Self-adaptive control algorithm for waste gas treatment process of fiber setting machine

Abstract

The invention relates to the field of waste gas treatment, and discloses a fiber setting machine waste gas treatment process adaptive control algorithm comprising the following steps: step a, collecting and calculating a process entropy of waste gas treatment equipment; b, when the sudden change of the working condition is not detected, online optimization control is executed to enable the process entropy to tend to be minimized; c, when the sudden change of the working condition is detected, performance gradient feed-forward control is executed, and a feed-forward compensation amount is output; and step d, calculating the feed-forward compensation amount by using the Hessian matrix of the entropy to obtain a final compensation amount, and adjusting operation parameters of waste gas treatment based on the final compensation amount. According to the method, a plurality of originally independent optimization targets such as the stability of the purification effect, the operation stability of the key unit, the change rate of unit energy consumption and the health state of the equipment are constructed into the unified process entropy, and the overall comprehensive operation level of the waste gas treatment equipment is improved by minimizing the process entropy and searching a global optimal balance point.

Description

Technical Field

[0001] This invention relates to the field of waste gas treatment, specifically to an adaptive control algorithm for the waste gas treatment process of a fiber setting machine. Background Technology

[0002] Fiber setting machines are key heat treatment equipment in modern textile industries for fabric stretching and finishing. During the high-temperature operation of this equipment, dyes, auxiliaries, fiber oils, and fine fiber debris adhering to the fabric evaporate or sublimate upon heating, forming high-temperature exhaust gas containing a large amount of oil fumes, particulate matter (PM2.5), volatile organic compounds (VOCs), and other complex components. This type of exhaust gas is characterized by high pollutant concentrations, complex composition, and pungent odors. Direct emission into the atmosphere not only causes serious pollution to the ecological environment but also poses a potential threat to human health. With increasingly stringent global environmental regulations, efficient and stable purification of fiber setting machine exhaust gas has become a mandatory requirement and technical prerequisite for textile enterprises to achieve green production and sustainable development.

[0003] Currently, while the technology for treating exhaust gas from fiber setting machines is relatively mature in terms of purification equipment itself, significant limitations remain in process control technology. Existing control methods mainly fall into two categories: one is open-loop control based on fixed parameters or human experience. To cope with peak pollution loads, this method typically sets operating parameters such as induced draft fan frequency and spray pump pressure at relatively high constant values. While this "worst-case" design can ensure compliance under extreme conditions, it results in significant energy waste under most stable or low-load conditions. The other category employs traditional feedback control strategies such as PID (proportional-integral-derivative). This method uses a single indicator, such as the pollutant concentration at the exhaust gas outlet, for feedback adjustment. However, due to the significant inertia and lag time of the exhaust gas treatment system, when upstream setting processes (such as fabric running speed and setting temperature) experience sudden changes, the response of the feedback control often exhibits a significant delay, making it difficult to avoid instantaneous exceedances of pollutant emissions.

[0004] To address the response speed issue, some technical solutions have introduced feedforward control, which monitors changes in upstream operating conditions to proactively compensate and adjust. However, traditional feedforward control heavily relies on a pre-set, static system model. Waste gas treatment systems are complex, time-varying systems whose treatment efficiency dynamically changes due to factors such as progressive clogging of filter units, aging of spray fluid, and natural performance degradation of equipment. This leads to a gradual mismatch between the static feedforward model and the actual system characteristics, ultimately causing overshoot or undershoot compensation, thus compromising long-term operational reliability. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an adaptive control algorithm for the exhaust gas treatment process of fiber setting machines. This algorithm solves the significant deficiencies in the control strategies of existing technologies, namely, the difficulty in coordinating economic optimization under steady-state operation with rapid response during sudden changes in operating conditions, and the general lack of adaptive capability to the time-varying characteristics of the system, which makes it difficult to guarantee the reliability and economy of its long-term operation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine, comprising the following steps:

[0007] First, to comprehensively quantify the operating status of waste gas treatment equipment, this invention constructs a comprehensive performance index, namely process entropy. This process entropy is obtained by calculating and fusing one or more process state variables, including: pollutant concentration at the waste gas outlet, spray system pressure or flow rate, filter unit pressure difference, total power of the waste gas treatment equipment, waste gas flow rate, and vibration signal of the waste gas treatment equipment.

[0008] In one specific embodiment, the process entropy can be calculated by weighted summing of the variance of pollutant concentration at the exhaust gas outlet, the coefficient of variation of spray system pressure or flow rate, the coefficient of variation of filter unit pressure difference, the rate of change of energy consumption per unit of exhaust gas treatment, and the spectral entropy of equipment vibration signal.

[0009] The control algorithm of this invention switches and coordinates between two control modes depending on whether the system is in a state of sudden change in operating conditions.

[0010] Steady-state online optimization control:

[0011] When no sudden changes in operating conditions originating from variations in the fabric running speed of the fiber setting machine and the set temperature of the setting process are detected, the system performs online optimization control. In this mode, the controller applies small, continuous probing disturbances to the control variables of one or more waste gas treatment devices (including the speed or pressure of the spray system circulating pump, the set frequency of the induced draft fan, the heating temperature of the gas whitening unit, and the opening degree of the damper).

[0012] Simultaneously, the controller, using online identification algorithms such as recursive least squares, estimates in real time the local model characteristics of the process entropy at the current operating point with respect to the control variable, i.e., the entropy gradient and the Hessian matrix of the entropy, based on the response of the process entropy caused by the disturbance. The identification of the Hessian matrix of the entropy is obtained through online estimation using recursive least squares.

