Multi-variable dynamic collaborative optimization control method for waste incineration flue gas deacidification

Abstract

The embodiment of the application discloses a kind of garbage incineration flue gas deacidification multivariable dynamic collaborative optimization control methods, it is related to garbage incineration flue gas deacidification intelligent control technical field.The method includes: obtaining real-time data is preprocessed, and temperature-pressure coupling coefficient K T‑P It is calculated;Real-time data after preprocessing and the temperature-pressure coupling coefficient K T‑P , rule weight W i It is calculated;The rule weight W i After fuzzy rule matching, preliminary control amount ΔQ slurry It is output, wherein ΔQ slurry It is slurry flow variation;Forecast-energy efficiency optimization, by objective function and constraint condition, quadratic programming is solved optimal control amount, and optimization instruction is output to actuating mechanism;According to current system state and pre-set reward function, rule base is updated.The present application can solve the problem of low control precision and high operating cost in the process of garbage incineration flue gas desulfurization and deacidification, improve the environmental benefits and economic benefits of garbage incineration plant.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for flue gas desulfurization in waste incineration, and in particular to a multivariate dynamic collaborative optimization control method for flue gas desulfurization in waste incineration. Background Technology

[0002] With the continuous development of society and increasingly stringent environmental protection requirements, waste incineration has become one of the key methods for urban waste management. However, the flue gas produced during waste incineration contains a large amount of acidic and harmful gases such as HCl and SO2, which will cause serious environmental pollution if not effectively purified.

[0003] The current "Standard for Pollutant Control of Municipal Solid Waste Incineration" (GB18485-2014) is no longer able to meet current environmental protection needs. Zhejiang, Hebei and other places have launched pilot projects to upgrade the standards and issued relevant plans, which put forward stricter requirements for emission indicators and purification facilities of waste incineration plants.

[0004] In the flue gas purification system of waste-to-energy plants, the SDA (Spray Dryer Absorber) flue gas desulfurization process is a core component. However, the control process of this process is characterized by large time delays, nonlinearity, and multiple variables. Currently, most municipal solid waste incineration power plants use traditional PID (Proportional-Integral-Derivative) control algorithms and conventional fuzzy control + predictive control methods, which struggle to accurately control the amount of quicklime slurry added when flue gas pollutant concentrations fluctuate significantly. This not only leads to a substantial increase in quicklime consumption and power consumption but may also result in emissions exceeding standards, seriously impacting environmental and economic benefits. Existing methods often lack effective control and accurate calculation of these factors.

[0005] Currently, most municipal solid waste incineration power plants employ traditional PID control algorithms, with flue gas desulfurization control programs designed and executed within the flue gas purification system's PLC (Programmable Logic Controller) or DCS (Distributed Control System). The specific process involves adjusting the opening of the lime slurry solution valve based on real-time monitoring of acidic component (such as HCl and SO2) levels in the emitted flue gas using a traditional PID controller. When the detected acidic component content is high, the controller increases the injection rate of lime slurry solution, thereby enhancing the acid-base neutralization effect; conversely, when the acidic component content decreases to a lower level, the controller reduces the amount of lime slurry solution used. This control method ensures that the acidic component content in the emitted flue gas remains within the standards stipulated by environmental protection authorities.

[0006] Traditional PID control has limitations. While widely used in industrial automation, its linear control characteristics are ill-suited for nonlinear, time-varying systems in waste incineration flue gas desulfurization. Specific drawbacks are as follows:

[0007] Parameter tuning is difficult: The concentrations of HCl and SO2 in waste incineration flue gas are significantly affected by the composition of waste (such as chlorinated plastics and sulfides) and combustion efficiency, exhibiting significant fluctuations (instantaneous changes can reach ±100%). Traditional PID control requires manual tuning of parameters (Kp, Ki, Kd) based on experience, but when operating conditions change frequently, fixed parameters cannot guarantee control effectiveness.

[0008] Overshoot and oscillation risks: The flue gas desulfurization system has a large time delay (it takes 20-50 seconds from slurry adjustment to concentration feedback), and the PID integral term is prone to accumulate errors, which can lead to overshoot or continuous oscillation.

[0009] Multivariable coupling problem: Traditional PID control mainly uses single-loop and cascade regulation, which cannot solve the strong coupling effect between temperature, pressure and pollutant concentration. For example, an increase in flue gas temperature will reduce the atomization efficiency of the slurry and the deacidification efficiency in the deacidification tower, resulting in a decrease in the removal rate of acidic gases and an increase in the consumption of deacidification materials.

[0010] The application of fuzzy control and predictive control combined with feedforward control also has limitations. Fuzzy control solves nonlinear problems by simulating human experience, but it still faces the following bottlenecks in waste incineration scenarios:

[0011] Rule bases rely on human experience: existing fuzzy rule bases are mostly designed based on limited operating condition data, such as only considering moderate fluctuations in HCl concentration (±10 mg / m³). 3 The system does not cover extreme operating conditions (such as a sudden increase in HCl concentration to 50 mg / m³ due to the mixing of large amounts of chlorine-containing waste into the garbage). 3 For example, a waste incineration plant in Zhejiang has a fuzzy rule library containing 50 rules, but when processing waste with more than 30% chlorinated plastics, the lack of corresponding rules resulted in HCl emissions exceeding the standard three times that day.

