Coal-fired power plant boiler with temperature adjusting structure and temperature adjusting control method and system

Abstract

The invention discloses a coal-fired power plant boiler with a temperature adjusting structure, the coal-fired power plant boiler is a T-shaped boiler, convection vertical shafts are arranged on the two sides of the T-shaped boiler respectively, the convection vertical shafts are connected with a hearth through horizontal flues, two tail flues are arranged in each convection vertical shaft in parallel, and the tail flues are connected with the T-shaped boiler. A low-temperature reheater is arranged at the position, close to the hearth, of the tail flue, a low-temperature superheater is arranged at the position, away from the hearth, of the tail flue, and an adjustable flue baffle is arranged at the tail end of each flue. The invention further discloses a temperature regulation control method and system thereof, a boiler flue digital twinborn model is constructed for training to obtain an intelligent agent, boiler real-time state parameters are collected, a reward function is set for training a strategy network, and control actions are output based on the strategy network. The problems that the T-shaped furnace is small in temperature adjustment amplitude, adjustment lags behind and the response rate is low are solved, and quick response and stability of the temperature are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to a power station boiler, a temperature control method and system thereof. BACKGROUND

[0002] The existing supercritical T-shaped boiler currently uses a steam-steam heat exchanger to adjust the temperature. The steam-steam heat exchanger has a large volume, a lagging adjustment, and a limited adjustment range, which is difficult to meet the requirements of rapid peak shaving and rapid load change.

[0003] A single flue gas damper can be used to adjust the temperature of the flue gas at the tail of the T-shaped boiler. This damper configuration is more suitable for stable load conditions and scenarios with low adjustment accuracy requirements. However, the single damper on one side has a coupling effect on the low-temperature reheating and the low-temperature superheating, making it difficult to independently optimize the flow rate and temperature of different flues for step-by-step temperature adjustment, affecting the temperature adjustment effect and response control effect. Uneven distribution of flue gas can also cause local wear or ash accumulation, requiring frequent manual intervention.

[0004] In addition, the control method of the prior art also has difficulty in dealing with the coupling effect of multiple dampers on both sides of the steam temperature, such as the PID controller which is difficult to adapt to system dynamic changes, resulting in overshoot or slow adjustment. Traditional control methods face many limitations in flue gas damper control and cannot achieve ideal results. SUMMARY

[0005] To overcome the above-mentioned defects of the prior art, the present application aims to provide a coal-fired power station boiler with a temperature adjustment structure to solve the problem of small temperature adjustment range, lagging adjustment, and slow response rate of the T-shaped boiler, which is difficult to adapt to the requirements of rapid peak shaving and rapid load change. Another task of the present application is to provide a temperature adjustment control method and a corresponding temperature adjustment control system for a coal-fired power station boiler with a temperature adjustment structure, which can achieve rapid response and stability of temperature under variable load conditions.

[0006] The technical scheme of the present application is as follows: a coal-fired power station boiler with a temperature adjustment structure, wherein the coal-fired power station boiler is a T-shaped boiler, two sides of the T-shaped boiler are provided with a convection shaft, the convection shaft is connected with a furnace by a horizontal flue, two tail flues are arranged in parallel in each convection shaft, a low-temperature reheater is arranged near the furnace, a low-temperature superheater is arranged away from the furnace, and an adjustable flue damper is arranged at the end of each flue.

[0007] Further, the tail flue provided with the low-temperature superheater is further provided with an economizer, and the economizer is located behind the low-temperature superheater in the flue gas flow direction.

[0008] Further, the flue damper is a double-leaf flat structure.

[0009] Another technical solution of the present invention is a temperature control method for a coal-fired power plant boiler with a temperature regulation structure, based on the aforementioned coal-fired power plant boiler with a temperature regulation structure, comprising:

[0010] A digital twin model of the boiler flue based on the actual system was constructed to simulate the temperature and pressure under different flue damper openings, and an intelligent agent was trained based on historical operating data.

[0011] The real-time status parameters of the T-type boiler are collected and input into the strategy network to obtain the distribution of all actions. A reward function is set to calculate the corresponding reward amount for each action, select the optimal reward, and use the reward to update the strategy network using the PPO algorithm. The real-time status parameters include the temperature of the low-temperature superheater, the temperature of the low-temperature reheater, the boiler load, and the current opening of each flue damper. The action is the target opening of each flue damper.