[0013] The controller then uses the quasi-Newton method to calculate the adjustment amount of the control variables to minimize the process entropy based on the identified entropy gradient and the Hessian matrix of the entropy, thereby continuously pushing the equipment operating state towards the optimal. In a preferred embodiment, the amplitude of the probing disturbance can be adaptively adjusted based on the norm of the Hessian matrix of the entropy. Specifically, the larger the norm of the Hessian matrix of the entropy, the more sensitive the system is currently, and the smaller the amplitude of the probing disturbance will be automatically adjusted to ensure the stability of the optimization process.

[0014] Cooperative feedforward control:

[0015] When a sudden change in operating conditions is detected, the system immediately initiates performance gradient feedforward control and coordinates with the online optimization module.

[0016] First, the controller quantifies the disturbance change caused by the sudden change in operating condition, and then multiplies this disturbance change by a preset performance gradient matrix to quickly calculate a preliminary feedforward compensation amount. The mathematical expression for this calculation process is as follows:

[0017] ;

[0018] In the formula, Is The feedforward compensation amount, calculated at each time step, is used to offset the effects of disturbances. It is a vector of control variables. Column vectors of the same dimension; It is a preset performance gradient matrix. If the control variables have Dimension, disturbance variables have If the dimension is 1, then the matrix is ​​a single dimension. Matrix; In order to be in The disturbance changes obtained at any time are confirmed and quantified for use in feedforward control.

[0019] Furthermore, instead of directly using the feedforward compensation amount calculated based on the static model, the Hessian matrix, which is continuously updated under steady-state optimization control mode and contains the latest dynamic sensitivity information of the system, is used to refine and correct the feedforward compensation amount online. The specific calculation steps include: constructing a diagonal adjustment matrix based on the Hessian matrix of the aforementioned entropy. Its construction formula is:

[0020] ;

[0021] In the formula, Is A diagonal adjustment matrix that is constantly constructed to refine the feedforward compensation amount; This indicates that the elements within the parentheses are used as the main diagonal elements, and the remaining elements are zero, to construct a diagonal matrix; It is a preset, positive-zero adjustment sensitivity coefficient, which is used to adjust the response strength of the adjustment matrix to changes in the elements of the Hessian matrix; exist Hessian matrix of entropy identified at each moment The The first diagonal element directly reflects the process entropy on the first... The local curvature of the change of a control variable, i.e., sensitivity; For absolute value operations; Control the dimensions of variables.

[0022] After constructing the adjustment matrix, the final feedforward control adjustment amount used to compensate for disturbances is calculated using the following formula, which is the feedforward compensation amount after being corrected by the Hessian matrix of the entropy. The formula is expressed as:

[0023] ;

[0024] In the formula, It is the final compensation amount obtained after calculation of the feedforward compensation amount; through this calculation, a clear explanation can be obtained from the construction of the adjustment matrix:

[0025] For the Each control variable, and its corresponding feedforward compensation component, is multiplied by an adjustment factor. When the system is at the When the response is sensitive in each control direction, the corresponding diagonal elements of the Hessian matrix The absolute value of the denominator will be relatively large. This leads to a larger denominator. The increase causes the regulatory factor to decrease. (Approaching 0). Therefore, in the direction the system is sensitive to (where the absolute value of the corresponding element in the Hessian matrix is ​​large), the suppression effect on the feedforward compensation is stronger. Conversely, when the system is sluggish in that direction, Smaller, regulating factor As the value approaches 1, the feedforward compensation is essentially unaffected.

[0026] Subsequently, the feedforward compensation amount is corrected using the adjustment matrix to obtain the final compensation amount, the mathematical expression of which is:

[0027] ;

[0028] This calculation yields a new control variable vector. It may include:

[0029] ;

[0030] Each component within it will be converted into specific physical commands through the control system's data output interface and sent to the corresponding actuator.

[0031] The technical effect of this correction mechanism is that, in the Hessian matrix of entropy, the larger absolute values ​​of the elements indicate that the system is sensitive to response in that control direction. Therefore, the corresponding element in the adjustment matrix will have a smaller value, resulting in a stronger suppression effect on the corresponding component of the feedforward compensation, thus avoiding control overshoot. Conversely, in the direction where the system response is sluggish, the feedforward compensation remains essentially unchanged.

[0032] Finally, the controller adjusts the control variables based on the refined final compensation amount, thereby achieving rapid, accurate and stable compensation for sudden changes in operating conditions.

[0033] This invention provides an adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine. It has the following beneficial effects:

[0034] 1. This invention constructs a comprehensive performance index called process entropy, which unifies multiple originally independent optimization objectives, such as the stability of purification effect, the operational stability of key units, the rate of change of unit energy consumption, and the health status of equipment, into a single optimization framework. By minimizing process entropy, it can find the globally optimal balance point that takes into account purification, stability, energy saving, and equipment lifespan, thereby improving the overall comprehensive operation level of the waste gas treatment equipment.

[0035] 2. This invention employs a technical solution combining online optimization and feedforward compensation. When the operating conditions are stable or changing slowly, online optimization control is performed to continuously approach the optimal operating point. When a sudden change in production load is detected, the system immediately switches to rapid performance gradient feedforward control for predictive compensation. This allows the algorithm to cope with performance drift under steady-state conditions and effectively suppress severe impacts during transient processes, thereby maintaining the high efficiency and stability of the waste gas treatment system under various complex operating conditions such as fiber setting machine start-up, shutdown, and speed changes.

[0036] 3. This invention utilizes the Hessian matrix, which reflects the current sensitivity of the system and is identified in real time during the online optimization process, to dynamically correct and refine the feedforward compensation amount. By suppressing the compensation amplitude in the response-sensitive direction and providing sufficient compensation in the response-sluggish direction, the feedforward compensation based on the static historical model is transformed into dynamic compensation, thereby ensuring the efficiency and accuracy of the control strategy. Attached Figure Description

[0037] Figure 1 This is a schematic diagram of the algorithm flow of the present invention;

[0038] Figure 2 This is a schematic diagram illustrating the process entropy calculation of the present invention;

[0039] Figure 3 This is a flowchart of the steady-state optimization control of the present invention;

[0040] Figure 4 This is a flowchart of the feedforward compensation and correction process for sudden changes in operating conditions in this invention;

[0041] Figure 5 This is a schematic diagram of the process entropy surface and optimization process of the present invention. Detailed Implementation

[0042] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] To better understand the present invention, the above content will be described in detail below with reference to specific embodiments.

