Method and system for collaborative control of fuel quantity and blast air quantity of lime rotary kiln based on multi-objective dynamic optimization

Abstract

The application discloses a method and system for collaborative control of fuel quantity and air blast quantity of a lime rotary kiln based on multi-target dynamic optimization. A layered architecture is adopted: the upper layer is triggered according to a first period, three target functions and constraint and preference modes are constructed, an adaptive multi-target particle swarm algorithm is run to output a Pareto frontier, and optimal temperature set values and optimal air-fuel ratio set values are selected; the lower layer is triggered according to a second period, a model predictive controller adopts a two-input two-output dynamic matrix model, a rolling optimization calculation is performed on a control increment, and the control increment is output to a coal injection valve and a fan frequency converter to drive a rotary kiln body, and optimal set values and NOx concentration feedback closed loops are output. The application realizes three-target Pareto collaborative optimization, can improve the qualified rate of activity, and reduce comprehensive energy consumption and NOx emission. The application solves the problem that existing control only focuses on a single temperature index, adopts a fixed air-fuel ratio or a single-loop PID, and cannot take into account three targets of lime activity, comprehensive energy consumption and NOx emission.

Description

Technical Field

[0001] This invention relates to the field of industrial kiln combustion control technology, and in particular to a method and system for coordinated control of fuel quantity and air volume in a lime rotary kiln based on multi-objective dynamic optimization. Background Technology

[0002] Lime kilns are strongly coupled, multivariable systems. Fuel quantity (gas / pulverized coal) provides heat, while air volume provides oxygen and influences the airflow and temperature fields. Together, they determine the calcination temperature, the kiln atmosphere (oxidizing / reducing atmosphere), and the flue gas residence time, thus directly affecting lime activity, unit heat consumption, and the amount of thermal NOx generated. How to synergistically reduce energy consumption and emissions while ensuring quality (lime activity, unit heat consumption, and the amount of thermal NOx generated) has become a pressing problem for the industry.

[0003] Existing similar solutions can be mainly divided into two categories: Fuzzy control based on expert rules (such as fuzzy PID used in some field applications): This approach transforms operator experience into a rule base, for example, "If the temperature and NOx are high, then reduce both fuel and airflow simultaneously." The drawbacks of this method are: the rule base cannot cover all operating conditions, and the rules themselves are often heuristic, failing to guarantee finding the optimal solution. When operating conditions change beyond the defined range of the rules, the control effect deteriorates drastically.

[0004] Combustion control based on single objective optimization: This scheme aims to maximize thermal efficiency and finds the optimal air-fuel ratio through an optimization algorithm.

[0005] Disadvantages: Focusing solely on thermal efficiency may lead to temperatures being controlled at the lower end of the acceptable range in pursuit of minimum heat consumption. This can easily result in underburning (a quality incident) if raw material fluctuations occur. Alternatively, increasing combustion temperature to maximize thermal efficiency can cause NOx levels to spike. It cannot handle conflicts between objectives.

[0006] In summary, the control schemes commonly used in industrial settings currently have the following shortcomings: Limitations of single-loop PID control: Typically, the kiln tail temperature or calcination zone temperature is used as the primary controlled variable, fuel quantity as the manipulated variable, and the air volume is controlled by the ratio of air-fuel ratio to fuel quantity. This approach can only guarantee temperature stability but cannot handle conflicts between multiple objectives. For example, increasing fuel and air volume to improve output may not change the temperature, but NOx emissions may drastically exceed limits.

[0007] Disadvantages of a fixed air-fuel ratio: Traditional ratio control uses a fixed air-fuel ratio coefficient. However, with fluctuations in fuel calorific value, changes in raw material particle size, and kiln aging, the optimal air-fuel ratio (i.e., the excess air coefficient that achieves complete combustion and low NOx generation) is dynamically changing. A fixed ratio inevitably leads to incomplete combustion (black smoke, high heat consumption) or excessive combustion (high NOx, high heat loss) under certain operating conditions. Furthermore, this method lacks a multi-objective coordination mechanism: when making adjustments, on-site operators often prioritize ensuring quality (temperature), followed by preventing environmental violations (reducing airflow and NOx), and only consider energy consumption last. This manual compromise is slow to react, highly volatile, and unable to accurately quantify the optimal balance point between various objectives, resulting in the lime kiln operating under suboptimal conditions for extended periods. Summary of the Invention

[0008] The technical problem this invention aims to solve is to provide a method and system for coordinated control of fuel quantity and air volume in a rotary kiln for quicklime calcination (mainly using a rotary kiln). This method addresses the technical challenge of traditional control strategies, which focus solely on temperature when adjusting fuel quantity and air volume. This approach fails to simultaneously consider the three conflicting objectives of product quality (activity), energy consumption (coal / electricity consumption), and environmental emissions (NOx concentration), resulting in a trade-off between these objectives and an inability to achieve optimal overall operation.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method and system for coordinated control of fuel quantity and air volume in a lime rotary kiln based on multi-objective dynamic optimization is proposed. The method automatically optimizes the setpoints for temperature and air-fuel ratio, and achieves dynamic decoupling of fuel and air volume through model predictive control, so that the rotary kiln always operates at the Pareto optimal frontier that takes into account quality, energy consumption and environmental protection.

[0010] A method for coordinated control of fuel quantity and air volume in a lime rotary kiln based on multi-objective dynamic optimization includes the following steps: S1. Collect real-time operating data, including output, coal calorific value, coal injection rate, primary air volume, secondary air volume, actual calcination zone temperature T_act, actual air-fuel ratio λ_act, and NOx concentration in kiln tail flue gas. S2, Upper-level optimization: Triggered in the first cycle, a multi-objective function is constructed with lime activity, comprehensive energy consumption per ton of lime and NOx concentration in kiln tail flue gas as objectives, and a constraint and preference mode is established; an adaptive multi-objective particle swarm optimization algorithm is run to output the Pareto front, and the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt are selected according to the preference mode; S3, Send to lower-level controller: Triggered every second cycle, the model prediction controller reads T_opt, λ_opt and the current feedback values ​​T_act, λ_act, calculates the control increment of pulverized coal injection and blast volume through rolling optimization, and outputs control commands to the actuator; S4. The actuator adjusts the opening of the pulverized coal injection valve and the frequency of the blower frequency converter according to the control command, decomposes the total air volume command into primary air command and secondary air command, drives the rotary kiln body to run, and returns to step S1.

