Intelligent control method of waste gas recovery system

Abstract

The invention relates to the technical field of industrial waste gas purification and resource recovery, in particular to an intelligent control method of a waste gas recovery system. Step 2, constructing a distributed communication architecture; 3, collecting and fusing data in real time; 4, constructing and operating a digital twinborn model; step 5, intelligently optimizing and generating a control instruction; step 6, performing security arbitration and instruction execution; step 7, performing monitoring execution and closed loop feedback; step 8, model and strategy self-learning updating; by constructing a'general control-branch control 'distributed communication architecture and combining double-link redundant communication and a local security arbitration mechanism, the risk of centralized control is effectively dispersed, the robustness of the system under the condition of local faults or network interruption is enhanced, control areas are conveniently and flexibly increased or adjusted according to technological process changes, and the system reliability is improved. And the modularization level and the engineering adaptability of the system are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial waste gas purification and resource recovery, and particularly relates to an intelligent control method of a waste gas recovery system. BACKGROUND

[0002] If the waste gas generated in the industrial production process is directly discharged without effective treatment, not only environmental pollution will be caused, but also the waste of valuable components in the waste gas will be caused. Therefore, the waste gas recovery system is very important in many industries such as chemical industry, spraying and semiconductor. The traditional waste gas recovery system control mainly adopts a centralized control strategy based on fixed rules or simple PID regulation, and the response speed, processing efficiency and self-adaptive ability of the control strategy are often difficult to cope with complex and changeable working conditions.

[0003] For example, a control method, device and system for waste gas collection with the application number CN201210418013.1 and the authorization announcement date of 20141112, the main content of which includes: receiving the state parameters of the waste gas in the current feeding bin detected by the detection device, determining the waste gas collection parameters corresponding to the current waste gas state parameters according to the corresponding relationship between the waste gas state parameters and the waste gas collection parameters, and collecting and controlling the waste gas in the feeding bin by using the determined waste gas collection parameters. In this way, the speed of waste gas collection and the size of the opening degree of the air suction valve can be adjusted in real time according to the waste gas collection parameters determined according to the current waste gas state parameters, so that the waste gas can be timely, effectively and quickly transferred to the waste gas treatment tower when there is too much waste gas in the feeding bin, avoiding the problem of waste gas overflow, and the load of the waste gas collection equipment can be reduced and energy consumption can be saved when there is relatively less waste gas in the feeding bin, with appropriate frequency and appropriate opening degree of the air suction valve.

[0004] The waste gas recovery control method of the prior art adopts a centralized architecture, has poor scalability and high risk, and the decision depends on static rules, lacking multi-objective dynamic optimization capability. In addition, the safety mechanism is weak, the abnormal switching is not smooth, lacks high-fidelity digital twin model support, cannot realize predictive simulation and verification, and the system does not have self-learning ability, making it difficult to maintain optimal performance for a long time. Therefore, it is urgent to design an intelligent control method of a waste gas recovery system to solve the above problems. SUMMARY

[0005] The purpose of the present application is to provide an intelligent control method of a waste gas recovery system to solve the above problems in the prior art.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme:

[0007] An intelligent control method of a waste gas recovery system, comprising the following steps:

[0008] Step 1. Arrange sensors and actuators: Deploy sensor networks at key nodes throughout the system, connect sensors, variable frequency fans, control valves, and other waste gas recovery system actuators to their corresponding regional control nodes.

[0009] It should be noted that:

[0010] The sensor network deployed throughout the waste gas treatment process includes at least PID / FID concentration meters or NDIR concentration meters for monitoring waste gas composition and concentration, thermal flow meters or ultrasonic flow meters for monitoring flow, temperature and humidity sensors, pressure transmitters, vibration sensors and current sensors for monitoring equipment status, and energy consumption metering modules.

[0011] The actuators include variable frequency fans, electric control valves, pumps, and heaters, which are connected to the corresponding regional control nodes through field bus or hardwiring.

[0012] Step 2. Build a distributed communication architecture: Build a hierarchical control architecture consisting of one master node and multiple regional control nodes, communicate between nodes through redundant industrial network to ensure reliable data transmission.

[0013] It should be noted that:

[0014] The master node and each regional control node communicate redundantly through dual links of industrial Ethernet and 5G / 4G wireless network, and are configured with automatic switching mechanism and local instruction cache function in case of communication interruption. During communication interruption, each control node can maintain system safe operation according to the last valid instruction and local guaranteed control logic.