[0044] Please see the appendix Figure 1 -Appendix Figure 5 This invention provides an adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine, which includes the following steps:

[0045] The first step, step a, involves collecting variables from the waste gas treatment equipment in real time. This data, reflecting multiple dimensions of the equipment's performance, operational stability, energy efficiency, and health status, is then integrated into a single scalar value using a specific mathematical model (such as weighted summation). The higher the entropy value in this process, the further the system deviates from its ideal operating state.

[0046] The second step is step b. When the system is in a stable or slowly changing operating condition without affecting normal system operation, a small, pre-set probing disturbance is actively applied to one or more control variables (such as the frequency of the induced draft fan, the speed of the spray pump, etc.). By observing how these disturbances cause changes in the process entropy response, the system can identify the local geometric characteristics of the "entropy surface" near the current operating point in real time. Specifically, this is manifested as the entropy gradient (indicating the optimal descent direction) and the Hessian matrix of entropy (characterizing the concavity / convexity of the surface or the system's sensitivity to control). Based on these two key pieces of information, the controller can make intelligent decisions, continuously and incrementally adjusting the control variables to drive the system autonomously towards the minimum point of process entropy (i.e., the optimal operating point).

[0047] Next is step c. When a sudden change in the operating conditions from the front-end production process is detected (such as a significant adjustment in the stenter speed or temperature), the algorithm immediately switches to a fast-response performance gradient feedforward control mode. This mode calculates the specific change in the operating condition disturbance and, in conjunction with a preset performance gradient model, quickly calculates a feedforward compensation amount to offset the impact of the disturbance. The purpose of this step is to perform predictive and compensatory coarse adjustments to prevent significant degradation of system performance due to sudden changes in operating conditions.

[0048] Finally, step d involves using the Hessian matrix, which is identified in real-time by the steady-state optimization module in step b, to correct the feedforward compensation amount calculated in step c. The principle is that the Hessian matrix accurately reflects the system's current sensitivity to changes in different control variables. By adjusting the feedforward compensation amount using this sensitivity information (e.g., appropriately reducing the compensation amplitude in directions where the system response is sensitive), the final compensation amount can be calculated. Based on this adjustment of the control variables, it is ensured that the system can respond quickly to sudden changes while avoiding control overshoot and oscillations, achieving a smooth transition.

[0049] In a specific embodiment:

[0050] By deploying a series of sensors at key nodes of the exhaust gas treatment system, multiple process state variables that characterize the system's operating status are collected in real time.

[0051] Among these parameters, the pollutant concentration at the exhaust gas outlet is the most direct indicator for evaluating the purification performance of the system. An online oil fume concentration detector or photoionization detector is installed on the duct before the main exhaust stack to continuously measure the pollutant content in the treated gas. The 4-20mA standard analog signal output by the detector is sampled by the controller's A / D conversion module to obtain the discrete-time series of pollutant concentration values ​​at the exhaust gas outlet. .

[0052] As a key unit for waste gas pretreatment and cooling, the operational stability of the spray system is crucial. Therefore, a pressure transmitter or electromagnetic flow meter is installed on the main circulation pipeline of the spray system to monitor the pressure or flow rate. This measurement signal is then converted from an analog-to-digital converter (A / D converter) to obtain the spray system pressure or flow rate value quantifying the spray operating conditions. .

[0053] The fiber filter unit is the core purification component of this system, and its differential pressure changes can effectively reflect the degree of clogging and the health status of the filter media. By connecting the positive and negative pressure terminals of a differential pressure transmitter to its inlet and outlet sides respectively, the differential pressure value of the filter unit can be obtained in real time. .

[0054] To monitor the overall energy consumption level of the system, intelligent power meters are installed in the main circuit supplying power to the waste gas treatment equipment (including induced draft fans, water pumps, etc.) to measure the total power of the waste gas treatment equipment. Simultaneously, a Pitot tube flow meter or a thermal gas mass flow meter is installed in the main air duct of the system to measure the flow rate of the treated waste gas. The combination of these two variables is used to subsequently calculate the energy consumption per unit of waste gas treatment.

[0055] In addition, to monitor the mechanical health of critical equipment, accelerometers or vibration transmitters are fixed to the bearing housings of key rotating equipment such as induced draft fans. The vibration signals collected by the sensors are conditioned and sampled to obtain the vibration signal amplitude of the exhaust gas treatment equipment, reflecting the equipment's operational stability. .

[0056] In each control cycle The central controller combines the various process state variables, which are collected in real time and digitally processed, into a column vector, namely the process state variable vector. The mathematical expression is:

[0057]

[0058] in, This refers to the sequence number of the discrete control cycle, i.e., the time. This is the matrix transpose operator.

[0059] In this way, the physical operating state of complex waste gas treatment equipment can be successfully transformed into a structured mathematical object that the controller can recognize and process, and provide the necessary input for the subsequent calculation of process entropy.

[0060] After digitally characterizing the operating status of the waste gas treatment equipment, intervention in the waste gas treatment process is achieved by adjusting the operating status of one or more actuators. In a preferred embodiment, these control variables and their specific physical implementations may include:

[0061] One is the speed or pressure of the spray system's circulating pump. This variable is controlled by a frequency converter (VFD) connected to the motor of the spray circulating pump. By outputting a frequency setpoint to the VFD, the speed of the pump motor is adjusted by changing its output frequency, thereby directly changing the circulation flow rate and spray pressure of the spray liquid to meet the washing and cooling requirements under different exhaust gas loads.