[0011] In the above technical solution, the constraints in step S2 include the calcination zone temperature boundary, the fan surge boundary, and the range of flue gas oxygen content; the preference modes include quality priority mode, energy saving priority mode, and low nitrogen mode.

[0012] In the above technical solution, the adaptive multi-objective particle swarm algorithm in step S2 introduces an adaptive mutation operator to increase particle diversity when it detects drastic changes in operating conditions.

[0013] Furthermore, the drastic changes in operating conditions include at least one of the following: a change rate in feed rate exceeding 10%, a change rate in coal calorific value exceeding 10%, a change rate in calcination zone temperature exceeding 5% or an absolute change exceeding 50°C, a change rate in air-fuel ratio exceeding 8%, and a change rate in NOx concentration exceeding 15%.

[0014] In the above technical solution, the comprehensive energy consumption per ton of lime in step S2 is calculated by multiplying the coal injection quantity and the calorific value of the coal by the output, and the power consumption of the blower is also calculated.

[0015] Specifically, the heat consumption of the fan is calculated by multiplying the sum of the power of the primary fan, secondary fan and induced draft fan by 3600 kJ / kWh and then dividing by the output.

[0016] In the above technical solution, the lime activity in step S2 is estimated by a soft measurement model based on the actual calcination zone temperature T_act.

[0017] In the above technical solution, the activity of lime is obtained periodically from offline testing to correct the soft measurement model.

[0018] Ideally, the lime activity should be obtained offline every 4 to 8 hours, and the soft measurement model should be corrected using the deviation compensation method or the recursive least squares method.

[0019] Preferably, when the deviation between three consecutive test values ​​and the soft measurement estimate exceeds 8%, the piecewise linear reconstruction of the soft measurement model is triggered.

[0020] In the above technical solution, the model prediction controller in step S3 adopts a two-input two-output dynamic matrix model with pulverized coal injection rate and total blast volume as inputs and actual calcination zone temperature T_act and actual air-fuel ratio λ_act as outputs.

[0021] In the above technical solution, the dynamic matrix model is obtained through step response tests or historical data identification, and includes the hysteresis characteristics of pulverized coal injection rate on actual calcination zone temperature T_act and the coupled influence of total blast volume on actual calcination zone temperature T_act and actual air-fuel ratio λ_act.

[0022] In the above technical solution, the goal of the rolling optimization in step S3 is to minimize the deviation between T_act and T_opt and the deviation between λ_act and λ_opt, while minimizing the control increment magnitude.

[0023] In the above technical solution, the actuator in step S4 includes a pulverized coal injection valve, a primary air fan frequency converter, and a secondary air fan frequency converter.

[0024] A multi-objective dynamic optimization-based coordinated control system for fuel quantity and air volume in a lime rotary kiln, characterized in that it includes: The optimization layer is triggered in each first cycle. It receives real-time operating data, constructs a multi-objective function and constraint and preference pattern with lime activity, comprehensive energy consumption per ton of lime and NOx concentration in kiln tail flue gas as objectives, runs an adaptive multi-objective particle swarm algorithm, outputs the Pareto front, and selects the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt according to the preference pattern. The control layer is used to trigger every second cycle. The model prediction controller reads T_opt, λ_opt and the current feedback values ​​of the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act. It calculates the control increment of the pulverized coal injection and blast volume through rolling optimization and outputs control commands. The actuator, including a pulverized coal injection valve, a primary air fan frequency converter, and a secondary air fan frequency converter, is used to receive the control command, adjust the opening degree of the pulverized coal injection valve and the frequency of the fan frequency converter, and decompose the total air volume command into a primary air command and a secondary air command. The controlled object layer, including the rotary kiln body, is used to receive primary and secondary air and pulverized coal injection, output the actual calcination zone temperature T_act, the actual air-fuel ratio λ_act, and the NOx concentration, and feed them back to the optimization layer and the control layer.

[0025] In the above technical solution, the optimization layer includes: a multi-objective function construction unit, used to construct a multi-objective function with lime activity, comprehensive energy consumption per ton of lime, and NOx concentration in kiln tail flue gas as objectives; a constraint and preference mode unit, used to set constraint and preference modes; an adaptive multi-objective particle swarm algorithm unit, used to run the adaptive multi-objective particle swarm algorithm; and a Pareto front output unit, used to output the Pareto front and select T_opt and λ_opt according to the preference mode.

[0026] In the above technical solution, the control layer includes: a model prediction controller for reading T_opt, λ_opt, T_act, and λ_act; a rolling optimization unit for calculating the control increments of pulverized coal injection and blasting volume; and a control command output unit for outputting control commands to the actuator.

[0027] In the above technical solution, the optimization layer further includes a soft measurement unit for estimating lime activity based on the actual calcination zone temperature T_act; and a feedback correction unit for periodically acquiring offline lime activity data and correcting the soft measurement unit.

[0028] In the above technical solution, the model prediction controller adopts a two-input two-output dynamic matrix model. The inputs are the pulverized coal injection rate and the total blast volume, and the outputs are the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act.

[0029] The preferred first cycle is 20 or 30 minutes, and the second cycle is 1 minute or 1.5 minutes.