[0015] Step 3. Real-time data acquisition and fusion: Each control node periodically acquires sensor data, filters and processes abnormal data, and generates a unified system operating state vector by the master node.

[0016] It should be noted that:

[0017] Each control node acquires data at a cycle of 1-10 seconds and uses Kalman filter algorithm for noise elimination.

[0018] Abnormal data beyond the reasonable range is marked and excluded, and backup sensor data or historical time series data is used for interpolation completion.

[0019] The master node uses time alignment algorithm to unify the time stamp of the collected data, and generates a standardized operating data set representing the global operating state of the system (including waste gas process parameters, equipment state parameters, and environmental parameters).

[0020] Step four. Construct and run the digital twin model: the master node constructs a digital twin model based on a hybrid model of mechanism and data driving, and inputs the working condition data set into the model, so that the digital twin model runs synchronously with the physical system;

[0021] It should be noted that:

[0022] The constructed digital twin model is a hybrid model that combines a mechanism model based on the quality and energy conservation law of the waste gas recovery process and a data-driven model trained based on historical operation data;

[0023] The data-driven model is constructed using a least squares support vector machine (LSSVM) or neural network algorithm to correct the prediction bias of the mechanism model;

[0024] The simulation rehearsal is used to predict the change trend of key indicators such as adsorption saturation, outlet pollutant concentration and system comprehensive energy consumption in the next 15-30 minutes.

[0025] Step five. Intelligent optimization to generate control instructions: the master node uses the synchronized digital twin model to generate a global optimization instruction set containing the process parameter set values of each actuator, with the highest waste gas recovery rate, the lowest system energy consumption and equipment loss as the comprehensive optimization objective, under the preset safety and environmental protection constraints;

[0026] It should be noted that:

[0027] The online optimization solution adopts a multi-objective optimization algorithm with constraints to generate a global optimization instruction set by solving the following comprehensive objective function online:

[0028] Maximize J = α·η - β·E - γ·D;

[0029] Wherein, η is the real-time waste gas recovery efficiency, E is the system comprehensive energy consumption, D is the key equipment loss index, and α, β, γ are self-learning adjustment weight coefficients, and all are positive numbers;

[0030] The constraint conditions for solving include: device safe operation limit value, environmental protection emission standard, and physical feasible region of decision variables such as main recovery unit switching period, regeneration unit working parameter, fan speed and valve opening.

[0031] Step six. Safety arbitration and instruction execution: after receiving the instructions, the sub-control node calls its locally stored physical safety rule set for independent verification, and executes if it passes, and refuses to execute the instruction and automatically switches to the preset bottom safety mode if the verification fails, and simultaneously alarms the master control;

[0032] It should be noted that:

[0033] The local physical safety rule set at least includes:

[0034] The area material mass conservation verification rule, the equipment mechanical action limit rule, the process parameter safety threshold rule, and the actuator action rate limit rule;

[0035] The result of the safety arbitration is divided into three categories: verification pass, verification warning, and verification failure, wherein:

[0036] When the verification fails, the sub-control node refuses to execute the global optimization instruction and automatically switches to the local stored bottom safety mode operation, and sends an alarm to the master control node;

[0037] When the verification warning, the sub-control node sends a warning information to the master control node while executing the instruction, and strengthens the state monitoring of the region;

[0038] In step six:

[0039] When the system runs in the bottom safety mode due to safety arbitration failure, if it needs to restore the optimization operation, the recovery step is executed, as shown below:

[0040] The master control node performs safety simulation on the recovery process from the current state to the optimization target state based on the digital twin model, and automatically issues a gradual control right return instruction sequence to the sub-control node after verification.

[0041] Step 7. Monitoring execution and closed-loop feedback: After execution, compare the deviation between the actual response and the model prediction, and if it is within the specified range, maintain it, otherwise, perform local fine-tuning or generate a new instruction by the master control, forming a closed loop.