[0062] Secondly, the set frequency of the induced draft fan is achieved through a high-power frequency converter connected to the main motor of the induced draft fan. The control algorithm can calculate the optimal operating frequency of the induced draft fan based on the results of optimization or feedforward compensation, and send this frequency setpoint to the frequency converter to precisely control the total air volume of the system.

[0063] Thirdly, there is the heating temperature of the gas de-whitening unit. For systems equipped with a gas de-whitening unit, the heating temperature needs to be controlled to eliminate the visual "white smoke" and ensure emission compliance. This control is achieved by adjusting the power supplied to the electric heating element inside the de-whitening unit. A temperature setpoint is output, which acts on an independent temperature control loop (e.g., a PID controller). This loop then adjusts the heating power by controlling the conduction angle of a silicon controlled rectifier or solid-state relay, ultimately stabilizing the outlet gas temperature at the setpoint.

[0064] Fourthly, there is the valve opening. In the exhaust gas treatment pipeline, a valve driven by an electric actuator can be installed. This actuator outputs a signal representing the target valve opening (e.g., 0-100%), such as a 4-20mA current signal, to the valve's positioner. The positioner drives the electric actuator to precisely achieve the set valve opening value, thereby adjusting the airflow distribution or overall resistance within the system.

[0065] In each control cycle The calculated set values ​​for each of the above physical quantities are combined into a multi-dimensional column vector, namely the control variable vector. The mathematical expression is as follows:

[0066] ;

[0067] In the formula, In order to be in The frequency setting value sent to the inverter of the spray system circulating pump at all times corresponds to the speed of the spray system circulating pump. In order to be in The frequency setting value sent to the frequency converter of the induced draft fan at all times corresponds to the set frequency of the induced draft fan. In order to be in The temperature setpoint is constantly sent to the temperature control loop of the gas whitening unit, which corresponds to the heating temperature of the gas whitening unit; In order to be in The position setting value is constantly sent to the electric actuator of the air valve, which corresponds to the opening degree of the air valve.

[0068] To ensure equipment safety and process stability, the aforementioned control vectors are pre-configured in the controller. Each component has its allowed operating range, i.e., upper and lower limits, and these constraints are based on the equipment's design specifications and process safety requirements. This allows for the establishment of a correspondence between abstract control algorithms and specific physical actuators.

[0069] After identifying the control output variables, in order to achieve predictive control of system disturbances, it is also necessary to obtain the core production process parameters from the source of the waste gas, namely the fiber setting machine itself.

[0070] The sudden changes in operating conditions addressed in this embodiment stem from variations in the fabric running speed of the fiber setting machine and the set temperature of the setting process. These two parameters directly determine the total amount of waste gas entering the waste gas treatment equipment per unit time, its temperature, and the concentration of pollutants, making them the main sources of disturbance causing drastic fluctuations in the load of the waste gas treatment system.

[0071] To obtain this critical disturbance information, a real-time data communication link is established between the central controller of the exhaust gas treatment equipment and the upper-level control system of the fiber setting machine (such as its main PLC or DCS system). This communication link can be implemented by laying an industrial Ethernet cable between the two and configuring standard industrial communication protocols, such as Modbus-TCP, OPCUA, or Profinet.

[0072] The central controller of the exhaust gas treatment equipment acts as a client or master station in this communication link, periodically initiating data read requests to the stenter control system with the same control cycle k as the aforementioned process state variable acquisition. The target address of the request is configured as a specific data register in the stenter controller that stores the current running speed of the fabric and the set temperature of the current process section.

[0073] In each control cycle The controller combines the instantaneous values ​​of these two key process parameters into a two-dimensional column vector, which is the measurable disturbance variable vector. The mathematical expression is as follows:

[0074] ;

[0075] In the formula, Is The fabric running speed of the fiber setting machine is obtained from the setting machine control system in real time; Is The setting temperature for the setting process is constantly obtained from the setting machine control system.

[0076] By establishing this data acquisition channel, load changes from the production source can be detected in advance. This is a real-time updated vector of measurable disturbance variables. This signal serves as the direct input signal for subsequent performance gradient feedforward control.

[0077] Because the exhaust gas treatment process of fiber setting machines involves complex gas-liquid two-phase flow, mass and heat transfer, chemical reactions, and the dynamic degradation of filter media performance, this invention characterizes the operating state of the exhaust gas treatment equipment using process entropy. According to this definition, a higher process entropy value indicates a more chaotic and unstable operating state, or worse performance in a certain aspect, meaning it is further from the ideal state. Conversely, a lower process entropy value indicates a more stable and efficient overall operating state of the exhaust gas treatment equipment. Therefore, by adjusting the aforementioned control variable vector... The goal is to seek and maintain a process entropy value that is as low as possible.

[0078] Process entropy is achieved by mathematically quantifying and weighting multiple process state variables that characterize the system's performance in different dimensions. This calculation method transforms the abstract system operating state into a concrete, computable scalar.

[0079] In a preferred embodiment, process entropy In the The formula for calculating each control cycle is as follows:

[0080] ;

[0081] In the formula, In order to be in The process entropy value calculated at each time step; , , , , These correspond to the non-negative weight coefficients of each entropy term, and their values ​​are preset according to specific process requirements and optimization priorities. Calculate the specified variable in the past Variance within a control cycle time window; Is The concentration of pollutants at the exhaust gas outlet is collected at all times; It calculates the past value of a specified variable. The coefficient of variation within a control cycle time window; Is The pressure or flow rate of the sprinkler system is collected in real time; Is The pressure difference of the filter unit is collected and calculated at all times; Is The total power of the waste gas treatment equipment collected at all times; Is The flow rate of the treated waste gas is collected in real time; It calculates the spectral entropy of a specified signal; Is Vibration signals from the waste gas treatment equipment are collected continuously. Among them:

[0082] It is the variance of pollutant concentration at the exhaust gas outlet, used to characterize the stability of the purification effect.