[0030] In summary, this invention discloses a method and system for coordinated control of fuel quantity and blast volume in a lime rotary kiln based on multi-objective dynamic optimization. It adopts a hierarchical architecture: the upper layer is triggered in the first cycle, constructing a three-objective function and constraint and preference mode, running an adaptive multi-objective particle swarm optimization algorithm to output the Pareto front, and selecting the optimal temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt; the lower layer is triggered in the second cycle, with a model predictive controller using a two-input, two-output dynamic matrix model, continuously optimizing and calculating the control increment, outputting to the pulverized coal injection valve and the blower frequency converter to drive the rotary kiln body, and outputting T_act, λ_act, and NOx concentration feedback closed loop.

[0031] The upper layer is a multi-objective dynamic optimization layer. Using an evolutionary algorithm, under real-time constraints, it calculates the optimal calcination temperature setpoint and the optimal air-fuel ratio setpoint through rolling optimization with the objectives of product quality (activity), energy consumption (heat consumption), and environmental protection (NOx). The lower layer is a model prediction and collaborative control layer. It receives the setpoints from the upper layer, performs decoupled collaborative control of fuel quantity and air volume, quickly tracks the setpoints, and suppresses disturbances.

[0032] Multi-objective optimization function and constraint model: Construct a comprehensive evaluation function with lime activity (based on temperature soft measurement or laboratory feedback), comprehensive energy consumption per ton of lime (converted to coal consumption + electricity consumption), and NOx concentration in kiln tail flue gas as optimization objectives. Simultaneously, establish key process constraint models, including: upper and lower limits of calcination zone temperature, upper limit of kiln tail temperature (to prevent equipment damage), blower surge boundary, and range of oxygen content in flue gas.

[0033] A dynamic optimization mechanism based on evolutionary algorithms: Considering the time-varying operating conditions of lime kilns (such as changes in feed rate and coal calorific value), an adaptive multi-objective particle swarm optimization algorithm (AMOPSO) or a decomposition-based multi-objective evolutionary algorithm (MOEA / D) is designed. This algorithm can quickly search for a set of Pareto optimal solutions satisfying the constraints within the feasible region using real-time data in each control cycle (e.g., every 15-30 minutes). Based on preset preferences (e.g., "quality priority," "energy saving priority," or "environmental protection priority"), it selects the final "compromise optimal solution" as the setpoint for the next layer. The key points are the algorithm's speed and online adaptive capability.

[0034] Fuel-airflow dynamic decoupled coordinated control: The upper layer provides the target values ​​(T_set, λ_set, where λ is the air-fuel ratio). The lower-level controller utilizes the multivariate processing capabilities of Model Predictive Control (MPC) to construct a two-input, two-output predictive model for fuel quantity and blower volume. Through MPC's rolling optimization, it automatically calculates control commands for fuel valve position and blower speed (or valve opening), achieving dynamic coordination between the two and avoiding drastic fluctuations in one variable caused by adjusting the other.

[0035] This invention can automatically find the optimal matching relationship between fuel quantity and air volume based on real-time operating conditions, so that the lime kiln always operates on the "Pareto optimal" frontier that takes into account quality, energy consumption and environmental protection, thereby maximizing comprehensive benefits.

[0036] Compared with the prior art, the present invention has the following specific beneficial effects: (1) Achieving optimal comprehensive indicators (Pareto improvement): Integrating previously conflicting quality, energy consumption, and environmental protection indicators into a unified optimization framework. Practical applications show that, while ensuring that the activity qualification rate does not decrease or even increases, NOx emission concentration can be reduced by 10%-20%, while reducing comprehensive energy consumption by 3%-5%. This truly achieves the synergy of quality improvement, cost reduction, and carbon reduction.

[0037] (2) Strong adaptability to operating conditions: The multi-objective optimization algorithm can automatically adjust the air-fuel ratio and temperature setpoint according to the decrease in fuel calorific value, or automatically switch to low emission mode when environmental pressure is high (such as production restrictions in autumn and winter), and control NOx at a lower level without seriously sacrificing quality. This flexibility is unmatched by traditional control.

[0038] (3) Reduce the burden on operators and realize intelligent decision-making: The system automatically completes complex multi-objective trade-off calculations, freeing operators from tedious manual balancing. They only need to set the preference mode according to the company's strategy, thus realizing intelligent control decision-making.

[0039] (4) A creative combination of layered architecture is realized: the multi-objective evolutionary algorithm is combined with the MPC control model, and the timing separation and functional decoupling are achieved through the intermediate variable of "set value", rather than simple serial or parallel superposition, which solves the inherent conflict between the two algorithms.

[0040] (5) Unconventional selection of decision variables: The “calcination zone temperature setpoint” and “air-fuel ratio setpoint” are used as the only decision variables at the upper level, rather than directly optimizing the control quantity. It is recognized that these two parameters are the strategic hub of the coupled combustion system and are a creative discovery of the process mechanism.

[0041] (6) Dedicated design of 2×2 dynamic matrix MPC: The air-fuel ratio is transformed from an algebraic calculation to an independent controlled variable. A dedicated dynamic matrix model is constructed to describe the cross-coupling of fuel-air volume with temperature-air-fuel ratio, realizing implicit dynamic decoupling and breaking through the coupling oscillation of traditional ratio control.

[0042] (7) Pareto frontier preservation and preference selection: reject the weighted scalar simplification of existing multi-objective evolutionary algorithms, insist on preserving the Pareto non-dominated solution set, and design an operating mode preference switching mechanism to achieve strategic-level human-machine collaborative decision-making.

[0043] (8) Activity Dual Time Scale Closed Loop: In response to the industry pain point of long quality inspection cycle of lime kiln, an innovative design of soft measurement real-time prediction + offline test periodic correction mechanism is used to fill the gap in closed loop control of quality indicators.