[0042] It should be noted that:

[0043] The comparison result of the deviation between the actual response and the model prediction is classified according to the preset deviation threshold:

[0044] When the deviation value is less than or equal to the first threshold value, the current control strategy is maintained;

[0045] When the deviation value is greater than the first threshold value and less than or equal to the second threshold value, the local parameter fine-tuning algorithm is started by the corresponding sub-control node for compensation control;

[0046] When the deviation value is greater than the second threshold value or a sudden abnormality is detected, the master control node is triggered to re-execute step five to generate a correction instruction, or to start fault tracing analysis, wherein (the first threshold value is 3%, and the second threshold value is 10%);

[0047] When fault tracing analysis:

[0048] When the execution deviation is seriously out of limit or the safety arbitration verification fails, the digital twin model is started by the master control node to perform reverse simulation, and the abnormality is reproduced by adjusting the virtual equipment state parameters to locate the root cause of the fault;

[0049] When the system needs to recover from the bottom safety mode to the normal optimization mode, the master control node automatically performs a controlled and gradual control right return sequence after safety simulation of the recovery process based on the digital twin model.

[0050] Step eight. Model and strategy self-learning update: periodically calibrate the digital twin model parameters with actual data, and dynamically optimize the weight coefficients in the multi-objective optimization model and the safety threshold values in the local physical safety rule set based on historical performance and cases through machine learning algorithms to realize continuous evolution of the system

[0051] It should be noted that:

[0052] The model and strategy self-learning update in step eight further comprises:

[0053] Periodically compare the digital twin model prediction data with the actual operation data, and when the average deviation continuously exceeds 5%, online calibrate the adsorption capacity and heat transfer coefficient key parameters of the model;

[0054] Based on historical operation performance data and safety arbitration cases, dynamically optimize the weight coefficients in the multi-objective optimization model and the safety rule threshold values through machine learning algorithms;

[0055] At the same time, based on the equipment state trend data, predict the remaining service life of the core components, and automatically generate predictive maintenance work orders.

[0056] In the above technical solution, the intelligent control method of the waste gas recovery system provided by the application has the following beneficial effects:

[0057] (1) The application improves the overall reliability, self-adaptability and expandability of the system. By constructing a "master control-subordinate control" distributed communication architecture, combining double-link redundant communication and local safety arbitration mechanism, not only the risk of centralized control is effectively dispersed, the robustness of the system in the case of local fault or network interruption is enhanced, but also the control area can be flexibly increased or adjusted according to the process flow change, significantly improving the modular level and engineering adaptability of the system.

[0058] (2) The application realizes intelligent decision-making and predictive control based on multi-objective dynamic optimization, by constructing a digital twin hybrid model that integrates mechanism and data-driven, the system can simulate and predict future working condition trends with high fidelity, based on this, a multi-objective optimization algorithm with constraints is used for online rolling solution, which can dynamically find the optimal balance point between recovery efficiency, energy consumption and equipment loss in real time, and generate global optimization instructions, thereby breaking through the limitations of traditional static rule control, and realizing the continuous optimization of system comprehensive operation efficiency.

[0059] (3) The application constructs a multi-level and active safety protection and closed-loop regulation system, by setting local physical safety rule set on the sub-control node for instruction verification and safety arbitration, forming a double protection of "global optimization of overall control-sub-control local safety bottom-up", combined with deviation grading processing and closed-loop feedback mechanism after execution, and reverse simulation tracing ability in fault, realizing the whole process control from instruction generation, safety verification, execution monitoring to abnormal handling, greatly improving the intrinsic safety level and the smooth switching and recovery ability under abnormal working conditions of the system.

[0060] (4) The application realizes self-learning and optimization of the system with the accumulation of running time through periodic online calibration of model parameters, dynamic adjustment of optimization weights and safety thresholds based on historical data and machine learning algorithm, and predictive maintenance based on device state trend, effectively resisting model drift and device performance degradation, maintaining high performance and high reliability in the long term, reducing the difficulty and cost of long-term operation and maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0062] Figure 1 The method flowchart provided for the intelligent control method embodiment of the waste gas recovery system of the present application.

[0063] Figure 2 The distributed control system architecture diagram provided for the intelligent control method embodiment of the waste gas recovery system of the present application.

[0064] Figure 3 The data acquisition and fusion flowchart provided for the intelligent control method embodiment of the waste gas recovery system of the present application.

[0065] Figure 4 The safety arbitration logic flowchart provided for the intelligent control method embodiment of the waste gas recovery system of the present application.

[0066] Figure 5 The fault tracing and system recovery flowchart is provided for the intelligent control method embodiment of the waste gas recovery system.