[0012] Based on the trained policy network, the optimal action is obtained according to the real-time state parameters of the T-type boiler, namely, controlling the opening of the flue damper.

[0013] Furthermore, the reward function reflects at least one of three indicators: minimum deviation between average steam temperature and rated temperature, improvement in boiler thermal efficiency, and minimum NOx / SO2 emissions.

[0014] Furthermore, when obtaining the corresponding optimal action, i.e. controlling the opening degree of the flue damper, an action constraint space is set. The action constraint space includes the opening degree limit of a single flue damper and the total opening degree limit of multiple flue dampers.

[0015] Furthermore, the total opening of the multi-flue damper is limited to the sum of the openings of two flue dampers in the same convection shaft.

[0016] Another technical solution of the present invention is a temperature control system for a coal-fired power plant boiler with a temperature regulation structure, wherein the coal-fired power plant boiler is the aforementioned coal-fired power plant boiler with a temperature regulation structure, and the temperature control system includes:

[0017] The intelligent agent construction unit is used to build a digital twin model of the boiler flue based on the actual system, simulate the temperature and pressure under different flue damper openings, and train the intelligent agent based on historical operating data.

[0018] The data acquisition and status variable unit is used to acquire real-time status parameters of the T-type boiler;

[0019] The policy gradient reinforcement learning unit inputs the real-time state parameters into the policy network to obtain the distribution of all actions, sets a reward function to calculate the corresponding reward amount for each action, selects the optimal reward, and uses the reward to update the policy network using the PPO algorithm. The real-time state parameters include the temperature of the low-temperature superheater, the temperature of the low-temperature reheater, the boiler load, and the current opening degree of each flue damper. The action is the target opening degree of each flue damper.

[0020] The control action output unit is used to obtain the corresponding optimal action, i.e., control the opening degree of the flue damper, based on the real-time state parameters of the T-type boiler and the trained policy network.

[0021] Compared with the prior art, the advantages of the technical solution of the present invention are as follows:

[0022] This invention fully utilizes the T-type boiler to divert the flue gas to both sides, effectively solving the problems of insufficient tail-end heating surface and arrangement space. Two baffles are placed on each side for adjustment, reducing the coupling effect between the low-temperature superheater and the primary or secondary low-temperature reheater, effectively adjusting the superheater / reheater deviation (deviation on both sides of the superheater is less than 2%), and improving response speed and sensitivity. The independent adjustment of the four flue gas baffles effectively avoids response lag and large temperature fluctuations caused by single baffle adjustment, achieving a smooth transition and reducing adjustment fluctuations. It also balances the heat absorption of each heating surface, avoiding localized overheating or low-temperature corrosion, and extending equipment life. Optimizing the flue gas path reduces ineffective heat exchange, lowers the exhaust gas temperature (typically by 1-3℃), and improves the overall boiler efficiency. Simultaneously, while the four baffles increase the number of baffles, they reduce baffle movement, resulting in better control and reduced maintenance costs such as actuator wear.

[0023] This invention employs an intelligent control method combining reinforcement learning and digital twin technology, simultaneously meeting engineering requirements for safety, reliability, and adaptability. The intelligent control system, through coordinated action, can quickly and accurately make control judgments on damper adjustment feedback, addressing the characteristics of specific T-type boilers such as strong coupling, time-varying characteristics, nonlinearity, and multi-objective conflicts. It provides stable control methods with good steam temperature stability, timely dynamic compensation, and small temperature fluctuations. This can reduce reheat steam temperature fluctuations by 60%; improve unit thermal efficiency by 0.3%; reduce damper operation frequency by 30%, extending actuator life; shorten adjustment time during load changes by 50%; achieve a 99.5% consistency rate between command and actual opening degree; and improve reheat steam temperature control accuracy by 40% under various load conditions.

[0024] Overall, this invention effectively eliminates the temperature deviation at the steam outlets on both sides of the superheater and reheater, without affecting combustion and thermal efficiency in the furnace, thereby improving the unit's economy while achieving efficient and accurate temperature control. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of a coal-fired power plant boiler with a temperature regulation structure.

[0026] Figure 2 This is a schematic diagram of the temperature control method for a coal-fired power plant boiler with a temperature regulation structure. Detailed Implementation

[0027] The present invention will be further described below with reference to embodiments, but these are not intended to limit the scope of the invention.