[0012] Long-term performance degradation: The fuzzy rule base lacks a self-learning mechanism and cannot adapt to changes such as equipment aging and sensor drift after long-term operation. For example, when the efficiency of the slurry pump decreases, the flow command output by the original rule cannot achieve the expected effect.

[0013] Rule conflicts and redundancy: Under complex operating conditions, multiple rules may trigger contradictory instructions. For example, the simultaneous existence of rules stating "high HCl concentration requires increasing slurry flow" and "high temperature requires decreasing slurry flow" can cause frequent system oscillations. Experimental verification: In a scenario simulating a simultaneous increase in flue gas temperature and HCl concentration, the slurry flow fluctuation of traditional fuzzy control reaches ±20%, which is three times greater than that of single-variable operating conditions.

[0014] The challenges of implementing predictive control: Predictive control (such as model predictive control (MPC)) relies on high-precision models and data, but faces the following challenges in the waste incineration scenario:

[0015] Model inaccuracy: The composition of waste incineration flue gas is complex, and traditional ARIMA (Autoregressive Integrated Moving Average) models and state-space models are insufficient to accurately describe the dynamic characteristics of HCl / SO2. For example, HCl concentration is affected by the reaction rate of Cl- with metal compounds, and its kinetic equation contains multiple nonlinear terms. Existing models often simplify this to a linear relationship, leading to prediction errors. For instance, when the calorific value of waste changes by 20%, the predicted HCl concentration deviation reaches 8 mg / m³. 3 This triggers erroneous adjustments.

[0016] Data quality bottleneck: The harsh environment of flue gas detection (high temperature, high dust) makes sensor signals susceptible to noise interference. For example, in an 800℃ environment, the signal-to-noise ratio (SNR) of a laser spectrometer drops below 10dB due to thermal radiation noise, resulting in high-frequency jitter in the measured data.

[0017] Computational resource constraints: Real-time prediction requires online solving of optimization problems (such as quadratic programming), which places high demands on the controller's computing power. For example, the MPC algorithm requires 50ms for a single iteration, while the waste incineration control cycle needs to be ≤100ms, causing the controller to operate under overload.

[0018] In summary, with increasingly stringent environmental protection requirements and the rapid development of automation technology, these control methods have gradually revealed their limitations, such as slow response speed, limited accuracy, and high resource consumption costs. Therefore, more advanced and efficient automated control methods are needed to optimize the flue gas desulfurization process, improve overall operating efficiency and environmental performance, and compensate for the many shortcomings of previous control methods in the flue gas desulfurization process of waste incineration power plants. Summary of the Invention

[0019] In view of this, embodiments of the present invention provide a multivariate dynamic collaborative optimization control method for desulfurization and deacidification of waste incineration flue gas, in order to solve the problems of low control accuracy and high operating costs in the existing desulfurization and deacidification process of waste incineration flue gas, and improve the environmental and economic benefits of waste incineration plants.

[0020] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0021] A multivariable dynamic collaborative optimization control method for acid removal from waste incineration flue gas includes:

[0022] Step S101: Acquire real-time data, preprocess it, and calculate the temperature-pressure coupling coefficient K. T-P ;

[0023] Step S102: Utilize the preprocessed real-time data and the temperature-pressure coupling coefficient K T-P Calculate the rule weight W i ;

[0024] Step S103: Utilize the rule weight W i After fuzzy rule matching, the initial control quantity ΔQ is output. slurry , where ΔQ slurry This represents the change in slurry flow rate;

[0025] Step S104: Perform predictive-energy efficiency optimization. Solve for the optimal control quantity through quadratic programming using the objective function and constraints, and output the optimized command to the actuator.

[0026] Step S105: Update the rule base according to the current system status and the pre-set reward function.

[0027] The present invention has the following beneficial effects:

[0028] Improved control precision: By comprehensively utilizing multi-variable dynamic collaborative optimization control, high-precision control of the flue gas desulfurization and deacidification process is achieved, and the system response time is improved by 28.1%.

[0029] Energy saving and consumption reduction: Through multi-variable coordinated control and precise parameter adjustment, the use of quicklime slurry, process water and energy is optimized, reducing operating costs and energy consumption. The operating cost of quicklime is reduced by 7.7%.

[0030] Enhanced system stability: Advanced control algorithms and real-time monitoring mechanisms improve the system's adaptability to changes in operating conditions and disturbances, enhancing system stability and reliability, and reducing the number of overshoots in the control system by 16.7%. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating the multivariate dynamic collaborative optimization control method for deacidification of waste incineration flue gas according to the present invention.

[0033] Figure 2 This is a schematic diagram of the multivariable dynamic collaborative optimization control method for deacidification of waste incineration flue gas according to the present invention.

[0034] Figure 3 This is a schematic diagram of the dynamic weight control principle in this invention;

[0035] Figure 4 This is a schematic diagram of the prediction-energy efficiency integrated optimization principle in this invention;

[0036] Figure 5 This is a schematic diagram of the reinforcement learning rule base in this invention. Detailed Implementation

[0037] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0038] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0039] The core shortcomings of existing technologies can be summarized as follows: linear control cannot handle nonlinear systems, fuzzy rules lack dynamic optimization capabilities, and prediction models are limited by data quality and computing power. These problems directly lead to a triple dilemma for waste incineration plants: large emission fluctuations, high operating costs, and high difficulty in retrofitting.