[0067] Figure 6 The self-learning update module architecture diagram is provided for the intelligent control method embodiment of the waste gas recovery system. DETAILED DESCRIPTION

[0068] In order to make the technical personnel in the art better understand the technical solutions of the present application, the present application will be further described in detail below with reference to the drawings.

[0069] As shown in the figure, the intelligent control method of the waste gas recovery system provided by the embodiment of the present application comprises the following steps: Figures 1-6

[0070] Step one. Arranging sensors and actuators: deploying sensor networks at key nodes of the whole system, connecting sensors, frequency conversion fans, regulating valves and other actuators of the waste gas recovery system to their corresponding regional control nodes respectively

[0071] It should be noted that:

[0072] The sensor network deployed in the whole process of waste gas treatment at least includes: PID / FID concentration meter or NDIR concentration meter for monitoring the composition and concentration of waste gas, thermal flowmeter or ultrasonic flowmeter for monitoring flow, temperature and humidity sensor, pressure transmitter, vibration sensor and current sensor for monitoring equipment state, and energy consumption metering module;

[0073] The actuators include frequency conversion fans, electric regulating valves, pumps and heaters, and are connected to the corresponding regional control nodes through field bus or hardwiring.

[0074] Step two. Building a distributed communication architecture: building a hierarchical control architecture containing one master control node and multiple regional control nodes, communicating between nodes through redundant industrial network to ensure reliable data transmission;

[0075] It should be noted that:

[0076] The master control node and each regional control node communicate redundantly through the dual link composed of industrial Ethernet and 5G / 4G wireless network, and are configured with automatic switching mechanism and local instruction cache function in case of communication interruption, and during the communication interruption, each control node can maintain the safe operation of the system according to the last valid instruction and local bottom control logic.

[0077] ​Step three. Real-time data acquisition and fusion: Each sub-control node periodically acquires sensor data, performs filtering and abnormality processing, and then fuses the data to generate a unified system operating condition state vector by the master control node;

[0078] It should be noted that:

[0079] Each sub-control node acquires data at a period of 1-10 seconds and uses Kalman filtering algorithm for noise elimination;

[0080] Abnormal data beyond the reasonable range are marked and removed, and backup sensor data or historical time series data are used for interpolation completion;

[0081] The master control node uses time alignment algorithm to unify the time stamp of the collected data, and fuses to generate a standardized operating condition data set (including exhaust process parameters, equipment state parameters and environmental parameters) representing the global operating state of the system.

[0082] Step four. Building and running digital twin model: The master control node builds a digital twin model based on a hybrid model of mechanism and data-driven model, and inputs the operating condition data set into the model, so that the digital twin model runs synchronously with the physical system;

[0083] It should be noted that:

[0084] The built digital twin model is a hybrid model that combines a mechanism model based on the mass and energy conservation laws of the exhaust gas recovery process, and a data-driven model trained based on historical operating data;

[0085] The data-driven model is constructed using least squares support vector machine (LSSVM) or neural network algorithm, which is used to correct the prediction bias of the mechanism model;

[0086] Simulation rehearsal is used to predict the trends of key indicators such as adsorption saturation, outlet pollutant concentration and system comprehensive energy consumption in the next 15-30 minutes.

[0087] Step five. Intelligent optimization to generate control instructions: The master control node uses the synchronized digital twin model to generate a global optimization instruction set containing the process parameter set values of each actuator, with the highest exhaust gas recovery rate, the lowest system energy consumption and equipment wear as the comprehensive optimization objective, under the preset safety and environmental protection constraints;

[0088] It should be noted that:

[0089] The online optimization solution uses a multi-objective optimization algorithm with constraints to generate a global optimization instruction set by solving the following comprehensive objective function online:

[0090] ;

[0091] Where η is the real-time waste gas recovery efficiency, E is the overall energy consumption of the system, D is the loss index of key equipment, and α, β, γ are self-learning adjusted weight coefficients, all of which are positive numbers.

[0092] The constraints to be solved include: equipment safety operation limits, environmental emission standards, and the physical feasible region of decision variables such as the main recovery unit switching cycle, regeneration unit operating parameters, fan speed and valve opening.

[0093] Step Six. Security Arbitration and Command Execution: After receiving the command, the sub-control node calls its locally stored physical security rule set for independent verification. If the verification passes, the command is executed; if the verification fails, the command is rejected and automatically switched to the preset safety mode, while simultaneously sending an alarm to the central control.

