Agent-driven pulverized coal boiler combustion optimization closed-loop control method and system

Abstract

The invention relates to the technical field of boiler combustion optimization, and discloses an Agent-driven pulverized coal boiler combustion optimization closed-loop control method and system, and the method comprises the steps: obtaining combustion data in the combustion process of a pulverized coal boiler; according to the combustion data and a preset combustion task, generating an initial combustion control strategy through a multi-agent layered cooperation mode of the power system; according to the combustion data and an initial combustion control strategy, optimizing the initial combustion control strategy by predicting the combustion state of the boiler in a preset future time period to obtain an optimized combustion strategy, and guiding the boiler to carry out combustion adjustment so as to carry out closed-loop control on the combustion optimization process of the gas turbine; through closed-loop control over the combustion optimization process of the pulverized coal boiler, different combustion tasks can be responded in time, the control strategy is automatically adjusted according to the combustion tasks and the combustion state of the boiler, the control performance is continuously improved, the combustion state meets the task requirement all the time, and stable operation of the gas turbine is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of boiler combustion optimization, and particularly to an Agent-driven closed-loop control method and system for pulverized coal boiler combustion optimization. Background Art

[0002] Traditional boiler combustion control methods adopt a centralized control mode. During the optimization process of the control mode, the overall control scheme needs to be optimized, resulting in low optimization efficiency and insufficiently timely optimization response to the control process. In the process of boiler combustion optimization, there is a lack of intelligent control and adaptive control, which cannot meet the application requirements of different combustion tasks and combustion scenarios.

[0003] For example, Chinese Patent Application No. CN115437245A discloses a boiler optimization control method based on a combustion state prediction model, which includes using a random forest machine learning algorithm and a model online update function to construct a boiler combustion state prediction model, and then obtaining the optimal adjustable state parameters under the current working conditions through a combustion state classification algorithm to guide the boiler to perform combustion adjustment, so as to achieve the self-optimization function of the combustion state under multi-objective optimization, and enable the boiler to achieve closed-loop optimization and control.

[0004] The above existing technologies have the problems raised in this background art: adopting a centralized control mode, with low optimization efficiency, resulting in insufficiently timely optimization response to the control process; lacking intelligent control and adaptive control, and unable to meet the application requirements of different combustion tasks and combustion scenarios. To solve any one of the above problems, the present application proposes an Agent-driven closed-loop control method and system for pulverized coal boiler combustion optimization. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the main purpose of the present invention is to provide an Agent-driven closed-loop control method and system for pulverized coal boiler combustion optimization, which can effectively solve the problems in the background art. The specific technical solutions of the present invention are as follows:

[0006] An Agent-driven closed-loop control method for pulverized coal boiler combustion optimization includes:

[0007] Obtaining combustion data during the combustion process of the pulverized coal boiler;

[0008] Combining the combustion data with a preset combustion task, and generating an initial combustion control strategy through the hierarchical cooperation of multiple intelligent agents in the power system;

[0009] According to the combustion data and the initial combustion control strategy, by predicting the boiler combustion state in a preset future period, optimizing the initial combustion control strategy to obtain an optimized combustion strategy, and guiding the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine.

[0010] Specifically, based on the combustion data and in combination with a preset combustion task, an initial combustion control strategy is generated through a hierarchical cooperation method of multiple agents in the power system, including:

[0011] Decompose the preset combustion task to construct a task network;

[0012] According to the task network, find multiple agents in the preset multi-agent network that execute corresponding tasks to obtain an agent combination;

[0013] According to the combustion data, each agent in the agent combination formulates a corresponding control strategy to obtain an initial combustion control strategy.

[0014] Specifically, the decomposition of the preset combustion task to construct a task network includes:

[0015] Decompose the preset combustion task to obtain multiple subtasks;

[0016] According to the refinement degree of the subtasks, perform hierarchical processing on the multiple subtasks to obtain multi-level subtasks;

[0017] According to the dependency relationship between adjacent-level subtasks, connect the subtasks between adjacent levels to obtain a task network, where at least one subtask in the lower-level subtasks is connected to a subtask in the higher-level subtasks among the adjacent-level subtasks.

[0018] Specifically, according to the task network, finding multiple agents in the preset multi-agent network that execute corresponding tasks to obtain an agent combination includes:

[0019] Initialize the multi-agent combination according to the boiler combustion process to obtain a preset multi-agent network;

[0020] Calculate a task adjacency matrix according to the topological structure of the task network;

[0021] Extract features of each task in the task network through a preset task feature extraction model to obtain task features;

[0022] Combined with the task adjacency matrix and task features, calculate the reward value of each agent corresponding to each task through a preset agent selection model;

[0023] According to the reward value, select the agent with the highest reward value corresponding to each task from high to low in the preset multi-agent network according to the task network hierarchical relationship to obtain multiple agents;

[0024] Connect the multiple agents accordingly according to the connection relationship of the subtasks in the task network to obtain an agent combination.