[0040] This invention presents a systematic solution to the aforementioned pain points, addressing the problems of low control precision and high operating costs caused by the high volatility (instantaneous changes in HCl / SO2 concentration of ±100%) and strong coupling (interaction between temperature, pressure, and pollutant concentration) in the desulfurization and deacidification process of waste incineration flue gas. This invention proposes a multivariate dynamic collaborative optimization control method, aiming to provide a high-precision, adaptive, and low-cost intelligent control method for desulfurization and deacidification of waste incineration flue gas. This achieves comprehensive and high-precision control of the flue gas deacidification process, ensuring stable and economical HCl and SO2 emissions, reducing operating costs, and minimizing environmental impact.

[0041] This invention provides a multivariate dynamic collaborative optimization control method for acid removal from waste incineration flue gas, such as... Figure 1-2 As shown, it includes:

[0042] Step S101: Acquire real-time (flue gas) data, preprocess it, and calculate the temperature-pressure coupling coefficient K. T-P ;

[0043] As an optional embodiment, in step S101, the real-time data includes HCl concentration, SO2 concentration, temperature, and pressure. Preprocessing can be various conventional processing methods in the art, such as filtering and noise reduction, which will not be elaborated here.

[0044] like Figure 3 As shown, in specific implementation, input variables may also include:

[0045] e HClHCl concentration deviation (set value - measured value);

[0046] Δe HCl : Rate of change in HCl concentration deviation.

[0047] As another optional embodiment, the temperature-pressure coupling coefficient K T-P The calculation formula is:

[0048]

[0049] (Temperature range: T) min =120℃, T max =180℃; Pressure range: P min = -500Pa, P max =0Pa)

[0050] Parameter definition:

[0051] T: Real-time temperature value (unit: °C);

[0052] T min Lower limit of temperature (120℃); T max Temperature limit (180℃);

[0053] P: Real-time pressure value (unit: Pa);

[0054] P min Lower pressure limit (-500 Pa); P max Pressure limit (0 Pa).

[0055] Temperature Item Normalizing temperature T to the [0,1] interval reflects its relative position within its extreme range. When T = T min When T = T, the temperature term is 0; when T = T max At that time, the temperature term value was 1.

[0056] stress item Similarly, the pressure P is normalized to [0,1].

[0057] Weighting: The weights of T and P are 0.3 and 0.7 respectively, indicating that in the waste incineration process, the fluctuation of pressure P has a greater impact on the acid removal efficiency.

[0058] Overall result: K T-P The value range is [0,1], and the larger the value, the better the overall performance of T and P.

[0059] It is worth noting that if T or P exceeds the extreme value range, additional processing is required (such as truncating or expanding the extreme value).

[0060] Step S102: Utilize the preprocessed real-time data and the temperature-pressure coupling coefficient K T-P Calculate the rule weight W i ;

[0061] This step involves a dynamic weighted collaborative control algorithm.

[0062] As an optional embodiment, a weight adaptive mechanism is adopted (i.e., the rule weights are dynamically adjusted according to real-time operating conditions). In step S102, the rule weight W i The calculation formula is:

[0063] W i =α·(C SO2 / C HCl )+(1-α)·K T-P

[0064] Among them, C SO2 / C HCl The SO2 / HCl concentration ratio is used, and the weighting coefficient covers extreme operating conditions.

[0065] α is an adaptive coefficient, employing an adaptive mechanism: initially α = 0.6, according to C HCl / C SO2 The concentration ratio is dynamically adjusted. When the typical ratio is 1:2 to 1:5, α is automatically adapted to 0.5 to 0.7 to ensure increased weighting under high SO2 conditions.

[0066] Example rule (W under high HCl concentration and low temperature conditions) i calculate):

[0067] Real-time operating data: HCl = 50 mg / Nm³ 3 (Set value 10mg / Nm) 3 , deviation e HCl = -40mg / Nm 3 (This belongs to the category of "negative big NB").

[0068] SO2 = 100 mg / Nm 3 Concentration ratio of C SO2 / C HCl =100 / 50=2 (equal to the lower limit, then α=0.5).

[0069] Temperature T = 130℃ (T min =120℃, normalized value (130-120) / (180-120) = 0.167).

[0070] Pressure P = -300 Pa (P min = -500Pa, normalized value (-300+500) / 500 = 0.4).

[0071] The core problem is that the HCl concentration far exceeds the set value, requiring a significant increase in slurry volume. However, the deacidification efficiency is good at this temperature, allowing for a moderate reduction in the slurry volume increase. This needs to be addressed by using W... i dynamic equilibrium.

[0072] Temperature-pressure coupling coefficient K T-P =0.3×0.167+0.7×0.4=0.33;

[0073] Dynamic weight W i =α·(C SO2 / C HCl )+(1-α)·K T-P =0.5×2+0.5×0.33=1.165;

[0074] Compared with traditional fuzzy control: Without a weighting mechanism, insufficient slurry production may occur due to the conflict between the rules of "slurry production due to excessive pollutants" and "reducing slurry production at appropriate temperature", thereby prolonging the time of exceeding the standard.