[0083] It is the coefficient of variation of the pressure or flow rate of the spray system, used to quantify the stability of the spray unit's operating conditions. The stability of the spray operation directly affects the processing efficiency of subsequent units and can accurately reflect whether the unit is operating under stable conditions.

[0084] It is the coefficient of variation of the pressure difference of the filter unit, which is used to characterize the health status of the core purification components (such as fiber cloth). Drastic fluctuations in pressure difference are an early sign of abnormal phenomena such as blockage, penetration or caking of the filter media. By incorporating this item into the process entropy, a stable operating condition that is conducive to extending the life of the filter media can be actively maintained.

[0085] It is the rate of change of energy consumption per unit of waste gas treatment, which is intended to monitor and optimize energy efficiency. This item calculates the changing trend of energy required to treat a unit volume or mass of waste gas. A positive rate of change indicates that the system energy efficiency is deteriorating. By minimizing it, the waste gas treatment equipment can find a more energy-efficient operating point.

[0086] It is the spectral entropy of the equipment vibration signal, used to quantify the mechanical health status of key rotating equipment (such as induced draft fans). Healthy mechanical equipment usually has a relatively simple vibration signal spectrum, with energy concentrated on a few frequencies and a low spectral entropy. When the equipment experiences wear, imbalance, or other faults, its vibration spectrum becomes more complex, and the spectral entropy increases accordingly. This makes the equipment health status one of the optimization targets.

[0087] By weighting and summing the above five entropy terms, a process entropy index that can comprehensively reflect the operating status of waste gas treatment equipment is constructed.

[0088] After completing the mathematical construction and calculation of the process entropy, online optimization control is executed. The goal of this control is to autonomously and continuously find the optimal combination of control variables that minimizes the calculated process entropy. To achieve this goal, it is necessary to first determine the optimal combination of control variables based on the calculated process entropy. With control variable vector The local dynamic relationship between them. This relationship can be abstractly understood as a multi-dimensional entropy surface. This invention uses an online identification process to obtain the local geometric characteristics of this surface near the current operation point in real time;

[0089] Specifically:

[0090] The first step is to... in each control cycle The calculated nominal control variables Above, a pre-designed, tiny probing perturbation vector is actively superimposed. The magnitude and direction of this disturbance vector are designed to ensure that it sufficiently excites the system dynamics without causing any perceptible impact on the actual production process. The final expression for the actual control quantity applied to the actuator is:

[0091] ;

[0092] The second step involves calculating the change in the control quantity after the application of the probing disturbance. :

[0093] ;

[0094] In the formula, In order to be in The vector of control variables applied to the exhaust gas purification equipment at all times; In order to be in The vector of control variables applied to the exhaust gas purification equipment at all times;

[0095] Then calculate the resulting change in process entropy. Its formula is:

[0096] ;

[0097] In the formula, In order to be in The change in process entropy calculated at each moment; In order to be in The change in process entropy calculated at each moment;

[0098] At this point, a set of input and output data about the local dynamics can be obtained;

[0099] Based on this data, this invention uses a second-order Taylor expansion to approximate the process entropy surface near the current operating point, and its expression is:

[0100] ;

[0101] This formula can be rewritten in the standard linear regression form:

[0102] ;

[0103] In the formula, for Measurable output at any given time; It is the change in the control variable The data vector formed by combining the second-order terms is called the regression vector; It is a parameter vector to be identified, whose internal elements correspond to the entropy gradient of the process entropy with respect to the control variables. The various components, and the Hessian matrix of entropy. Each of the individual elements.

[0104] The third step is to identify the parameter vector. The algorithm employs recursive least squares for online identification. In each control cycle, iteratively updates the parameter estimates using newly acquired data. The update law of recursive least squares consists of the following set of formulas:

[0105] ;

[0106] ;

[0107] ;

[0108] ;

[0109] In the formula, In order to be in Covariance matrix at time: In order to be in Covariance matrix at time: It is the transpose operator; In order to be in Time interval for parameter vector The estimated value; In order to be in Time interval for parameter vector The estimated value; In order to be in The prediction error at any given time; In order to be in The gain vector at time step; exist The covariance matrix at time t; Forgetting factor is a constant with a value between (0,1) used to adjust the weighting of historical data in the algorithm so that it can track the dynamic characteristics of time-varying systems.

[0110] Through the above recursive calculations, an updated parameter vector estimate can be obtained at the end of each control cycle. By analyzing this vector, we can output the entropy gradient of the process entropy with respect to the control variables and the Hessian matrix of the entropy. The entropy gradient indicates the direction in which the process entropy decreases the most, while the Hessian matrix of the entropy describes the local curvature of the entropy surface, reflecting the sensitivity to changes in different control variables.

[0111] In the online identification process, applying probing perturbations is a crucial step in obtaining dynamic information about the system to identify the entropy gradient and the Hessian matrix of entropy. However, perturbations with fixed amplitudes have inherent limitations: if the amplitude is too large, it will cause unnecessary interference to the stable operation of the system; if the amplitude is too small, it will result in a weak response signal of the process entropy and a low signal-to-noise ratio, thus affecting the accuracy and convergence speed of the identification algorithm.

[0112] Therefore, the Hessian matrix of the entropy identified in the previous control cycle is used to assess the current sensitivity of the system in real time, and the magnitude of the next probing disturbance is dynamically adjusted accordingly. Specifically:

[0113] The basic principle is that the norm of the Hessian matrix of entropy is a measure of the local average curvature of the process entropy surface. The magnitude of this norm directly reflects the overall sensitivity of the process entropy to changes in the control variable. The larger the norm, the more sensitive the system is to changes in the control variable, and the steeper the entropy surface is at that point; conversely, the smaller the norm, the more sluggish the system response, and the flatter the entropy surface. Therefore, by using the Hessian matrix of entropy identified in the previous control cycle, the current sensitivity of the system can be assessed in real time, and the magnitude of the next probing disturbance can be dynamically adjusted accordingly.