[0044] The aforementioned inventive discoveries and designs enable this invention to achieve significant results simultaneously in three dimensions: improved activity qualification rate, reduced overall energy consumption, and reduced NOx emissions. It also supports adaptive operating conditions, flexible mode switching, and robust long-term models. Compared with existing technologies, it has outstanding substantive features and significant progress, and has strong industry applicability. Attached Figure Description

[0045] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a schematic diagram of the structure of the coordinated control system for fuel quantity and air volume of a lime rotary kiln based on multi-objective dynamic optimization according to the present invention.

[0046] Figure 2 This is a control flowchart of the method for coordinated control of fuel quantity and air volume in a lime rotary kiln based on multi-objective dynamic optimization, as described in this invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0048] like Figure 1 As shown, this is a multi-objective dynamic optimization-based coordinated control system for fuel quantity and air volume in a lime kiln, implemented according to the present invention.

[0049] I. Overall Architecture like Figure 1 As shown, this system is a multi-objective dynamic optimization-based coordinated control system for fuel quantity and air volume in a lime rotary kiln. It adopts a hierarchical closed-loop architecture, comprising four functional modules from top to bottom: an optimization layer, a control layer, an actuator layer, and a controlled object layer. The modules interact via an industrial communication network, forming a complete control closed loop of "slow strategic optimization—rapid tactical tracking—precise execution—process feedback."

[0050] II. Optimization Layer The optimization layer is the strategic decision-making layer of the system, which is triggered once every 20 or 30 minutes or a set period. It receives real-time operating data from the controlled object layer, constructs a multi-objective function and constraint and preference mode with lime activity, comprehensive energy consumption per ton of lime and NOx concentration in kiln tail flue gas as objectives, runs an adaptive multi-objective particle swarm algorithm, outputs the Pareto front, selects the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt according to the preference mode, and sends them down to the control layer.

[0051] The optimization layer specifically includes the following functional units: (1) Multi-objective function construction unit This unit is used to construct a three-objective optimization function aimed at maximizing lime activity, minimizing the overall energy consumption per ton of lime, and minimizing the NOx concentration in the kiln tail flue gas. Lime activity is predicted using a soft sensor model based on the actual calcination zone temperature. This model establishes a steady-state mapping relationship between calcination zone temperature and lime activity, obtained through regression fitting using historical operating data. The overall energy consumption per ton of lime is calculated by multiplying the pulverized coal injection rate and the calorific value of the coal by the output, and the power consumption of the primary and secondary air blowers is also factored in. NOx concentration is predicted using a NOx formation mechanism model. This model takes the calcination zone temperature and air-fuel ratio as inputs and is based on a thermal NOx formation mechanism, or a data-driven model can be used instead. The three objective functions are independent of each other and together constitute a vector optimization problem.

[0052] (2) Constraint and Preference Pattern Unit This unit is used to set process constraints and operating mode preferences. Process constraints include hard constraints and soft constraints: hard constraints include the upper and lower limits of calcination zone temperature (equipment safety and quality boundaries, preventing over-burning or under-burning), the blower surge boundary (preventing primary or secondary blowers from entering the stall zone), and the range of flue gas oxygen content (ensuring complete fuel combustion, preventing CO generation, and suppressing NOx increases caused by excessive oxygen combustion); soft constraints include limits on the rate of change of pulverized coal injection and the rate of change of air volume, etc. Operating mode preferences include a quality-first mode (prioritizing solutions with high activity in the Pareto front, suitable for periods with high order quality requirements), an energy-saving-first mode (prioritizing solutions with low overall energy consumption, suitable for periods with high fuel costs), and a low-NOx mode (prioritizing solutions with low NOx emissions, suitable for periods of environmental protection-related production restrictions or periods with ultra-low emission requirements in autumn and winter).

[0053] (3) Adaptive multi-objective particle swarm algorithm unit This unit is used to run the Adaptive Multi-Objective Particle Swarm Optimization (AMOPSO) algorithm or the Decomposition-Based Multi-Objective Evolutionary Algorithm (MOEA / D). The algorithm initializes a swarm of particles, each representing a set of candidate solutions, i.e., a combination of calcination zone temperature setpoints and air-fuel ratio setpoints. Using current operating data, the objective function model is substituted to calculate three objective values ​​for each particle. The individual and global optima of the particles are updated according to the Pareto dominance relationship, forming the Pareto front. When drastic changes in operating conditions are detected (such as sudden changes in feed rate exceeding a set threshold, coal calorific value fluctuations exceeding ±10%, or kiln lining collapse causing sudden temperature changes exceeding ±30°C), the particle mutation probability is adaptively increased, introducing random perturbations to improve population diversity and avoid the algorithm getting trapped in local optima. After the algorithm iterates to convergence or reaches the maximum number of iterations, it outputs the Pareto optimal solution set.

[0054] (4) Pareto Leading Line Output Unit This unit receives the Pareto front non-dominated solution set output by the adaptive multi-objective particle swarm optimization unit and selects the compromise optimal solution from the Pareto front using either the TOPSIS method (a ranking method based on ideal solution similarity) or a linear weighted method, based on the operating mode preference selected by the constraint and preference mode unit. The TOPSIS method calculates the Euclidean distance between each solution and the positive ideal solution (optimal value of each objective) and the negative ideal solution (worst value of each objective), selecting the solution closest to the positive ideal solution and farthest from the negative ideal solution as the compromise optimal solution. The output results are the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt, which are transmitted to the model predictive controller in the control layer via the industrial communication network.

[0055] III. Control Layer The control layer is the tactical execution layer of the system, which is triggered once every 1 minute, 1.5 minutes or a set period. The model prediction controller reads the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt issued by the optimization layer, as well as the current feedback values ​​of the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act from the controlled object layer. It calculates the control increment of pulverized coal injection and blast volume through rolling optimization and outputs control commands to the actuator.