[0028] Please combine Figure 1 As shown, the coal-fired power plant boiler with a temperature-regulating structure involved in this embodiment of the invention is a double reheat T-type boiler, that is, the boiler adopts a single furnace body structure, the heating surface is arranged in a "T" shape, and convection shafts 2 are arranged on both sides of the furnace 1. The top of the convection shafts 2 is connected to the horizontal flues 3 on both sides of the boiler, thus communicating with the furnace 1. The horizontal flues are respectively equipped with a final high-temperature reheater 4, a primary high-temperature reheater 5, and a secondary high-temperature reheater 6. Each convection shaft 2 is divided into two parallel tail flues by a vertical partition wall (there are four tail flues in total for the two convection shafts, i.e., "four flues"). The front tail flues on both sides of the furnace 1 (i.e., the tail flues relatively close to the furnace 1) are used to arrange the primary low-temperature reheater 7 and the secondary low-temperature reheater 8, and the rear tail flues (i.e., the tail flues relatively far from the furnace 1) are used to arrange the low-temperature superheaters 9a, 9b and the economizer 10.

[0029] Adjustable flue dampers are installed at the bottom of each tail flue. Adjusting the damper opening changes the medium flow rate, thus affecting heat exchange efficiency. It also alters the flue gas flow resistance in each flue to regulate flue gas flow distribution, thereby controlling different target temperatures. There are four independent adjustable dampers, named First Damper 11, Second Damper 12, Third Damper 13, and Fourth Damper 14, corresponding to the First Low-Temperature Superheater 9a, the First Low-Temperature Reheater 7, the Second Low-Temperature Reheater 8, and the Second Low-Temperature Superheater 9b. The combination of damper openings determines the distribution ratio of flue gas among the heating surfaces, thus affecting the reheat steam temperature. Each damper blade is a split-plate structure, streamlined, and formed in a single stamping process.

[0030] Please combine Figure 2 As shown, the temperature control system of this coal-fired power plant boiler with a temperature regulation structure includes an agent construction unit, a data acquisition and state variable unit, a policy gradient reinforcement learning unit, and a control action output unit. Each unit combines reinforcement learning with digital twin technology to implement the temperature control method. The specific working principle is as follows:

[0031] The established agent building blocks are used to construct a digital twin model of the flue gas duct based on the actual system, simulating the distribution of temperature, pressure, etc., under different baffle angles. Historical operating data is collected from the existing control system's historical operations, and the agents are fully validated and trained in a real-time simulation environment. After training, the optimal strategy is transferred to the target system, and historical operating data guides the initialization of the strategy network. When all four flue gas baffles require coordinated adjustment, each baffle can be considered as an agent. By sharing state information or reward signals, the temperature coupling relationship between the low-temperature reheater and the primary or secondary low-temperature reheater in the T-type reheat boiler system is coordinated, achieving coordinated operation of the four flue gas baffles. The simulation twin model needs to be built based on a CFD or thermodynamic model and can be quickly validated on a digital twin platform (such as MATLAB / Simulink or Python+PyTorch), and the RL agent trained before being transferred to the physical system. Multi-agent cooperation of the four flue gas baffles can be implemented using the MADDPG or COMA method.