[0114] Therefore, in a specific embodiment, the calculation steps for the adjustment rule are as follows:

[0115] First, calculate in The Hessian matrix estimate of the entropy obtained by time step identification The Frobenius norm is expressed by the following formula:

[0116] ;

[0117] in, For matrix The Line number Column elements, To control the dimensions of the variables.

[0118] Then, based on that norm value, in the first... The amplitude vector of the probe disturbance applied in each control cycle Update using the following formula:

[0119] ;

[0120] In the formula, In order to be in The amplitude vector of the probe perturbation used at any given time; The maximum perturbation amplitude is defined as the preset base amplitude vector when the system response is least sensitive; This is a positive zero adjustment sensitivity coefficient used to regulate the degree to which the perturbation amplitude changes with the norm; In order to be in The Frobenius norm of the Hessian matrix of the entropy identified at each moment.

[0121] Through the above calculation and adjustment steps, a closed-loop adaptive mechanism can be achieved, thereby ensuring that the larger the norm of the Hessian matrix of the entropy, the smaller the amplitude of the probing disturbance. When the system is operating in a sensitive region, the disturbance is automatically reduced to ensure system stability; when the system is operating in a sluggish region, the disturbance is automatically increased to ensure identification accuracy.

[0122] After obtaining a real-time estimate of the local geometric properties of the entropy surface, online optimization control uses this information to calculate the adjustment amount of the control variable that can reduce the process entropy.

[0123] In this step, the present invention employs a control decision strategy based on the quasi-Newton method. This strategy utilizes both first-order information (entropy gradient) and second-order information (Hessian matrix of entropy) to determine the optimal descent direction and step size near the current operating point, thereby adjusting the control variables according to the entropy gradient and the Hessian matrix of entropy.

[0124] Specifically, in each control cycle The controller is based on the identified entropy gradient estimate. The Hessian matrix estimate of entropy Calculate the optimization adjustment amount used to update the nominal control variables. The core iterative formula for optimization based on the quasi-Newton method is:

[0125] ;

[0126] In the formula, In order to be in The control variable adjustment vector calculated at each time step and used for online optimization; A learning rate or step size coefficient with a value in the range (0,1] is used to control the convergence speed and stability of the optimization process; In order to be in The estimated value of the Hessian matrix of the entropy obtained at each time step; This is the operation for finding the inversion of a matrix. In order to be in The estimated value of the entropy gradient obtained at each time step.

[0127] To improve the numerical stability of the algorithm in actual operation and prevent [problems caused by] [other factors]. If an abnormal condition causes the calculation to fail, the matrix is ​​regularized before performing the inversion operation. A preferred implementation is the Levenberg-Marquardt method, expressed as:

[0128] ;

[0129] In the formula, If it is a positive number, it is called the regularization parameter; Is with The identity matrix of the same dimension; this correction ensures that the matrix to be inverted is always a positive definite matrix, thus guaranteeing the correctness of the optimization direction and the robustness of the calculation.

[0130] And by Nominal control variables acting on the previous period To update the nominal control variables for the current period. The expression is:

[0131] ;

[0132] This updated This refers to the new operational center point after moving in the direction of minimizing process entropy under steady-state optimization mode. By repeating this calculation and update process in each control cycle, the system is continuously and smoothly driven to converge towards the globally optimal operating point.

[0133] However, when the exhaust gas treatment system faces significant step disturbances from the upstream production process, relying solely on this feedback optimization mechanism results in insufficient response speed and cannot effectively suppress the impact of the disturbances on system performance. Therefore, this invention introduces performance gradient feedforward control, the first step of which is to rapidly detect and quantify sudden changes in operating conditions.

[0134] The detection and quantification process is based on the measurable perturbation variable vector mentioned above. This is achieved through continuous monitoring. In each control cycle... First, calculate the change in the disturbance vector relative to the previous period. The expression is:

[0135] ;

[0136] Subsequently, the controller compares the absolute values ​​of each component of the change vector with a pre-set disturbance change threshold, which is also in vector form. The comparison is performed. Each component of the threshold vector defines the minimum change that the corresponding perturbation variable (such as cloth speed or temperature) is considered to have undergone a "sudden change".

[0137] The occurrence of sudden changes in operating conditions, i.e., through judgment Does the absolute value of any component exceed The threshold value of the corresponding component is used to determine this. Once this condition is met, it is determined that the current control cycle is in progress. A sudden change in operating conditions occurred.

[0138] After confirming the occurrence of a sudden change in operating conditions, the algorithm of this invention will immediately calculate the disturbance change amount of the sudden change. In a specific embodiment, this disturbance change amount is defined as the disturbance change amount calculated in the current period that has been confirmed to have triggered the change determination. To clarify its purpose, it is labeled as the feedforward perturbation change. Its expression is:

[0139] ;

[0140] Therefore, through the above steps, real-time monitoring and rapid response to changes in front-end operating conditions are achieved. Once a disturbance exceeding the preset amplitude is detected, it is immediately and accurately quantified, and the quantification result is used as the input signal to start performance gradient feedforward control, providing a basis for subsequent rapid compensation calculations.

[0141] After accurately quantifying the disturbance change of the sudden change in operating conditions, based on this disturbance information, control compensation actions that can preemptively offset the main impact of the disturbance on system performance are calculated.

[0142] To achieve this goal, a performance gradient matrix is ​​pre-established. This matrix is ​​a linearized model that directly characterizes how each control variable needs to respond to changes in each measurable disturbance variable to maintain the stability of the process entropy, quantifying the static gain relationship between disturbance changes and the required control compensation.