[0094] It should be noted that:

[0095] The local physical security rule set should include at least:

[0096] Regional material mass conservation verification rules, equipment mechanical motion limit rules, process parameter safety threshold rules, and actuator motion rate limit rules;

[0097] The results of security arbitration are divided into three categories: verification passed, verification alarm, and verification failed.

[0098] When verification fails, the sub-control node refuses to execute the global optimization command and automatically switches to the local storage safety mode, while sending an alarm to the central control node.

[0099] When verifying an alarm, the sub-control node sends an early warning message to the central control node while executing the instruction, and strengthens the status monitoring of its area;

[0100] In step six:

[0101] If the system is running in a safety-assured mode due to a security arbitration failure, and you need to restore optimized operation, perform the recovery steps as follows:

[0102] The central control node performs a security simulation of the recovery process from the current state to the optimized target state based on a digital twin model. After successful verification, it automatically issues a sequence of progressive control handover instructions to the sub-control nodes.

[0103] Step 7. Monitoring Execution and Closed-Loop Feedback: After execution, compare the deviation between the actual response and the model prediction. If it is within the specified range, maintain the result; otherwise, perform local fine-tuning or have the central control regenerate the instruction to form a closed loop.

[0104] It should be noted that:

[0105] The deviation between the actual response and the model prediction is compared, and the result is classified according to a preset deviation threshold:

[0106] When the deviation value is less than or equal to the first threshold value, the current control strategy is maintained;

[0107] When the deviation value is greater than the first threshold value and less than or equal to the second threshold value, a local parameter fine-tuning algorithm is started by the corresponding sub-control node for compensation control;

[0108] When the deviation value is greater than the second threshold value or a sudden abnormality is detected, the master control node is triggered to re-execute step five to generate a correction instruction, or a fault tracing analysis is started, wherein (the first threshold value is 3%, and the second threshold value is 10%);

[0109] During fault tracing analysis:

[0110] When the execution deviation is severely out of limits or the safety arbitration verification fails, the digital twin model is started by the master control node for reverse simulation, and the abnormality is reproduced by adjusting the virtual device state parameters to locate the fault root cause;

[0111] When the system needs to recover from the bottom safety mode to the normal optimization mode, the master control node automatically executes a controlled and gradual control right return sequence after simulating the recovery process based on the digital twin model.

[0112] Step eight. Model and strategy self-learning update: periodically calibrate the digital twin model parameters with actual data, and dynamically optimize the weight coefficients in the multi-objective optimization model and the safety threshold values in the local physical safety rule set based on historical performance and cases through machine learning algorithms to realize continuous evolution of the system

[0113] It should be noted that:

[0114] The model and strategy self-learning update in step eight further includes:

[0115] Periodically compare the digital twin model prediction data with the actual operation data, and when the average deviation continuously exceeds 5%, calibrate the adsorption capacity and heat transfer coefficient key parameters of the model online;

[0116] Based on historical operation performance data and safety arbitration cases, dynamically optimize the weight coefficients in the multi-objective optimization model and the safety rule threshold values through machine learning algorithms;

[0117] At the same time, based on the device state trend data, predict the remaining service life of the core components, and automatically generate predictive maintenance work orders.

[0118] Embodiment:

[0119] The system processing scale of this embodiment is 10000 m³ / h, the target VOCs recovery rate is ≥95%, and the emission concentration is ≤10 mg / m³.

[0120] I. System hardware and network configuration

[0121] The system hardware configuration is as follows:

[0122] Sensor network: install PID online concentration instrument (range 0-1000 ppm) and thermal flow meter at the exhaust gas inlet; install temperature and pressure transmitters at key process points such as zeolite rotary adsorption area, desorption area, condenser inlet and outlet; install current and vibration sensors on the circulating fan and desorption heater; install liquid level and pH meter on the recovery liquid tank.

[0123] Actuator: including variable frequency circulating fan (110 kW), electric regulating valve, gas heater (200 kW), etc.

[0124] Control node:

[0125] Regional control node: use three high-performance PLCs (such as Siemens S7-1200) as "pretreatment and air intake control node", "adsorption-desorption rotary wheel control node", and "condensation and recovery control node". Each node is directly connected to the sensors and actuators in its jurisdiction area.

