to the optimal initial values obtained through offline training. By executing the above self-check program, the present invention can effectively address the system drift problem during long-term operation. The zero-drift calibration of the pressure sensing unit ensures the accuracy of the measurement data, the recalculation of the reference pressure calibration value adapts to the ambient temperature change, and the reset of the fuzzy neural network control weights guarantees the adaptability and robustness of the control algorithm.

[0077] As a preferred implementation, a counter can be set in the system to record the number of consecutive triggering of the re-optimization instruction. When the counter reaches the preset threshold (e.g., 5 times), the self-check program is automatically triggered. After the self-check program is completed, the counter is automatically cleared. In addition, after each successful execution of the pressure control cycle, the counter can be decremented by 1 to avoid unnecessary self-checks caused by accidental fluctuations. The specific implementation example is as follows: Suppose the pressure control method of the present invention is used in a gas cylinder filling station. The system sets the threshold for consecutive triggering of the re-optimization instruction to 5 times. During a long-term operation, the system detects the following situations:

[0078] Day 1: The system operates normally without triggering any re-optimization instructions. Day 2: One re-optimization instruction is triggered, and the counter value is 1. Day 3: Two re-optimization instructions are triggered, and the counter value increases to 3. Day 4: One pressure control cycle is successfully executed, and the counter value decreases to 2. Day 5: Three re-optimization instructions are continuously triggered, and the counter value increases to 5. At this time, the system automatically starts a self-check program:

[0079] 1. Perform zero-drift calibration on the pressure sensing unit: Use a standard pressure source to calibrate at three points: 0 MPa, 5 MPa, and 10 MPa, and correct the sensor output curve.

[0080] 2. Recalculate the reference pressure calibration value: The current ambient temperature is 28°C, while the initial calibration temperature is 20°C. Assuming the initial calibration pressure is 10 MPa, the new reference pressure is calculated as follows:

[0081] P2 = 10 MPa * (301.15 K / 293.15 K) ≈ 10.27 MPa

[0082] 3. Reset the weights of the fuzzy neural network control: Restore the weight matrix in the five-layer network structure to the optimal initial value after offline training.

[0083] After the self-check program is completed, the counter is cleared, and the system resumes normal operation. In this way, the present invention can maintain the stability and accuracy of the system during long-term operation, effectively solving the problem that the execution logic of the threshold re-optimization instruction is not perfect enough.

[0084] Compared with the simple fixed-cycle calibration or manual intervention methods in the prior art, the execution logic of the threshold re-optimization instruction of the present invention dynamically determines whether to perform self-check according to the actual operation status of the system, avoiding unnecessary frequent calibrations. The self-check program covers three aspects: hardware, environmental factors, and control algorithms, ensuring the overall optimization of the system. The entire process does not require manual intervention, improving the autonomy and reliability of the system. Moreover, by dynamically updating the reference pressure calibration value, the system can more accurately adapt to environmental changes and improve the accuracy of pressure control.

[0085] Finally, it should be noted that: The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: They can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-stage pressure control method for gas cylinders based on an adaptive threshold, characterized in that, It includes the following steps: Step S1: Real-time collect the pressure value, ambient temperature, and gas flow rate data inside the gas cylinder through multi-source sensors; Step S2: Construct a pressure stage division model, and determine whether the gas cylinder enters the normal filling stage, critical adjustment stage, or safety release stage according to the dynamic deviation amount between the current pressure value and the preset reference threshold; Step S3: Dynamically calculate the pressure fluctuation characteristic value inside the gas cylinder using the sliding window algorithm, and generate an adaptive threshold adjustment coefficient α based on the matching degree between this characteristic value and the historical working condition database; Step S4: Perform a multiplication operation on the reference threshold and the adjustment coefficient α to generate the dynamic control threshold for each stage; Step S5: Activate the corresponding pressure regulation strategy according to the stage determination result and the dynamic control threshold, including: starting the PID proportional valve linear control in the normal filling stage, starting the fuzzy neural network control in the critical adjustment stage, and triggering the multi-stage decompression interlock mechanism in the safety release stage; Step S6: Perform a time-domain analysis on the execution effect through the pressure state feedback module. When the pressure change gradient inside the gas cylinder exceeds the set value within 3 consecutive sampling periods, trigger the threshold re-optimization instruction.