[0143] This performance gradient matrix can be obtained through offline system identification methods. Specifically, during the equipment commissioning phase or using historical operating data, a series of known disturbance steps (e.g., changing the speed of the stenter) are applied, and the optimal control variable adjustment required to maintain stable system performance is recorded. Then, parameter estimation techniques such as least squares are used to fit the elements of the matrix. Once determined, the matrix is ​​stored in the controller as a set of fixed parameters.

[0144] When a sudden change in operating conditions is detected, the controller immediately performs the following calculation to output the feedforward compensation amount, expressed as:

[0145] ;

[0146] In the formula, Is The feedforward compensation amount, calculated at each time step, is used to offset the effects of disturbances. It is a vector of control variables. Column vectors of the same dimension; It is a preset performance gradient matrix. If the control variables have Dimension, disturbance variables have If the dimension is 1, then the matrix is ​​a single dimension. Matrix; In order to be in The disturbance changes obtained at any time are confirmed and quantified for use in feedforward control.

[0147] This matrix multiplication operation can be completed instantaneously within the controller, thus ensuring the speed of feedforward compensation, and the calculated feedforward compensation amount... This represents a coarse-tuned control action based on static model predictions, designed to respond to changes in current disturbances.

[0148] Therefore, through this calculation, the algorithm of the present invention can predictively and quickly generate a compensatory control output based on changes in front-end process parameters, thereby proactively intervening before the impact of the disturbance is fully transmitted to the system output and causes the process entropy to deteriorate.

[0149] And the initial feedforward compensation amount for fast response is calculated. Subsequently, due to the performance gradient matrix It is a static model based on offline identification or historical data, reflecting the gain relationship of the system under a certain average or specific state. However, the actual dynamic characteristics of the exhaust gas treatment system are time-varying. For example, the gradual clogging of the filter unit, the change in the chemical composition of the spray liquid, and the influence of ambient temperature and humidity will all cause the system's response to the same control input to change.

[0150] Therefore, the feedforward compensation calculated based on the static model is only a rough estimate for the current operating conditions. If it is executed directly without adjustment, it will lead to control overshoot and oscillation when the actual response of the exhaust gas treatment equipment is more sensitive than the model; while when the actual response of the system is less sensitive than the model, it will lead to insufficient compensation and inability to effectively suppress disturbances.

[0151] At the same time, the Hessian matrix of the entropy of the latest dynamic characteristics obtained from the optimization control can be used. The Hessian matrix of this entropy is not a static parameter, but a real-time quantification of the local curvature of the process entropy surface. It contains information on the sensitivity of the algorithm of this invention to changes in various control variables at the current precise operation point.

[0152] Therefore, at this point, the speed of feedforward control can be combined with the real-time accuracy of steady-state optimization control.

[0153] Specifically, the Hessian matrix of the entropy output in online optimization control is used to calculate the calculated feedforward compensation amount.

[0154] The principle is to use the Hessian matrix of entropy as a dynamic adjustment scheme tailored to the current operating conditions, applying it to the feedforward compensation amount generated by the static model. When the corresponding element value of the Hessian matrix is ​​large, it indicates that the exhaust gas treatment equipment is currently very sensitive to this control precision; a small change in control can cause a drastic change in process entropy. In this case, the compensation amplitude in this direction should be appropriately suppressed to avoid overshoot. Conversely, when the corresponding element value of the Hessian matrix is ​​small, it indicates that the system response is relatively sluggish in this dimension, and a relatively sufficient amount of compensation can be applied.

[0155] Therefore, by coordinating the two processes, the compensation action of feedforward control is no longer based on a fixed, historical system model, but is instead injected with real-time information about the current actual response characteristics. This allows the final compensation action to maintain the fast response characteristics of feedforward control while incorporating adaptive adjustments to the current dynamic characteristics of the system, resulting in more precise compensation actions and ultimately achieving a smooth transition of the transient process. Specifically, this includes:

[0156] This approach enables dynamic correction of the initial feedforward compensation amount using the Hessian matrix of real-time identified entropy. The core of this construction is to build an "adjustment matrix" that can transform the system's current sensitivity information into a component-level scaling factor for the feedforward compensation amount.

[0157] Specifically, when a sudden change in operating conditions is detected and Calculate the initial feedforward compensation amount at any time. Subsequently, the algorithm of this invention does not stop the operation of the online optimization module. Therefore, at this time, the algorithm can hold... The estimated value of the Hessian matrix of entropy at the moment it has just been updated. This invention utilizes the diagonal elements of the matrix to construct a diagonal adjustment matrix. ;

[0158] The adjustment matrix The construction formula is as follows:

[0159] ;

[0160] In the formula, Is A diagonal adjustment matrix that is constantly constructed to refine the feedforward compensation amount; This indicates that the elements within the parentheses are used as the main diagonal elements, and the remaining elements are zero, to construct a diagonal matrix; It is a preset, positive-zero adjustment sensitivity coefficient, which is used to adjust the response strength of the adjustment matrix to changes in the elements of the Hessian matrix; exist Hessian matrix of entropy identified at each moment The The first diagonal element directly reflects the process entropy on the first... The local curvature of the change of a control variable, i.e., sensitivity; For absolute value operations; Control the dimensions of variables.