[0075] Advantages of this invention: By calculating the dynamic weight Wi, the "Zhengda" slurry increase can be forcibly triggered. Combined with the correction by the prediction module, the system response speed can be effectively improved, solving the problem of fixed rule weights in traditional fuzzy control and avoiding system control delay or system oscillation caused by the simultaneous triggering of "slurry increase" and "slurry decrease" rules.

[0076] Step S103: Utilize the rule weight W i After fuzzy rule matching, the initial control quantity ΔQ is output. slurry , where ΔQ slurry This refers to the change in the flow rate of the (hydrated lime) slurry.

[0077] Step S104: Perform predictive-energy efficiency optimization. Solve for the optimal control quantity through quadratic programming using the objective function and constraints, and output the optimized command to the actuator.

[0078] This step involves a prediction-energy efficiency integrated optimization algorithm.

[0079] like Figure 4 As shown, in an optional embodiment, step S104 includes:

[0080] Step S1041: Based on real-time data and historical data, use the prediction module to output the predicted values ​​of HCl and SO2 concentrations within a future preset time period;

[0081] In this step, based on sampling of real-time and historical data, the prediction module outputs predicted values ​​for HCl / SO2 concentrations within a preset future time period (e.g., the next 5 minutes). The construction of the prediction module is a standard technique in this field and will not be elaborated upon here. Both real-time and historical data can include HCl concentration, SO2 concentration, temperature, and pressure.

[0082] Step S1042: Construct the objective function;

[0083] Preferably, in step S1042, the objective function is:

[0084] J=Σe 2 +λ·ΣQ slurry

[0085] Where e represents the deviation of the pollutant (HCl / SO2) concentration, which is the difference between the actual measured pollutant concentration and the set pollutant concentration.

[0086] The significance of summation by squares: We square the deviation *e* and then sum the squared deviation values ​​at all times. This is done to emphasize the impact of larger deviations on the objective function, because the larger the deviation, the larger its square value, and therefore its greater weight in the summation. Simultaneously, the squaring operation ensures that all deviation terms are non-negative, preventing positive and negative deviations from canceling each other out, thus allowing for a more accurate measurement of the overall deviation.

[0087] λ represents the energy efficiency weight, with a default value of 0.3. Its function is to adjust the relative importance between the two objectives of emission compliance and cost control, and it can be dynamically adjusted (0.1 to 0.5) according to emission requirements. Impact of value: A smaller λ value indicates a greater emphasis on emission compliance, meaning the goal is to get the pollutant concentration as close to the set value as possible with minimal deviation. A larger λ value, on the other hand, emphasizes cost control, aiming to minimize the amount of quicklime slurry used.

[0088] Q slurry This represents the flow rate of the slaked lime slurry. The flow rate directly affects the treatment cost and effectiveness. ΣQ slurry It involves summing up the flow rates of quicklime slurry at various points in time. This summation reflects the total amount of quicklime slurry used throughout the entire control cycle, which is an important metric for measuring treatment costs.

[0089] Step S1043: Solve for the optimal ΔQ using quadratic programming. slurry ;

[0090] In this step, the optimal ΔQ is solved using quadratic programming. slurry (Change in slurry flow rate, i.e., the increment of slurry flow rate that needs to be adjusted).

[0091] Preferably, in step S1043, the quadratic programming formula is:

[0092]

[0093] Among them, e HCL,te represents the deviation of HCl concentration at time t. SO2,t Q represents the deviation of SO2 concentration at time t, λ is the energy efficiency weight, and Q slurry,t ΔQ represents the slurry flow rate at time t. slurry It is the change in this flow rate, i.e. (Q) slurry,t =Q slurry,t-1 +ΔQ slurry N is the number of time steps in the prediction or optimization time domain, i.e., the number of discrete time points within the future preset time period.

[0094] Step S1044: Output optimization instructions under pre-set constraints.

[0095] In this step, optimization commands are output under constraints to adjust the slurry regulating valve.

[0096] Preferably, in step S1044, the pre-set constraint conditions are as follows:

[0097] Emission limit: HCl ≤ 10 mg / m³ 3 SO2 ≤ 50 mg / m³ 3 ;

[0098] Equipment limitations: Maximum flow rate Q of slurry pump max =500 l / min; the change in slurry flow rate ΔQslurry,t≤2.8%·Qslurry,t-1 (t≥2), that is, the change in slurry flow rate / rate does not exceed 2.8% of the flow rate at the previous moment.

[0099] This allows for dual-objective optimization, clearly demonstrating the balance between emission deviation and lime slaked cost, as well as the dynamic adjustment mechanism of the λ parameter, highlighting the synergistic optimization of economic efficiency and environmental protection.

[0100] Step S105: Update the rule base according to the current system status and the pre-set reward function.

[0101] This step involves reinforcement learning rule base algorithms. For example... Figure 5 As shown, in an optional embodiment, step S105 includes:

[0102] Step S1051: Initialize the fuzzy rule base and Q table;

[0103] In this step, after the process begins, the fuzzy rule base and Q-table are initialized first. The fuzzy rule base is a set of rules based on expert experience and initial settings, and the Q-table is used to store the value estimate for each state-action pair.