[0126] Master control node: use an industrial edge server (or high-performance PLC such as S7-1500) to deploy digital twin platform, optimization algorithm library and database.

[0127] Communication network: the master control node and three sub-control nodes are connected through industrial Ethernet (Profinet) main link and 5GCPE wireless backup link, forming a redundant network, and configuring heartbeat detection and link automatic switching strategy.

[0128] II. Control method running process

[0129] Data acquisition and fusion (step three): three sub-control nodes synchronously collect local data with 5 seconds as the period. After preliminary processing by the Kalman filter program built-in the node, the data is uploaded to the master control node through the redundant network. The master control node runs the time alignment algorithm to unify the data from different nodes to the same timestamp, and fuses to generate a global working condition state vector containing [inlet concentration, flow, temperature and pressure at each point, fan current, valve opening, energy consumption…].

[0130] Digital twin pre-rehearsal and optimization decision (steps four and five):

[0131] The total control node inputs the real-time working condition vector into the digital twin hybrid model maintained by it. The mechanism part of the model is constructed based on adsorption kinetics and heat and mass transfer equations; the data-driven part uses LSSVM (Least Squares Support Vector Machine) and is trained using three months of historical data to correct the deviation of the mechanism model.

[0132] After the model is synchronized with the physical system, based on the current state and recent trends, the running state 20 minutes in the future is pre- simulated, and the change curves of the runner saturation, outlet concentration and total power consumption are predicted.

[0133] The simulation shows that there is a risk of exceeding the outlet concentration at the end of the current adsorption cycle. The intelligent strategy engine is immediately started, with "Max J = 0.6 * recovery rate - 0.3 * unit energy consumption - 0.1 * equipment load index" as the objective function, bed temperature < 120 ℃ and outlet concentration < 10 mg / m³ as the constraints, and particle swarm optimization algorithm (PSO) is used for online rolling solution.

[0134] The global optimization instruction set is obtained: {adsorption cycle is adjusted from 30 minutes to 22 minutes; desorption heating power is reduced from 85% to 78%; circulating fan frequency is fine-tuned from 45 Hz to 43.5 Hz}. The instruction set is verified by the expert rule base of the total control layer (such as the rule of avoiding frequent switching) and then issued.

[0135] Distributed safety arbitration and execution (step six):

[0136] The instruction is issued to the "adsorption-desorption runner control node". Before driving the actuator, the local safety verifier of the sub-control PLC is called.

[0137] The verifier checks according to the solidified physical safety rule set: rule 1 (device limit): fan target frequency 43.5 Hz < nameplate maximum frequency 50 Hz, pass; rule 2 (process safety): desorption target temperature (calculated from 78% power) ≈ 165 ℃ < VOCs flash point 230 ℃ - 50 ℃ safety margin = 180 ℃, pass; rule 3 (trend reasonable): valve planned action time is within the mechanical allowable range, pass.

[0138] After all the rules are verified, the PLC executes the instruction and feeds back "execution confirmation" to the total control.

[0139] Suppose the total control instruction requires the desorption temperature to be set to 250 ℃ due to model error or data pollution, then the local safety verifier will trigger rule 2 verification failure. The PLC will refuse the dangerous instruction and immediately switch to the local stored safety mode (such as fixed at 120 ℃ safe temperature), while sending a "instruction verification failure: violates process safety rule, temperature exceeds limit" safety arbitration alarm to the total control and central monitoring room.

[0140] Closed-loop monitoring, traceability, and recovery (Step 7):

[0141] After the command execution, the system continuously monitors the actual outlet concentration. After 10 minutes, the monitored concentration is 35 mg / m³, while the digital twin model predicts 30 mg / m³, with a deviation of approximately 16.7%, exceeding the second threshold (e.g., 10%).

[0142] The local control node first attempts local fine-tuning, increasing the fan frequency from 43.5 Hz to 44.5 Hz. The concentration stabilizes at 29 mg / m³.

[0143] Meanwhile, the master control node triggers a reverse traceability analysis due to the deviation. A reverse simulation is initiated in the digital twin model, quickly simulating multiple fault hypotheses. Ultimately, the hypothesis of "20% reduction in condenser heat exchange efficiency" is determined to best replicate the measured deviation. The condenser icon is highlighted on the human-machine interface (HMI) flowchart, and the maintenance strategy "recommend cleaning the condenser E-101 heat exchange tubes" is pushed.