[0032] The established data acquisition and state variable unit is used to collect important parameter variables for the T-type boiler damper control. Real-time state parameters are collected through deployed sensors for flue gas temperature, flue gas flow rate, flue gas pressure, and damper opening feedback. A state processor is then used to determine state variables and construct the state. Among these, the most crucial sensor for flue gas temperature measurement uses Class II precision K-type thermocouples (chromium-aluminum-nickel-aluminum, -40℃~1300℃, ±0.75%t within 1300℃, φ6mm, with an Inconel 600 protective sleeve, insertion depth ≥8 times the pipe diameter), arranged on both sides of the damper. Redundant thermocouples are used, with 2-4 thermocouples on each side. In the intelligent control loop, a single measurement value is measured multiple times, and the average value is taken as the real-time data, improving the reliability of the detection data. The initial parameter for the superheater side flue gas temperature is 605℃, with a sensor range of 0℃~600℃; the initial parameter for the reheater side flue gas temperature is 623℃, with a sensor range of 0℃~800℃. The established policy gradient reinforcement learning unit adopts the policy gradient method of reinforcement learning, mainly using PPO-Clip. Data acquisition and the state(s) in the state variable unit are input into the policy network (the policy network can also receive inputs such as steam temperature deviation and historical state sequences) to obtain the distribution of all actions(a), and the maximum reward value is taken as the optimal policy. A penalty term is added to the policy gradient calculation; the smaller the penalty value, the better. In the overall expectation, when the data exceeds the expected range, the excess data is truncated in a timely manner to reduce computational complexity and learning difficulty. Specifically, the process can be as follows: data samples are extracted from the flue gas state parameters, and an action sequence is generated through the policy network, i.e., four selectable flue damper openings. The corresponding reward amount for each action is calculated through a reward function (which may include a penalty term). Finally, the optimal reward is selected, and the reward is used to update the policy through the PPO algorithm for further reinforcement learning. Through continuous training and learning, the openings of the four dampers are continuously optimized to obtain the optimal control action, i.e., the target opening, enabling it to respond quickly even under dynamic operating conditions. The reward function, based on the multi-objective characteristics of the T-type boiler, can be designed to reflect a single or combined indicator, such as minimizing the deviation between the average steam temperature and the rated temperature, improving boiler thermal efficiency (positive reward correlates with efficiency improvement), and minimizing NOx / SO2 emissions (negative reward is proportional to the concentration of NOx / SO2 exceeding the standard). As an optimal example, the reward function in this embodiment... Where R tracking =-λ1·∣T reheat -T set | represents the temperature deviation penalty, T reheat T is the reheat temperature. set To set the temperature; R load =-λ2∣P actual -P target| represents the dynamic load response, P actual t P represents the actual boiler power. target The target power of the boiler; For the baffle opening constraint term, u t is the baffle opening command, Ⅱ is the indicator function, and clip is the cutoff function, whose function is to forcibly restrict the baffle opening command ut to a physically allowed range (e.g., 0% to 100%, i.e., the [0,1] interval); R pressure =-λ4|ΔP furnace ∣·Ⅱ(∣ΔP furnace ∣>∈), where ΔP is the pressure fluctuation term. furnace R represents the rate of change of negative pressure, where ∈ is the threshold. smooth =-λ5∣u t -u t-1 | represents the baffle vibration term and motion smoothness; Rentropy = λ6·H(π(·|s) t ), where H is the policy entropy, and π(·∣s) is the exploration incentive term. t Let π be the policy function expression. π: The policy function, which determines how the agent (such as an AI controlling a baffle) chooses its action in a given state. ·: A placeholder representing the input variable of the policy function (here, all possible actions a). |st: The state st given the current time t. π(·|st) represents the probability distribution of all possible actions a output by policy π in state st; Renergy=-λ7·EnergyCost(u t ), where λ1, λ2, λ3, λ4, λ5, λ6, and λ7 are the coefficients of each term.

[0033] The established control action output unit is used to output control action variables, namely the control damper opening. The four-damper flue gas control adopts absolute value control: directly outputting the target opening of each damper (K1, K2, K3, K4), and setting certain action constraint space, such as opening limits and total opening constraints (e.g., K1+K2=M1, K3+K4=M2). These constraints and conditions can be adjusted according to the load rate. Furthermore, by using reinforcement learning to map the optimal damper opening under different loads (20% THA~100% BMCR), the fixed parameters of traditional PID control are replaced, ensuring reasonable distribution of the total flue gas volume and achieving multi-damper coordination. Finally, the action variables are fed back to the simulation twin model unit through experience playback. Through model retraining and strategy updates, the accuracy and applicability of intelligent control are continuously optimized.

[0034] Table of key states and action variables in flue gas damper control

[0035] Variable Type Specific Parameter Parameter Representation Typical Range / Unit State Variable First Low Temperature Superheater Temperature T1 ℃ Primary Reheat Temperature T2 ℃ Secondary Reheat Temperature T3 ℃ Second Low Temperature Superheater Temperature T4 ℃ Boiler Load LDM MW Current Opening of Each Damper [K1 0 , K2 0 , K3 0 , K4 0 ]]> ％ Action Variable First Damper Opening K1 ％ Second Damper Opening K2 ％ Third Damper Opening K3 ％ Fourth Damper Opening K4 ％

[0036] Based on the actual operating parameters of a certain boiler, the flue damper opening is controlled by the temperature control method of the coal-fired power plant boiler with temperature regulation structure of the present invention. As a result, the reheat steam temperature fluctuation is reduced by 60%, the unit thermal efficiency is increased by 0.3%, the damper action frequency is reduced by 30%, the life of the actuator is extended, the adjustment time during load changes is shortened by 50%, the consistency rate between command and actual opening reaches 99.5%, and the reheat steam temperature control accuracy is improved by 40% under various load conditions.