[0104] Step S1052: Quantify the current control state (such as emission deviation, energy consumption) into a state value S, and continuously acquire the current system state S;

[0105] Preferably, in step S1052, the formula for the current system state S is:

[0106]

[0107] State S consists of multiple normalized variables, where e HCl and e SO2 These are the concentration deviations of HCl and SO2, respectively, Q. slurry It is the slurry flow rate, ΔQ slurry It is the change in slurry flow rate, C max and Q max These are the maximum values ​​of the corresponding variables.

[0108] Step S1053: Determine whether the termination condition has been met. If the termination condition has not been met, proceed to step S1054. If the termination condition has been met, output the optimized fuzzy rule base and end the process.

[0109] In this step, it is determined whether the preset termination conditions have been met, such as reaching the maximum number of iterations or system state convergence. If the termination conditions are met, the optimized fuzzy rule base is output and the process ends; otherwise, execution continues.

[0110] Step S1054: Generate control action A based on the current rule base;

[0111] Step S1055: Execute the selected action A, the system state changes to a new state S', and calculate the reward R based on the state change and system performance;

[0112] Preferably, in step S1055, the reward function formula is:

[0113]

[0114] Among them, C meas For actual measured concentration, C lim For emission limits; Cost actual Cost is the actual cost. ref The reference cost is oscillation_penalty, which represents the system's oscillation behavior.

[0115] The reward R takes into account the emission compliance status (C) meas For actual measured concentration, C lim (emission limits), cost consumption (Cost) actual Cost is the actual cost. ref As a reference cost and considering the system's oscillation penalty, this formula optimizes the system's safe operation (away from C) by maximizing R. limCost control (below Cost) ref Performance in three aspects: stability (reducing oscillations).

[0116] Among them, performance and safety items

[0117] Objective: To evaluate the measured value C meas Relative to the limit value C lim Safety margin.

[0118] Effect: When C meas Approaching or exceeding C lim When scores drop significantly (or even become negative), it is encouraged to stay away from the extreme values ​​to ensure safety.

[0119] Weight: 10 (highest), highlighting the priority of security or stability.

[0120] Cost-benefit items

[0121] Objective: To measure the difference between actual costs and reference costs.

[0122] Effect: When the actual cost is lower than the reference value, the score increases; conversely, the score decreases, thus encouraging cost-saving.

[0123] Weight: 5, less important than performance and safety items.

[0124] Stability penalty: -2 oscillation_penalty

[0125] Objective: To suppress system oscillations or instabilities.

[0126] Effect: The more severe the oscillation, the more the total score will be deducted (e.g., the number of fluctuations, the amplitude, etc.).

[0127] Weight: 2, emphasizing stability but with low priority.

[0128] Step S1056: Update the Q table;

[0129] Preferably, step S1056 includes:

[0130] Update the Q-table based on the reward R and the new state S', and update the Q-value using the Q-learning algorithm, as shown in the following formula:

[0131] Q(S,A)←Q(S,A)+η·[R+γ·maxQ(S',A′)-Q(S,A)]

[0132] Where maxQ(S',A′) is the maximum expected reward for the next state, η is the learning rate, and γ is the discount factor.

[0133] In practice, the learning rate η = 0.1 and the discount factor γ = 0.9 are used to balance immediate rewards and future rewards.

[0134] Step S1057: Update the fuzzy rule base according to the new Q value;

[0135] In this step, the fuzzy rule base is updated based on the updated Q value to optimize the weight or action of the rules.

[0136] Step S1058: Update the current state to the new state S', go to step S1053, and continue the loop.

[0137] The interaction relationship between steps S101-S105 above is as follows:

[0138] (1) Dynamic weighting → Fuzzy control: Provides a rule-based weighting matrix;

[0139] (2) Fuzzy control → Predictive optimization: Transmitting initial control input;

[0140] (3) Predictive optimization → Actuator: Outputs final control command;

[0141] (4) Implementing agencies → reinforcement learning: feedback of emissions data and cost information;

[0142] (5) Reinforcement learning → Fuzzy control: Periodically update the rule base.

[0143] Figure 2 It fully demonstrates the closed-loop control process from data acquisition to control command output and then to rule base optimization, which is the core idea of ​​"multi-variable dynamic collaborative optimization" described in this application.

[0144] In one embodiment of the present invention, the following modifications can be made at the hardware level to achieve the above invention:

[0145] 1. Add a plug-and-play external control box:

[0146] Protocol communication module: Supports industrial protocols such as OPC UA, Modbus, Profinet, and Profibus-DP, and is compatible with common PLC and DCS control systems.

[0147] Virtual simulation unit: Built-in digital twin model, which can simulate control effects offline and does not affect the operation of the original system during debugging.

[0148] In practice, the plug-and-play external control box is connected to the original DCS or PLC system via industrial Ethernet or other communication methods, according to the actual communication protocol.

[0149] The advantage of plug-and-play external control boxes is that they can be seamlessly compatible with old control systems, reduce retrofit costs and risks, and enable plug-and-play control systems with a retrofit cycle of less than 30 days.

[0150] 2. Add testing instruments:

[0151] High-precision in-situ gas analysis instrument: HCl / SO2 laser / ultraviolet spectrometer (measurement range: HCl: 0—2000 mg / Nm3) 3 SO2: 0-1200 mg / Nm 3 ).