[0056] The control layer specifically includes the following functional units: (1) Model predictive controller This controller reads the optimal calcination zone temperature setpoint T_opt and optimal air-fuel ratio setpoint λ_opt from the upper optimization layer, as well as the current feedback values ​​from the controlled object layer: the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act. The model predictive controller employs a two-input, two-output dynamic matrix model. The two input variables are the pulverized coal injection rate and the total air volume, and the two output variables are the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act. The dynamic matrix model is obtained through step response experiments or historical data identification: step perturbations are applied to the pulverized coal injection rate and the total air volume, and the response curves of the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act are recorded. The step response coefficients of each channel are obtained by fitting using the least squares method and assembled into a dynamic matrix. The model includes the lag characteristics of the pulverized coal injection rate on the actual calcination zone temperature T_act (there is a lag of several minutes between the fuel being injected from the burner and the heat being transferred to the calcination zone) and the coupled effects of the total air volume on the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act (the change in air volume affects both the temperature distribution and the oxygen concentration distribution).

[0057] (2) Rolling optimization unit This unit is used to solve a constrained optimization problem based on the dynamic matrix model of the model predictive controller within each 1-minute control cycle. The objective function is to minimize the deviation between the predicted output and the setpoint, i.e., to minimize the deviation between T_act and T_opt, and the deviation between λ_act and λ_opt, while minimizing the control increment amplitude to suppress frequent actuator actions. Constraints include pulverized coal injection valve position boundaries (valve opening 0-100%), blower speed boundaries (inverter frequency upper and lower limits), and control increment rate limits (pulverized coal injection changes and air volume changes do not exceed set thresholds in adjacent cycles). The unit calculates the sequence of pulverized coal injection control increments and blower volume control increments to be applied in the future prediction time domain using quadratic programming or least squares methods. Only the first control increment is implemented, and the optimization is rolled over again in the next cycle.

[0058] (3) Control command output unit This unit converts the incremental control values ​​of pulverized coal injection and blast air volume calculated by the rolling optimization unit into actual control commands. The pulverized coal injection command is directly output to the pulverized coal injection valve; the total blast air volume command is further decomposed into primary air and secondary air commands, with the decomposition ratio determined based on the rotary kiln's combustion characteristics (primary air mainly adjusts the flame shape, accounting for approximately 15-25%; secondary air mainly provides combustion oxygen, accounting for approximately 75-85%), and output to the primary air fan frequency converter and secondary air fan frequency converter respectively. Control commands are issued via an industrial control network (such as Profinet or Ethernet / IP).

[0059] IV. Implementing Agency As the execution terminal of the system, the actuator includes a pulverized coal injection valve, a primary air fan frequency converter, and a secondary air fan frequency converter. It is used to receive control commands from the control layer, adjust the opening degree of the pulverized coal injection valve and the frequency of the fan frequency converter, decompose the total air volume command into primary air command and secondary air command, and supply pulverized coal and combustion air to the rotary kiln body of the controlled object layer.

[0060] (1) Pulverized coal injection valve The pulverized coal injection valve is installed at the end of the pulverized coal conveying pipeline and before the burner inlet. It receives pulverized coal injection commands from the control layer and adjusts the valve opening via an electric actuator to control the amount of pulverized coal injected into the rotary kiln burner per unit time. A calibrated flow characteristic curve exists between the valve opening and the pulverized coal flow rate. The control layer calculates the required valve opening based on the target pulverized coal injection rate.

[0061] (2) Primary air fan frequency converter The primary air fan is driven by a frequency converter. It receives primary air commands from the control layer and adjusts the fan speed by regulating the frequency converter's output frequency, thereby controlling the primary air volume. Primary air is delivered through the central duct of the burner, primarily for conveying pulverized coal, adjusting flame shape, and regulating oxygen concentration near the burner. Primary air accounts for 15-25% of the total blower volume.

[0062] (3) Secondary air fan frequency converter The secondary air fan is driven by a frequency converter. Receiving secondary air commands from the control layer, the frequency converter adjusts the fan speed to control the secondary air volume. The secondary air is then delivered into the kiln via a cooler or kiln hood. Its main functions are to provide oxygen for fuel combustion and to influence the temperature distribution and airflow within the kiln. Secondary air accounts for 75-85% of the total blower volume.

[0063] V. Controlled Object Layer The controlled object layer includes the rotary kiln body, which receives primary air, secondary air and pulverized coal from the actuator, performs fuel combustion and limestone calcination reaction, outputs the actual calcination zone temperature T_act, the actual air-fuel ratio λ_act and NOx concentration, and feeds them back to the optimization layer and the control layer.

[0064] (1) Rotary kiln body The rotary kiln body is the main equipment of the lime rotary kiln, including an inclined kiln shell, a kiln head burner, a kiln tail flue, support devices, and a transmission device. The kiln shell is lined with refractory material and rotates at a certain speed, causing the limestone material to tumble and advance within the kiln. The burner is installed at the kiln head, injecting a mixture of pulverized coal and primary air into the kiln for combustion, forming a high-temperature flame and calcination zone. Secondary air enters from the kiln head or cooler, mixes with the flue gas, and flows towards the kiln tail. Limestone enters from the kiln tail, undergoes preheating, calcination, and cooling processes within the kiln, decomposing into calcium oxide (active lime) and carbon dioxide, and is discharged from the kiln head.

[0065] (2) Sensor group The sensor array is distributed at key locations throughout the rotary kiln body to collect process parameters in real time. This includes: a calcination zone temperature sensor (using thermocouples or infrared thermometers, installed on the side of the kiln shell or through the kiln head observation hole, directly measuring or using soft measurement to obtain T_act); a kiln tail temperature sensor (thermocouple, installed in the kiln tail flue); a flue gas NOx concentration analyzer (CEMS system, installed in the kiln tail chimney, monitoring NOx concentration online); a primary air flow meter (installed in the primary air fan outlet pipe); a secondary air flow meter (installed in the secondary air fan outlet pipe); a pulverized coal flow meter (installed in the pulverized coal conveying pipe); a pulverized coal calorific value analyzer (offline testing or online soft measurement); and a feed rate metering device (installed at the kiln tail feed end).