[0161] After constructing the adjustment matrix, the final feedforward control adjustment used to compensate for disturbances is calculated using the following formula, which is the feedforward compensation amount after the Hessian matrix correction by entropy. The formula is expressed as:

[0162] ;

[0163] In the formula, It is the final compensation amount obtained after calculation of the feedforward compensation amount; through this calculation, a clear explanation can be obtained from the construction of the adjustment matrix:

[0164] For the Each control variable, and its corresponding feedforward compensation component, is multiplied by an adjustment factor. When the system is at the When the response is sensitive in each control direction, the corresponding diagonal elements of the Hessian matrix The absolute value of the denominator will be relatively large. This leads to a larger denominator. The increase causes the regulatory factor to decrease. (Approaching 0). Therefore, in the direction the system is sensitive to (where the absolute value of the corresponding element in the Hessian matrix is ​​large), the suppression effect on the feedforward compensation is stronger. Conversely, when the system is sluggish in that direction, Smaller, regulating factor As the value approaches 1, the feedforward compensation is essentially unaffected.

[0165] This calculation allows for the fusion of coarse feedforward compensation based on a static model with precise operational information of the waste gas treatment equipment based on real-time data, resulting in the calculated feedforward compensation amount, which is the final compensation amount.

[0166] In one specific embodiment, after obtaining the final compensation amount, based on the final compensation amount... To update and generate the current control cycle. The final control variable vector applied to the exhaust gas treatment equipment To distinguish them, the ones here will be... Recorded as .

[0167] This update process involves updating the control variable vector from the previous time step. The vector summation of the calculated final compensation amount is expressed as follows:

[0168] ;

[0169] This calculation yields a new control variable vector. It may include:

[0170] ;

[0171] Each component within it will be converted into specific physical commands through the control system's data output interface and sent to the corresponding actuator.

[0172] Specifically:

[0173] Sprinkler system circulating pump speed component The value is converted into a new frequency setting value (e.g., 35.2Hz) for the frequency converter of the spray system circulating pump, and transmitted to the frequency converter connected to the motor of the spray system circulating pump via communication or analog output, thereby controlling the speed of the spray system circulating pump.

[0174] induced draft fan set frequency component This will be used as a new frequency setting value (e.g., 48.5Hz) and sent to the induced draft fan frequency converter via industrial Ethernet communication or analog output module, thereby regulating the set frequency of the induced draft fan;

[0175] Heating temperature component of the gas dewhitening unit If the temperature is set, it will be updated to its independent temperature control loop as a new temperature setpoint, thereby regulating the temperature setpoint of the gas whitening unit temperature control loop and thus adjusting the heating temperature of the gas whitening unit.

[0176] damper opening component This signal will be converted into a standard current signal, which will drive the electric actuator of the damper to adjust to a new opening position, thereby regulating the opening of the damper.

[0177] Therefore, by transforming the final compensation amount into coordinated control actions of multiple physical actuators, compensation can be completed in advance before disturbances significantly affect process entropy, thus maintaining the stability and efficiency of the waste gas treatment process.

Claims

1. An adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine, characterized in that, Includes the following steps: Step a: Collect and calculate the process entropy of the waste gas treatment equipment during the waste gas treatment process. The process entropy is used to quantify the operating status of the waste gas treatment equipment. Step b: When no sudden change in operating condition is detected, online optimization control is performed: by applying a probing disturbance to the control variables of one or more waste gas treatment devices, and based on the response of the process entropy, outputting the entropy gradient of the process entropy with respect to the control variables and the Hessian matrix of the entropy, and adjusting the control variables based on the entropy gradient and the Hessian matrix of the entropy to minimize the process entropy; Step c: When a sudden change in operating condition is detected, perform performance gradient feedforward control: calculate the disturbance change of the sudden change in operating condition and output the feedforward compensation amount; Step d: Using the Hessian matrix of the entropy output in the online optimization control described in step b, calculate the feedforward compensation amount obtained in step c to obtain the final compensation amount, and adjust the control variable based on the final compensation amount. The control variable is used to regulate the operating parameters of the waste gas treatment equipment during waste gas treatment.

2. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, In step a, the calculation of the process entropy includes: weighted summation of the variance of pollutant concentration at the exhaust gas outlet, the coefficient of variation of spray system pressure or flow rate, the coefficient of variation of filter unit pressure difference, the rate of change of energy consumption per unit of exhaust gas treatment, and the spectral entropy of equipment vibration signal.

3. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, In step b, the adjustment of the control variable based on the entropy gradient and the Hessian matrix of entropy is carried out using the quasi-Newton method, which uses the inverse of the Hessian matrix of entropy and the entropy gradient to calculate the adjustment amount of the control variable.

4. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, In step c, the calculation of the feedforward compensation amount includes multiplying the disturbance change amount by a preset performance gradient matrix.

5. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, In step d, the step of calculating the feedforward compensation amount using the Hessian matrix of the entropy includes: An adjustment matrix based on the Hessian matrix of the entropy is constructed, and the adjustment matrix is ​​used to correct the feedforward compensation amount. The portion of the Hessian matrix of the entropy with a larger absolute value has a stronger suppression effect on the corresponding component of the feedforward compensation amount in the adjustment matrix.

6. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, In step b, the magnitude of the probing perturbation is adaptively adjusted based on the norm of the Hessian matrix of the entropy; the larger the norm of the Hessian matrix of the entropy, the smaller the magnitude of the probing perturbation.

7. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, The Hessian matrix of the entropy is identified by online estimation using the recursive least squares method.

8. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 2, characterized in that, The process entropy is based on one or more process state variables; the process state variables include: pollutant concentration at the exhaust gas outlet, spray system pressure or flow rate, filter unit pressure difference, total power of the exhaust gas treatment equipment, exhaust gas flow rate, and vibration signal of the exhaust gas treatment equipment.

9. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, The control variables include the rotational speed or spray pressure of the spray system circulating pump, the set frequency of the induced draft fan, the heating temperature of the gas dewhitening unit, and the opening degree of the air valve.

10. The adaptive control algorithm for the exhaust gas treatment process of a fiber setting machine according to claim 1, characterized in that, The sudden change in operating conditions originated from variations in the fabric running speed of the fiber setting machine and the set temperature of the setting process.

Images