[0152] In practice, a laser / ultraviolet spectrometer is installed at the inlet of the deacidification tower.

[0153] 3. Add intelligent actuators:

[0154] Variable frequency pump for quicklime slurry (flow rate adjustment accuracy ±1%).

[0155] Calcium slurry regulating valve and process water regulating valve (response time <0.5s).

[0156] In specific implementation, the framework computing unit is configured as follows: the SD card with the pre-installed lightweight LSTM model is written to JetsonNano, and the sampling period is set to 100ms.

[0157] The following debugging can be performed at the software level:

[0158] Virtual simulation verification: Import historical operating data, simulate a sudden change in HCl / SO2 concentration, and verify that the control response time is less than 2 seconds.

[0159] Dynamic weighting parameter adjustment: Based on the actual HCl / SO2 ratio range (1:2 to 1:5) of the waste incineration plant, the initial weighting coefficient α = 0.6 is set.

[0160] Finally, after running the system, the experimental data obtained are shown in Table 1:

[0161] Table 1 Experimental Data

[0162] index Traditional PID Existing fuzzy control Invention Solution <![CDATA[Average deviation of HCl / SO2]]> ±8.5 ±5.2 ±4.3 Hydrated lime consumption (kg / ton of waste) 12.8 11.7 10.8 Overshoot frequency (times / day) 3.2 1.8 1.5 Response time (seconds) 45 32 23

[0163] Note: The data comes from actual measurements taken at a waste incineration plant in Sichuan (processing capacity of 1,000 tons / day) from March to May 2023. The solution proposed in this invention reduces the operating cost of quicklime by 7.7% (the cost of fly ash treatment will also be reduced, but this was not calculated in this case).

[0164] As shown in Table 1, the average deviation of HCl / SO2 in this invention is smaller than that of traditional PID and existing fuzzy control, resulting in high adjustment accuracy; the slaked lime consumption of this invention is smaller than that of traditional PID and existing fuzzy control, resulting in low operating cost; the overshoot count of this invention is smaller than that of traditional PID and existing fuzzy control, resulting in strong stability; and the response time of this invention is smaller than that of traditional PID and existing fuzzy control, resulting in high sensitivity.

[0165] As can be seen from the above, the present invention specifically solves the following problems:

[0166] Achieve dynamic coordinated control of multiple variables (HCl, SO2, temperature, pressure) to improve stability under complex operating conditions;

[0167] The fuzzy rule base is optimized through a self-learning mechanism to reduce manual intervention;

[0168] Solve the control accuracy problem under conditions of strong coupling of multiple variables and high fluctuation of operating conditions;

[0169] Integrate energy efficiency optimization goals to reduce the amount of quicklime used and energy consumption;

[0170] Seamlessly compatible with legacy control systems, reducing retrofit costs and risks, enabling plug-and-play control systems with a retrofit cycle of less than 30 days.

[0171] In summary, the present invention has the following beneficial effects:

[0172] Improved control precision: By comprehensively utilizing multi-variable dynamic collaborative optimization control, high-precision control of the flue gas desulfurization and deacidification process is achieved, and the system response time is improved by 28.1%.

[0173] Energy saving and consumption reduction: Through multi-variable coordinated control and precise parameter adjustment, the use of quicklime slurry, process water and energy is optimized, reducing operating costs and energy consumption. The operating cost of quicklime is reduced by 7.7%.

[0174] Enhanced system stability: Advanced control algorithms and real-time monitoring mechanisms improve the system's adaptability to changes in operating conditions and disturbances, enhancing the system's stability and reliability, and reducing the number of overshoots in the control system by 16.7%.

[0175] Easy to maintain and expand: The independent external control box design facilitates system installation, maintenance and functional expansion. It can be seamlessly integrated with the original control system, and no production stoppage is required during commissioning. It does not affect the use of the original control system, which is conducive to the promotion and application of the technology.

[0176] The core innovation of this invention lies in: a dynamic weighted collaborative control algorithm (multivariable dynamic collaborative control algorithm) + a prediction-energy efficiency integrated optimization algorithm (energy efficiency target integration method) + a reinforcement learning rule base algorithm (self-learning rule optimization), which is particularly suitable for the highly volatile and strongly coupled environment of waste incineration flue gas desulfurization. Specific details are as follows:

[0177] Multivariable dynamic weighting mechanism: For the first time, the pressure coupling coefficient and the pollutant ratio are combined as weighting factors in the control method, which solves the problem that traditional fuzzy control ignores the influence of operating conditions (such as when the temperature fluctuates by ±30℃, the weight adjustment range of the rules in this invention reaches 35%, while the existing technology does not have this mechanism).

[0178] Energy efficiency and emission dual-objective optimization: By introducing the λ parameter (dynamically adjusted from 0.1 to 0.5), the cost of quicklime is reduced by 11.3% while ensuring emission compliance (verified by data from a waste incineration plant in Sichuan). The energy efficiency optimization objectives are integrated to reduce the amount of quicklime used and energy consumption.

[0179] Reinforcement learning rule base: The reinforcement learning mechanism updates the rule base every 15 minutes, and after long-term operation, the control accuracy is improved by 17.3% (compared to a fixed rule base without a learning mechanism).