[0066] (3) Feedback loop The feedback loop is implemented through an industrial communication network (such as Profinet, Ethernet / IP, or OPC-UA). The actual calcination zone temperature T_act and the actual air-fuel ratio λ_act (calculated from primary air volume, secondary air volume, and pulverized coal injection rate) are fed back to the control layer in real time as feedback values ​​for the rolling optimization of the model predictive controller. Real-time operating data (including output, calorific value of coal, pulverized coal injection rate, primary air volume, secondary air volume, T_act, λ_act, NOx concentration, etc.) are periodically summarized and fed back to the optimization layer as input for the multi-objective dynamic optimizer to update the objective function and constraints. The feedback loop forms a closed loop, enabling the system to continuously optimize based on actual operating conditions.

[0067] VI. Signal connection relationships between layers The signal connection relationships between the optimization layer, control layer, actuator layer, and controlled object layer are as follows: (1) Signal connection from optimization layer to control layer The optimization layer sends the optimal calcination zone temperature setpoint T_opt and optimal air-fuel ratio setpoint λ_opt, obtained every 20 minutes, to the model predictive controller in the control layer via an industrial communication network. This signal is a setpoint signal, representing the optimal operating target of the system under the current operating conditions.

[0068] (2) Signal connection from the control layer to the actuator The control layer sends the coal injection quantity command calculated every minute to the electric actuator of the coal injection valve through an industrial control network (such as the analog output module of a PLC), and sends the primary air command to the primary air fan frequency converter and the secondary air command to the secondary air fan frequency converter. This signal is a control command signal that directly drives the actuator to operate.

[0069] (3) Signal connection from the actuator to the controlled object layer The actuator supplies pulverized coal, primary air, and secondary air to the rotary kiln body through physical pipelines and equipment. The pulverized coal, regulated by the pulverized coal injection valve, is transported to the burner via pipelines; the airflow generated by the primary and secondary air fans is respectively sent to the kiln head or cooler via pipelines. This connection represents a physical connection of material and energy flow.

[0070] (4) Signal connection from the controlled object layer to the control layer The actual calcination zone temperature T_act (acquired by a temperature sensor) and the actual air-fuel ratio λ_act (calculated by an air flow meter and a pulverized coal flow meter) output by the rotary kiln body are fed back to the model predictive controller in the control layer in real time via sensor signal lines and a data acquisition module. This signal is a feedback signal used for MPC rolling optimization to calculate deviations.

[0071] (5) Signal connection from the controlled object layer to the optimization layer The real-time operating data output by the rotary kiln (including output, calorific value of coal, pulverized coal injection rate, primary air volume, secondary air volume, T_act, λ_act, NOx concentration, etc.) is periodically aggregated and fed back to the multi-objective dynamic optimizer in the optimization layer through a data acquisition system and industrial communication network. This signal is the operating data signal, used to update the optimization model and constraints.

[0072] VII. System Workflow The above four-layer structure constitutes a hierarchical closed-loop control system, and its workflow is as follows: Figure 2 As shown: The optimization layer receives real-time operating data every 20 minutes or at a set interval, constructs a three-objective optimization function and constraints, runs an adaptive multi-objective particle swarm optimization algorithm to output the Pareto front, selects the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt according to the preference mode, and sends them to the control layer. The control layer reads T_opt, λ_opt and the current feedback values ​​T_act, λ_act every minute, performs rolling optimization based on a two-input two-output dynamic matrix model, calculates the control increment of pulverized coal injection and air volume, and outputs pulverized coal injection command, primary air command and secondary air command to the actuator. The actuator adjusts the opening of the pulverized coal injection valve and the frequency of the blower frequency converter to supply pulverized coal and combustion air to the rotary kiln body. The rotary kiln body performs combustion and calcination reactions, outputs T_act, λ_act and NOx concentration, and returns them to the optimization layer and control layer through the feedback loop to enter the next control cycle.

[0073] This layered architecture achieves temporal separation between strategic-level multi-objective optimization and tactical-level dynamic decoupled tracking: the upper layer searches for the global optimum in a 20-minute cycle to adapt to slowly changing operating conditions; the lower layer quickly suppresses disturbances in a 1-minute cycle to ensure tracking accuracy. The two layers work together to ensure that the lime rotary kiln always operates at the Pareto optimal frontier, balancing quality, energy consumption, and environmental protection.

[0074] Example 2 The invention will now be described in detail with reference to a specific implementation example. This embodiment is applied to a rotary kiln with a daily output of 600 tons of active lime.

[0075] The control system architecture of this embodiment is as follows: the system consists of four parts: a data acquisition and preprocessing module, a multi-objective dynamic optimizer, a model prediction co-controller, and an actuator.

[0076] The implementation steps of the method in this embodiment are as follows: Step S1: Data acquisition and status awareness.

[0077] S11 Real-time Data Acquisition: Coal Injection Rate (F_fuel), Primary Air Volume (F_pri), Secondary Air Volume (F_sec), Total Blower Volume (F_air=F_pri+F_sec), Kiln Tail Temperature (T_tail, obtained by thermocouple), Calcination Zone Temperature (T_burn, which can be obtained by soft measurement or calculated based on infrared thermometry), Kiln Tail Flue Gas NOx Concentration (C_nox, obtained by CEMS system), Feed Rate (F_feed), Pulverized Coal Calorific Value (Q_fuel, offline laboratory manual input or soft measurement).

[0078] S12 Calculate the current heat consumption per ton of lime: E_cons=(F_fuel×Q_fuel) / F_feed.

[0079] Step S2: Multi-objective dynamic optimization (upper level, periodic operation, e.g., T1=20min).

[0080] S21 Build an optimization model: Decision variables: X=[T_sp,λ_sp], which are the setpoints for the calcination zone temperature and the total air-fuel ratio.