[0144] After maintenance is complete, the operator clicks "restore optimal operation" on the HMI. The system initiates an intelligent recovery (rebound) mechanism: the master control node instructs the digital twin model to simulate the entire process of smoothly transitioning from the current "reduced frequency operation" state to the "43.5 Hz" state set by the optimization command. After the simulation is successful, the system automatically and slowly executes the frequency down sequence, achieving a "soft landing" rather than an instantaneous switch.

[0145] Self-learning and optimization (Step 8):

[0146] Model calibration: Every week, the system automatically compares the adsorption cycle predicted by the digital twin model with the actual cycle. It is found that the model underestimates by an average of 5%, so the zeolite adsorption capacity parameter in the model is updated online using this week's data.

[0147] Strategy optimization: The monthly analysis report shows that under the condition of environmental humidity > 80%, the original conservative rules result in low energy efficiency. After analyzing historical data, the machine learning module suggests optimizing the humidity trigger threshold for load reduction to 85% and adjusting the load reduction amplitude. After the engineer reviews, the optimization is updated to the expert rule library.

[0148] Predictive maintenance: The system continuously analyzes the pressure difference growth trend and vibration data of the adsorption runner to predict its remaining service life, generating a "recommend checking the runner seal" work order two weeks in advance.

[0149] The foregoing merely illustrates some exemplary embodiments of the application, and it will be appreciated that those skilled in the art will be able to devise various modifications without departing from the spirit and scope of the application. The appended drawings and description are illustrative only, and are not intended to be limiting.

Claims

1. An intelligent control method of a waste gas recovery system, characterized by, The method comprises the following steps: Step 1: arranging sensors and actuators: at key nodes in the whole system, deploy sensor networks and actuators according to the preset control area plan; Step 2: building a distributed communication architecture: build a hierarchical control architecture comprising one master control node and multiple regional sub-control nodes, the master control node and each sub-control node are connected through a redundant communication network to build a distributed communication architecture; Step 3: real-time data acquisition and fusion: each sub-control node periodically acquires sensor data, performs filtering and abnormality processing, and then generates a unified system working condition state vector by the master control node; Step 4: building and running a digital twin model: the master control node builds a digital twin model based on a hybrid model of mechanism and data-driven, and inputs the working condition data set into the model, so that the digital twin model and the physical system run synchronously; Step 5: intelligent optimization to generate control instructions: the master control node uses the synchronized digital twin model to generate a global optimization instruction set comprising the set values of the process parameters of each actuator, with the highest waste gas recovery rate, the lowest system energy consumption and equipment loss as the comprehensive optimization objectives, and under the preset safety and environmental protection constraints; Step 6: safety arbitration and instruction execution: after receiving the instructions, the sub-control node calls the locally stored physical safety rule set for independent verification, and executes if passed, and refuses to execute if failed, and automatically switches to the preset bottom safety mode, and alarms the master control at the same time; Step 7: monitoring execution and closed-loop feedback: after execution, compare the deviation between the actual response and the model prediction, and maintain within the specified range, otherwise make local adjustments or generate new instructions by the master control to form a closed loop; Step 8: model and strategy self-learning and updating: periodically calibrate the digital twin model parameters with actual data, and dynamically optimize the weight coefficients in the multi-objective optimization model and the safety threshold in the local physical safety rule set based on historical performance and cases through machine learning algorithms, to realize continuous evolution of the system.

2. The intelligent control method of a waste gas recovery system according to claim 1, characterized in that, In step 1: The sensor network deployed in the whole waste gas treatment process at least comprises: PID / FID concentration meters or NDIR concentration meters for monitoring waste gas composition and concentration, thermal flow meters or ultrasonic flow meters for monitoring flow, temperature and humidity sensors, pressure transmitters, vibration sensors and current sensors for monitoring equipment status, and energy metering modules; The actuators include variable frequency fans, electric regulating valves, pumps and heaters.

3. The intelligent control method of a waste gas recovery system according to claim 1, characterized in that, In step 2: The master control node and each regional sub-control node are connected through a dual-link of industrial Ethernet and 5G / 4G wireless network for redundant communication, and are configured with an automatic switching mechanism and a local instruction caching function in case of communication interruption; The sensors and actuators deployed in step 1 are connected to their corresponding regional sub-control nodes through field buses or hardwired connections.