[0180] The core innovations and technical effects of this invention are reflected in:

[0181] (I) Domain Innovation of Dynamic Weighted Cooperative Control Algorithm

[0182] 1. Targeted design for multivariable coupling

[0183] In existing technologies, the application of dynamic weighting mechanisms in other fields (such as chemical process control) has not considered the specific coupling relationship (strong nonlinear interaction between temperature, pressure, and pollutant concentration) in waste incineration flue gas desulfurization. This invention, for the first time, incorporates the temperature-pressure coupling coefficient K... T-P Compared with pollutant concentration C SO2 / C HCl Combined as a weighting factor (i.e., formula W) i =α·(C SO2 / C HCl )+(1-α)·K T-P This solves the following problems that traditional methods could not handle:

[0184] 1) Adaptability to extreme working conditions: When the HCl concentration rises sharply (such as the instantaneous change of ±100% caused by the mixing of chlorine-containing plastics), the "slurry enhancement" rule is forcibly triggered by dynamic weights, avoiding the adjustment delay caused by rule conflicts in traditional fuzzy control (see the example in the manual, Wi calculation increases the slurry enhancement rate by 35%).

[0185] 2) Equipment characteristic matching: To address the issue of slurry pump efficiency decreasing over time, the weighting factor reflects changes in operating conditions in real time, compensating for control deviations caused by equipment aging (traditional methods rely on fixed rules and cannot be adaptive).

[0186] 2. Deep integration with waste incineration process parameters

[0187] Temperature range (T) min =120℃, T max =180℃), pressure range (P min =-500Pa, Pmax=0Pa) and the weighting (temperature weight 0.3, pressure weight 0.7) are all calibrated based on actual operating data of the deacidification tower, rather than general parameters. For example, the influence of pressure on atomization efficiency is quantified as a 70% weight, directly addressing the core pain point of "pressure fluctuations causing uneven slurry atomization" in the SDA process. This is a domain-specific optimization, rather than a simple transplantation of the existing weighting mechanism.

[0188] (II) Synergistic effect mechanism of the three algorithms (different from simple combination)

[0189] 1. Hierarchical progression from dynamic weights to fuzzy control to predictive optimization

[0190] Dynamic weights empower fuzzy control: Traditional fuzzy control uses fixed rule weights, leading to rule conflicts under complex operating conditions (such as simultaneous triggering of "high temperature reduction" and "high pollution increase"). This invention addresses this by using a dynamic weight matrix W. i Prioritize the rules to determine the initial control quantity ΔQ output by the fuzzy controller. slurry More closely matches real-time operating conditions (see instruction manual for details) Figure 3 After weight adjustment, the rule matching accuracy improved by 22%.

[0191] Predictive optimization corrects the defects of fuzzy control: The initial control quantity output by fuzzy control is based only on the current state. The prediction module (such as LSTM model) uses historical data to predict the concentration change in the next 5 minutes and performs secondary programming in combination with the energy efficiency objective function (J), which solves the "short-sightedness" problem of fuzzy control (for example, avoiding excessive slurry addition for short-term target achievement, which leads to long-term cost increases).

[0192] Reinforcement learning closed-loop optimization of the rule base: Traditional fuzzy control rule bases rely on human experience and cannot be updated. This invention updates the rule base every 15 minutes using the Q-learning algorithm, adapting to long-term changes such as equipment aging and sensor drift (see experimental data; control accuracy improved by 17.3% after long-term operation), forming a complete closed loop of "real-time control - predictive optimization - rule evolution" (e.g., ...). Figure 2 (As shown).

[0193] 2. Industry-specific aspects of dual-objective optimization and constraints

[0194] Objective function J = Σe 2 +λ·ΣQ slurry The energy efficiency weight λ (dynamically adjusted from 0.1 to 0.5) and constraints (emission limits, slurry pump flow rate change rate ≤ 2.8%) are designed to address the pain points of the waste incineration industry.

[0195] Balancing emission compliance with cost control: The λ parameter allows the system to flexibly switch between stringent environmental requirements (e.g., λ=0.1, focusing on emissions) and economic operation (e.g., λ=0.5, focusing on cost). Compared with traditional methods that only pursue compliance without considering costs (e.g., traditional PID quicklime consumption is 12.8 kg / ton of waste, while this invention reduces it to 10.8 kg / ton, a reduction of 15.6%), it has significant economic advantages.

[0196] The necessity of equipment safety constraints: The constraint of slurry flow rate change rate ≤2.8% prevents mechanical damage caused by frequent and large-scale adjustment of pump valves (traditional methods do not have this constraint, and the measured overshoot times reached 3.2 times / day, while the present invention reduces it to 1.5 times / day, a reduction of 53%), and improves equipment life.

[0197] (iii) The indivisibility of the technical solution

[0198] 1. Dynamic weight calculation depends on the input of the prediction module: K in step S102 T-P Temperature / pressure data needs to be filtered for noise by the prediction module (such as filtering when the sensor signal-to-noise ratio (SNR) is <10dB in high-temperature environments) to ensure the accuracy of weight calculation.