[0081] Objective function (minimization): f1(T,λ) = -activity (or converted to minimizing the deviation from the target activity, which is predicted by a temperature soft sensing model). f2(T,λ) = Comprehensive energy consumption E_cons (estimated by steady-state model, taking into account coal consumption and wind turbine power consumption); f3(T,λ) = NOx concentration C_nox (estimated by a NOx generation mechanism model or data model, with T and λ as inputs).

[0082] Constraints: T_min≤T_sp≤T_max (Equipment safety and quality boundary); λ_min≤λ_sp≤λ_max (combustion stability boundary, to prevent flameout or blow-off); O2_min≤Residual oxygen in flue gas≤O2_max (to ensure complete combustion and prevent CO formation); S22 Adaptive Multi-Objective Particle Swarm Optimization (AMOPSO): (1) Initialize a group of particles, each particle representing a set of candidate solutions [T_sp, λ_sp].

[0083] (2) Substitute the current operating data (such as the current feed amount F_feed and the calorific value of coal Q_fuel) into the objective function model to calculate the three objective values ​​for each particle.

[0084] (3) Update the individual optimality and global optimality (Pareto front) of the particle according to the Pareto dominance relationship.

[0085] (4) Introduce an adaptive mutation operator to increase particle diversity when drastic changes in operating conditions are detected (such as sudden changes in feed rate or sudden temperature changes caused by kiln skin collapse) to avoid getting trapped in local optima.

[0086] S23 Preference Selection and Output: After iteration, a set of Pareto front solutions were obtained.

[0087] Based on the operator's preset "operating mode" (e.g., normal mode, energy-saving mode, low-nitrogen mode), the TOPSIS method or linear weighted method is used to select a final compromise optimal solution [T_opt, λ_opt] from the Pareto front and send it to the lower-level controller.

[0088] Step S3: Fuel-air volume coordinated control (lower layer, fast cycle, e.g., T2=1min).

[0089] S31 constructs an MPC prediction model: Establish a 2x2 dynamic matrix model (identifiable through step response tests or historical data) with pulverized coal injection setpoints and total blast volume setpoints as inputs, and actual calcination zone temperature and actual air-fuel ratio as outputs. The model must include the hysteresis characteristics of fuel quantity with temperature, as well as the coupled effects of blast volume on temperature and air-fuel ratio.

[0090] S32 scrolling optimization: (1) Read the target values ​​T_opt and λ_opt sent down from the upper layer.

[0091] (2) Read the current feedback value T_act and the calculated actual air-fuel ratio λ_act=F_air / F_fuel.

[0092] (3) The MPC controller solves a constrained optimization problem and calculates the incremental sequence of coal injection control ΔF_fuel and the incremental sequence of air blowing control ΔF_air to be applied in the future time domain. The goal is to make the two controlled variables smoothly track the set value while minimizing the magnitude of the adjustment action.

[0093] S33 Decoupling Output: The core advantage of the MPC controller lies in its internal model, which automatically incorporates the coupling relationship between fuel and air volume. Therefore, the calculated ΔF_fuel and ΔF_air are coordinated. For example, when increasing the pulverized coal injection rate, it automatically calculates how much air volume needs to be increased to maintain the target air-fuel ratio and acts accordingly, achieving dynamic decoupling. Furthermore, for rotary kilns, the total blast volume command can be further decomposed into primary air commands (primarily adjusting flame shape) and secondary air commands (primarily supplying oxygen), implemented through simple proportional or rule-based allocation.

[0094] The final control commands are sent to the pulverized coal injection pipe regulating valve and the blower frequency converter (or damper).

[0095] Step S4: Feedback correction.

[0096] Periodically (e.g., every 4-8 hours), the activity of lime is obtained from offline testing and used to correct the activity prediction model in step S21, forming a closed-loop correction to prevent model drift.

[0097] Commonly used correction methods include: (1) Deviation compensation method: Calculate the average deviation Δ between the most recent 3-5 test values ​​and the soft measurement prediction, and add it to the model output; (2) Recursive least squares method: When the deviation exceeds 3%, recursively update the model regression parameters a and b, and take the forgetting factor as 0.95-0.99; (3) Model reconstruction: When the deviation exceeds 8% for three consecutive times, the model is judged to be invalid, and the piecewise linear model is retrained using the most recent 30-50 sets of data. After verifying that R²>0.85, it is put into operation. A closed-loop correction is formed to prevent model drift.

[0098] The above embodiments were demonstrated in an industrial test conducted on a rotary kiln for quicklime at a steel company. (1) The system operated continuously and stably for 3 months, experiencing different production loads, fluctuations in the calorific value of pulverized coal, and periodic changes in the kiln lining.

[0099] (2) Comparison of results: Quality: The pass rate of lime activity increased from 91% to 95.2%.

[0100] Energy consumption: The average comprehensive energy consumption per ton of ash decreased by 3.8% (coal consumption decreased significantly, while fan energy consumption increased slightly, but decreased after comprehensive conversion).

[0101] Environmental protection: Under the same operating conditions, the daily average NOx emission concentration decreased from 290 mg / m³ to 235 mg / m³, a reduction of approximately 19%.

[0102] (3) Conclusion: Experiments have shown that this method can effectively solve the multi-objective conflict problem, realize the green, low-carbon and efficient operation of rotary kiln, and has significant value for promotion and application.

[0103] Example 3 Based on Example 2, the optimization algorithm in Example 2 is replaced by the following alternative: The multi-objective particle swarm optimization (AMOPSO) algorithm in step S22 can be replaced by other multi-objective evolutionary algorithms, such as decomposition-based multi-objective evolutionary algorithm (MOEA / D) and non-dominated sorting genetic algorithm (NSGA-III), to adapt to different computing resources and problem complexity.