2. The multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that The specific implementation steps of the pressure stage division model in Step S2 include: S21: Establish the pressure-temperature compensation function f(T) = P base ×(1 + 0.023×(T - T0)); where T represents the actual temperature of the current environment; where T0 is the calibrated temperature, and f(T) represents the pressure value inside the gas cylinder after temperature compensation, which is a function of temperature T; where P base represents the base pressure value of the gas cylinder measured or set at the calibration temperature T0, which refers to the pressure value that should be in the cylinder at room temperature; S22: Calculate the normalized deviation amount S23: When |δ| ≤ 5%, it is determined as the normal filling stage. When 5% < |δ| ≤ 15%, it is determined as the critical adjustment stage. When |δ| > 15%, it is determined as the safety release stage.

3. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that, The sliding window algorithm uses a dynamic window size determination, and its specific steps include: S31: Calculate the minimum analysis period according to the volume V of the cylinder and the current speed v at which the gas flows out of or into the cylinder. S32: Set the window width where Δt is the sampling interval time; S33: Calculate the pressure fluctuation entropy value H = -∑(p i ·log2(p i )) within the window, where H is the pressure fluctuation entropy value used to quantify the uncertainty of pressure fluctuation, and p i is the probability of a certain pressure value occurring within the window.

4. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that, Starting the fuzzy neural network control in the critical adjustment stage includes constructing a five-layer network structure, including an input layer, a membership degree generation layer, a rule inference layer, a feature fusion layer, and an output defuzzification layer; the input layer includes the pressure deviation, deviation change rate, and temperature compensation factor, and the feature fusion layer uses an improved radial basis function, and its center vector is dynamically updated through an online clustering algorithm; the output layer generates the valve opening correction amount.

5. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 4, characterized in that, The dynamic update mechanism of the improved radial basis function includes: automatically merging similar feature clustering centers according to the real-time collected pressure change trend data. When a sudden change in the pressure fluctuation pattern is detected, create a new clustering node and freeze the working state of the historical low-frequency nodes.

6. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that Triggering the multi-stage decompression interlock mechanism in the safety release stage includes: When the gas cylinder enters the safety release stage, perform emergency decompression operations in sequence: a. First, open the main exhaust valve for primary decompression, and the duration is calculated according to a preset ratio based on the current pressure overlimit value; b. If the pressure value after primary decompression is still higher than the dynamic control threshold, activate the auxiliary pressure relief channel to perform secondary pulse decompression; c. When it is detected that the pressure drops back to the preset safe range, automatically close all pressure relief valves and generate an event log.

7. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that, The matching degree calculation method of the historical working condition database includes: extracting the pressure fluctuation spectrum characteristics within the current sliding window, comparing the similarity with the typical working condition feature vectors stored in the database, and dynamically adjusting the learning rate and update frequency of the control parameters according to the comparison results, where the historical working condition data includes three typical modes: normal filling, equipment aging, and sudden leakage.

8. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that, The configuration method of the multi-source sensors includes: deploying three pressure sensing nodes equidistantly in the circumferential direction on the inner wall of the gas cylinder, installing a temperature compensation sensor group on the top of the gas cylinder, and setting a bidirectional flow velocity detection device at the outlet pipeline. A cross-check mechanism is adopted between the sensor nodes to eliminate the single-point measurement error.

9. A multi-stage pressure control method for gas cylinders based on an adaptive threshold according to claim 1, characterized in that, The execution logic of the threshold re-optimization instruction includes: When the re-optimization instruction is continuously triggered more than the preset number of times, the system self-check program is automatically started: a. Calibrate the zero drift of the pressure sensing unit; b. Recalculate the reference pressure calibration value at the current ambient temperature; c. Reset the control weight matrix of the fuzzy neural network to the initial state.

Images