[0199] 2. The reward function in reinforcement learning includes the objective of prediction optimization: the cost term in the reward function. actual The cumulative slurry flow rate ΣQ in direct correlation prediction optimization slurry This forms a closed-loop feedback loop of "control-optimization-learning," none of which can be omitted.

[0200] 3. Industry-specific adaptability: All algorithm parameters (such as temperature and pressure range, λ value, and constraints) are customized based on the characteristics of waste incineration process, rather than parameter adjustments of general control algorithms, reflecting a targeted solution to technical problems in this field.

[0201] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A multivariable dynamic collaborative optimization control method for acid removal from waste incineration flue gas, characterized in that, include: Step S101: Acquire real-time data, preprocess it, and calculate the temperature-pressure coupling coefficient K. T-P ; Step S102: Utilize the preprocessed real-time data and the temperature-pressure coupling coefficient K T-P Calculate the rule weight W i ; Step S103: Utilize the rule weight W i After fuzzy rule matching, the initial control quantity ΔQ is output. slurry , where ΔQ slurry This represents the change in slurry flow rate; Step S104: Perform predictive-energy efficiency optimization. Solve for the optimal control quantity through quadratic programming using the objective function and constraints, and output the optimized command to the actuator. Step S105: Update the rule base according to the current system status and the preset reward function; In step S101, the real-time data includes HCl concentration, Concentration, temperature, and pressure; The temperature-pressure coupling coefficient K T-P The calculation formula is: ， Where T is the real-time temperature value, T min T is the lower limit of temperature. max P represents the upper limit of temperature, and P represents the real-time pressure value. min As the lower limit of pressure, P max This is the upper limit of pressure. In step S102, the rule weight W i The calculation formula is: ， Where α is the adaptive coefficient, according to Concentration ratio is dynamically adjusted. for Compared to the concentration of HCl; In step S104, the objective function is: J = Σe² + λ•ΣQ slurry , Where e represents HCl and Concentration deviation, λ is the energy efficiency weight, Q slurry Represents slurry flow rate; The constraints include emission limits and equipment limits; In step S105, the system state consists of normalized variables, including concentration deviation and slurry flow rate, and the reward function is based on the actual measured concentration, emission limit, cost, and system oscillation.

2. The multivariable dynamic collaborative optimization control method for desulfurization of waste incineration flue gas according to claim 1, characterized in that, Step S104 includes: Step S1041: Based on real-time data and historical data, use the prediction module to output HCl and [other data] for the future preset time period. Concentration prediction; Step S1042: Construct the objective function; Step S1043: Solve for the optimal ΔQ using quadratic programming. slurry ; Step S1044: Output optimization instructions under pre-set constraints; In step S1043, the quadratic programming formula is: Among them, e HCL,t e represents the deviation of HCl concentration at time t. SO2,t Represents time t Concentration deviation, λ is the energy efficiency weight, Q slurry,t denoted as slurry flow rate at time t, and N is the number of time steps in the prediction or optimization time domain; In step S1044, the pre-set constraints are as follows: Emission limit: HCl ≤ 10 mg / m³ ≤50mg / m³; Equipment limitations: Maximum flow rate Q of slurry pump max =500 l / min; the change in slurry flow rate ΔQslurry,t≤2.8%•Qslurry,t-1(t≥2), that is, the change in slurry flow rate does not exceed 2.8% of the flow rate at the previous moment.

3. The multivariable dynamic collaborative optimization control method for desulfurization of waste incineration flue gas according to claim 1, characterized in that, Step S105 includes: Step S1051: Initialize the fuzzy rule base and Q table; Step S1052: Quantize the current control state into a state value S, and continuously acquire the current system state S; Step S1053: Determine whether the termination condition has been met. If the termination condition has not been met, proceed to step S1054. If the termination condition has been met, output the optimized fuzzy rule base and end the process. Step S1054: Generate control action A based on the current rule base; Step S1055: Execute the selected action A, the system state changes to a new state S', and calculate the reward R based on the state change and system performance; Step S1056: Update the Q table; Step S1057: Update the fuzzy rule base according to the new Q value; Step S1058: Update the current state to the new state S', and proceed to step S1053.

4. The multivariable dynamic collaborative optimization control method for desulfurization of waste incineration flue gas according to claim 3, characterized in that, In step S1052, the formula for the current system state S is: ， Among them, e HCl and e SO2 They are HCl and Concentration deviation, Q slurry It is the slurry flow rate, ΔQ slurry It is the change in slurry flow rate, C max and Q max These are the maximum values ​​of the corresponding variables.

5. The multivariable dynamic collaborative optimization control method for desulfurization of waste incineration flue gas according to claim 3, characterized in that, In step S1055, the reward function formula is: ， Among them, C meas For actual measured concentration, C lim For emission limits; Cost actual Cost is the actual cost. ref The reference cost is oscillation_penalty, which represents the system's oscillation behavior.

6. The multivariable dynamic collaborative optimization control method for desulfurization of waste incineration flue gas according to claim 3, characterized in that, Step S1056 includes: Update the Q-table based on the reward R and the new state S', and update the Q-value using the Q-learning algorithm, as shown in the following formula: Q(S,A)←Q(S,A)+η•[R+γ•maxQ(S',A′)-Q(S,A)] Where maxQ(S',A′) is the maximum expected reward for the next state, η is the learning rate, and γ is the discount factor.

Images