[0104] Alternative solution for the lower-level controller: If the MPC controller in step S3 is not feasible on-site, a feedforward + decoupled PID scheme can be used as an alternative. This involves designing a decoupling compensator to cancel out the coupling effects between the pulverized coal injection and blast air channels, and then designing two separate PID controllers for adjustment. Although the control accuracy is slightly lower than MPC, the implementation threshold is lower.

[0105] Example 4 Based on Example 2, the air volume distribution scheme is replaced as follows: For rotary kilns, the air volume can be divided into primary air and secondary air. The scope of protection of this invention covers the coordinated control of the total air volume, but specifically regarding the allocation of primary and secondary air, a fixed ratio allocation can be adopted, or an intelligent allocation based on flame image recognition can be adopted. Further refined schemes of intelligent allocation also fall within the scope of protection of this invention.

[0106] Example 5 Based on Example 2, an alternative objective function extension scheme could be: the multi-objective framework of this invention has good scalability. New objective dimensions can be added as needed, such as "the highest temperature of the kiln shell". ” (Preventing kiln crust formation), "carbon emission intensity" ” And so on, to achieve more goals while optimizing.

[0107] Example 6 Based on Example 2, this method can be applied to other applications: it is not only applicable to lime rotary kilns, but also to all combustion systems involving fuel and air ratio adjustment, such as coal-fired power plant boilers, cement rotary kilns, and metallurgical heating furnaces, and which need to balance efficiency, emissions, and safety.

[0108] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for coordinated control of fuel quantity and air volume in a lime rotary kiln based on multi-objective dynamic optimization, characterized in that, Includes the following steps: S1. Collect real-time operating data, including output, coal calorific value, coal injection rate, primary air volume, secondary air volume, actual calcination zone temperature T_act, actual air-fuel ratio λ_act, and NOx concentration in kiln tail flue gas. S2. Upper-level optimization: Triggered in the first cycle, a multi-objective function is constructed with lime activity, comprehensive energy consumption per ton of lime, and NOx concentration in kiln tail flue gas as objectives, and constraint and preference modes are established; an adaptive multi-objective particle swarm optimization algorithm is run to output the Pareto front, and the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt are selected according to the preference mode; the constraints include the calcination zone temperature boundary, the fan surge boundary, and the range of flue gas oxygen content; the preference modes include quality priority mode, energy saving priority mode, and low-NOx mode; S3. Input to the lower-level model predictive controller: The model predictive controller adopts a two-input, two-output dynamic matrix model with pulverized coal injection rate and total blast volume as inputs and actual calcination zone temperature T_act and actual air-fuel ratio λ_act as outputs; every second cycle, the model predictive controller reads T_opt, λ_opt and the current feedback values ​​T_act, λ_act, calculates the control increment of pulverized coal injection rate and blast volume through rolling optimization, and outputs control commands to the actuator; S4. The actuator adjusts the opening of the pulverized coal injection valve and the frequency of the blower frequency converter according to the control command, decomposes the total air volume command into primary air command and secondary air command, drives the rotary kiln body to run, and returns to step S1.

2. The method according to claim 1, characterized in that, The comprehensive energy consumption per ton of lime mentioned in step S2 is calculated by multiplying the coal injection quantity and the calorific value of the coal by the output, and the power consumption of the blower is also calculated.

3. The method according to claim 1, characterized in that, The lime activity in step S2 is estimated using a soft measurement model based on the actual calcination zone temperature T_act.

4. The method according to claim 3, characterized in that, The dynamic matrix model is obtained through step response experiments or historical data identification, and includes the hysteresis characteristics of pulverized coal injection rate on actual calcination zone temperature T_act and the coupled influence of total blast volume on actual calcination zone temperature T_act and actual air-fuel ratio λ_act.

5. The method according to claim 1, characterized in that, The goal of the rolling optimization in step S3 is to minimize the deviation between T_act and T_opt and the deviation between λ_act and λ_opt, while minimizing the control increment magnitude.

6. The method according to claim 1, characterized in that, The actuators mentioned in step S4 include a pulverized coal injection valve, a primary air fan frequency converter, and a secondary air fan frequency converter.

7. A coordinated control system for fuel quantity and air volume in a lime rotary kiln based on multi-objective dynamic optimization, characterized in that... To implement the method according to any one of claims 1-6, comprising: The optimization layer is triggered in each first cycle. It receives real-time operating data, constructs a multi-objective function and constraint and preference pattern with lime activity, comprehensive energy consumption per ton of lime and NOx concentration in kiln tail flue gas as objectives, runs an adaptive multi-objective particle swarm algorithm, outputs the Pareto front, and selects the optimal calcination zone temperature setpoint T_opt and the optimal air-fuel ratio setpoint λ_opt according to the preference pattern. The control layer is used to trigger every second cycle. The model prediction controller reads T_opt, λ_opt and the current feedback values ​​of the actual calcination zone temperature T_act and the actual air-fuel ratio λ_act. It calculates the control increment of the pulverized coal injection and blast volume through rolling optimization and outputs control commands. The actuator, including a pulverized coal injection valve, a primary air fan frequency converter, and a secondary air fan frequency converter, is used to receive the control command, adjust the opening of the pulverized coal injection valve and the frequency of the fan frequency converter, and decompose the total air volume command into a primary air command and a secondary air command. The controlled object layer, including the rotary kiln body, is used to receive primary air and secondary air and pulverized coal injection, output the actual calcination zone temperature T_act, the actual air-fuel ratio λ_act and NOx concentration, and feed them back to the optimization layer and the control layer. The optimization layer includes: a multi-objective function construction unit for constructing a multi-objective function with the objectives of lime activity, comprehensive energy consumption per ton of lime, and NOx concentration in kiln tail flue gas; a constraint and preference mode unit for setting constraint and preference modes; an adaptive multi-objective particle swarm algorithm unit for running the adaptive multi-objective particle swarm algorithm; and a Pareto front output unit for outputting the Pareto front and selecting T_opt and λ_opt according to the preference mode.

Images