4. The intelligent control method of a waste gas recovery system according to claim 1, characterized in that, In step 3: Each sub-control node acquires data at a cycle of 1-10 seconds, and uses Kalman filtering algorithm for noise elimination; Abnormal data beyond the reasonable range are marked and excluded, and are interpolated and completed by using backup sensor data or historical time series data; The total control node uses a time alignment algorithm to timestamp the converged data, and fuses to generate a standardized working condition data set representing the global running state of the system.

5. The intelligent control method of a waste gas recovery system according to claim 1, characterized in that, In the fourth step: The constructed digital twin model is a hybrid model that fuses a mechanism model based on the quality and energy conservation law of waste gas recovery process and a data-driven model trained based on historical operation data; The data-driven model is constructed using a least squares support vector machine or a neural network algorithm, and is used to correct the prediction deviation of the mechanism model; The simulation rehearsal is used to predict the change trend of key indicators such as adsorption saturation, outlet pollutant concentration and system comprehensive energy consumption in the next 15-30 minutes.

6. The intelligent control method of a waste gas recovery system according to claim 1, wherein, In the fifth step: The online optimization solution adopts a multi-objective optimization algorithm with constraints to generate a global optimization instruction set by solving the following comprehensive objective function through online rolling: ； Wherein, η is the real-time waste gas recovery efficiency, E is the system comprehensive energy consumption, D is the key equipment loss index, and α, β, γ are self-learning adjustment weight coefficients, and all are positive numbers; The constraint conditions include: device safe operation limit value, environmental protection emission standard, and physical feasible region of decision variables such as main recovery unit switching period, regeneration unit working parameter, fan speed and valve opening.

7. The intelligent control method of a waste gas recovery system according to claim 1, characterized in that, In the sixth step: The local physical safety rule set at least includes: Regional material mass conservation verification rules, equipment mechanical action limit rules, process parameter safety threshold rules, and actuator action rate limit rules; The results of the safety arbitration are divided into three categories: verification pass, verification warning, and verification failure, wherein: When the verification fails, the sub-control node refuses to execute the global optimization instruction, and automatically switches to the local stored bottom safety mode operation, and sends an alarm to the total control node; When the verification warning, the sub-control node sends a warning information to the total control node while executing the instruction, and strengthens the state monitoring of the local area; In the sixth step: When the system runs in the bottom safety mode due to safety arbitration failure, if it needs to restore the optimization operation, the recovery step is executed, as shown below: The total control node simulates the recovery process from the current state to the optimization target state based on the digital twin model, and automatically issues a gradual control right return instruction sequence to the sub-control node after verification. 8.The intelligent control method of a waste gas recovery system according to claim 1, characterized in that, In the seventh step: The deviation between the actual response and the model prediction is compared, and the deviation threshold is classified according to the preset deviation threshold: When the deviation value is less than or equal to the first threshold, the current control strategy is maintained; When the deviation value is greater than the first threshold and less than or equal to the second threshold, the local parameter fine tuning algorithm is started by the corresponding sub-control node for compensation control; When the deviation value is greater than the second threshold or a sudden abnormality is detected, the total control node is triggered to re-execute step five to generate a correction instruction, or to start fault tracing analysis.

9. The intelligent control method of a waste gas recovery system according to claim 1, wherein, In the seventh step: When fault tracing analysis: When the execution deviation is seriously out of limit or the safety arbitration verification fails, the digital twin model is started by the total control node for reverse simulation, and the abnormality is reproduced by adjusting the virtual equipment state parameters to locate the fault root cause; When the system needs to recover from the bottom safe mode to the normal optimization mode, a controlled and gradual control right return sequence is automatically executed based on the digital twin model simulation of the recovery process by the total control node.

10. The intelligent control method of a waste gas recovery system according to claim 1, wherein, The model and strategy self-learning update in step eight further includes: Periodically compare the digital twin model prediction data with the actual operation data, and when the average deviation continuously exceeds 5%, online calibrate the key parameters of the adsorption capacity and the heat transfer coefficient of the model; Based on the historical operation efficiency data and safety arbitration cases, dynamically optimize the weight coefficients and safety rule thresholds in the multi-objective optimization model through machine learning algorithms; At the same time, based on the equipment state trend data, predict the remaining service life of the core components, and automatically generate predictive maintenance work orders.

Images