Claims

1. A coal-fired power plant boiler with a temperature regulating structure, characterized in that, The coal-fired power plant boiler is a T-type boiler. Convection shafts are set on both sides of the T-type boiler. The convection shafts are connected to the furnace through horizontal flues. Two tail flues are arranged in parallel in each convection shaft. The tail flue closer to the furnace is equipped with a low-temperature reheater, and the tail flue farther from the furnace is equipped with a low-temperature superheater. An adjustable flue baffle is set at the end of each flue.

2. The coal-fired power plant boiler with a temperature regulating structure according to claim 1, characterized in that, An economizer is also provided in the tail flue of the low-temperature superheater, and the economizer is located behind the low-temperature superheater in the direction of flue gas flow.

3. The coal-fired power plant boiler with a temperature regulating structure according to claim 1, characterized in that, The flue damper is a split flat plate structure.

4. A method for temperature control of a coal-fired power plant boiler with a temperature regulation structure, characterized in that, Based on the coal-fired power plant boiler with temperature regulation structure according to any one of claims 1 to 3, including: A digital twin model of the boiler flue based on the actual system was constructed to simulate the temperature and pressure under different flue damper openings, and an intelligent agent was trained based on historical operating data. The real-time status parameters of the T-type boiler are collected and input into the strategy network to obtain the distribution of all actions. A reward function is set to calculate the corresponding reward amount for each action, select the optimal reward, and use the reward to update the strategy network using the PPO algorithm. The real-time status parameters include the temperature of the low-temperature superheater, the temperature of the low-temperature reheater, the boiler load, and the current opening of each flue damper. The action is the target opening of each flue damper. Based on the trained policy network, the optimal action is obtained according to the real-time state parameters of the T-type boiler, namely, controlling the opening of the flue damper.

5. The temperature control method for a coal-fired power plant boiler with a temperature regulating structure according to claim 4, characterized in that, The reward function reflects at least one of three indicators: minimum deviation between average steam temperature and rated temperature, improvement in boiler thermal efficiency, and minimum NOx / SO2 emissions.

6. The temperature control method for a coal-fired power plant boiler with a temperature regulation structure according to claim 4, characterized in that, When obtaining the corresponding optimal action, i.e., controlling the opening of the flue damper, an action constraint space is set. The action constraint space includes the opening limit of a single flue damper and the total opening limit of multiple flue dampers.

7. The temperature control method for a coal-fired power plant boiler with a temperature regulating structure according to claim 6, characterized in that, The total opening of the multi-flue damper is limited to the sum of the openings of two flue dampers in the same convection shaft.

8. A temperature control system for a coal-fired power plant boiler with a temperature regulation structure, characterized in that, The coal-fired power plant boiler is a coal-fired power plant boiler with a temperature regulation structure as described in any one of claims 1 to 3, and the temperature regulation control system includes: The intelligent agent construction unit is used to build a digital twin model of the boiler flue based on the actual system, simulate the temperature and pressure under different flue damper openings, and train the intelligent agent based on historical operating data. The data acquisition and status variable unit is used to acquire real-time status parameters of the T-type boiler; The policy gradient reinforcement learning unit inputs the real-time state parameters into the policy network to obtain the distribution of all actions, sets a reward function to calculate the corresponding reward amount for each action, selects the optimal reward, and uses the reward to update the policy network using the PPO algorithm. The real-time state parameters include the temperature of the low-temperature superheater, the temperature of the low-temperature reheater, the boiler load, and the current opening degree of each flue damper. The action is the target opening degree of each flue damper. The control action output unit is used to obtain the corresponding optimal action, i.e., control the opening degree of the flue damper, based on the real-time state parameters of the T-type boiler and the trained policy network.

9. The temperature control system for a coal-fired power plant boiler with a temperature regulation structure according to claim 8, characterized in that, The reward function reflects at least one of three indicators: minimum deviation between average steam temperature and rated temperature, improvement in boiler thermal efficiency, and minimum NOx / SO2 emissions.

10. The temperature control system for a coal-fired power plant boiler with a temperature regulation structure according to claim 8, characterized in that, When obtaining the corresponding optimal action, i.e., controlling the opening of the flue damper, an action constraint space is set. The action constraint space includes the opening limit of a single flue damper and the total opening limit of multiple flue dampers.

Images