[0025] Specifically, according to the combustion data, each agent in the agent combination formulates a corresponding control strategy to obtain an initial combustion control strategy, including:

[0026] In the agent network, the top-level agent formulates a corresponding high-level control strategy according to the corresponding task;

[0027] In the hierarchical structure of the agent network from high to low, the lower-level agents between adjacent layers formulate control strategies according to the high-level control strategies of the corresponding connected high-level agents and in combination with the corresponding subtasks to obtain corresponding lower-level control strategies;

[0028] Until each agent in the agent network formulates a corresponding control strategy to obtain an initial combustion control strategy.

[0029] Specifically, according to the combustion data and the initial combustion control strategy, by predicting the boiler combustion state in a preset future period, the initial combustion control strategy is optimized to obtain an optimized combustion strategy to guide the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine, including:

[0030] According to the combustion data and the initial combustion control strategy, the boiler combustion data in a preset future period is predicted through a preset combustion prediction model, and the predicted combustion state is analyzed;

[0031] The predicted combustion state is compared with the preset combustion task, and the initial combustion control strategy is optimized to obtain an optimized combustion strategy;

[0032] According to the optimized combustion strategy, the combustion process of the pulverized coal boiler is controlled to realize the closed-loop control of the combustion optimization process of the gas turbine.

[0033] Specifically, the step of predicting the boiler combustion data in a preset future period through a preset combustion prediction model according to the combustion data and the initial combustion control strategy and analyzing to obtain the predicted combustion state includes:

[0034] According to the combustion data combined with the initial combustion control strategy, the boiler combustion data in a preset future period is predicted through a preset combustion prediction model to obtain predicted combustion data;

[0035] According to the predicted combustion data, the combustion state of the boiler is judged to obtain the predicted combustion state.

[0036] Specifically, the step of comparing the predicted combustion state with the preset combustion task, optimizing the initial combustion control strategy, and obtaining an optimized combustion strategy includes:

[0037] Compare the predicted combustion state with the preset combustion task to obtain the differential combustion data among the combustion data;

[0038] Update the combustion data sample pool according to the differential combustion data to obtain an updated sample pool;

[0039] Update the control strategy of the corresponding agent according to the updated sample pool to obtain an updated control strategy;

[0040] Optimize the control strategy of the corresponding agent in the initial combustion control strategy according to the updated control strategy to obtain an optimized control strategy.

[0041] Specifically, updating the control strategy of the corresponding agent according to the updated sample pool to obtain an updated control strategy includes:

[0042] Find the agent corresponding to the combustion task in the agent combination according to the newly added differential combustion data in the updated sample pool to obtain the agent to be updated;

[0043] Update the control strategy of the agent to be updated by using the updated sample pool in combination with the corresponding task to obtain an updated agent control strategy;

[0044] Update the corresponding control strategy of the agent connected to the agent to be updated in the agent combination according to the updated agent control strategy to obtain an updated control strategy.

[0045] An Agent-driven closed-loop control system for optimizing the combustion of pulverized coal boilers, used to implement the Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers, includes:

[0046] A data acquisition module that acquires combustion data during the combustion process of a pulverized coal boiler;

[0047] A control strategy formulation module that generates an initial combustion control strategy by means of multi-agent hierarchical cooperation according to the combustion data in combination with a preset combustion task;

[0048] A control process optimization module that optimizes the initial combustion control strategy by predicting the combustion state of the boiler in a preset future period according to the combustion data and the initial combustion control strategy to obtain an optimized combustion strategy, guiding the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine.

[0049] Compared with the prior art, the present application has the following beneficial effects:

[0050] This application optimizes the combustion control strategy through multi-agent hierarchical collaboration combined with combustion state prediction, realizes the closed-loop control of the combustion optimization process of pulverized coal boilers, can respond to different combustion tasks in a timely manner, automatically adjusts the control strategy according to the combustion tasks and the combustion state of the boiler, continuously improves the control performance, reduces manual intervention, improves production efficiency, realizes intelligent and adaptive control, enables the combustion state to always meet the task requirements, improves energy utilization efficiency, reduces fuel consumption, and ensures the stable operation of gas turbines. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a flowchart of the working process of the Agent-driven closed-loop combustion optimization control method for pulverized coal boilers in Embodiment 1 of the present invention;

[0052] Figure 2 It is a schematic structural diagram of the power system agent in Embodiment 1 of the present invention

[0053] Figure 3 It is a schematic diagram of the task network construction in Embodiment 1 of the present invention;

[0054] Figure 4 It is a schematic diagram of the agent combination construction in Embodiment 1 of the present invention;

[0055] Figure 5 It is a schematic structural diagram of the Agent-driven closed-loop combustion optimization control system for pulverized coal boilers in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following detailed description of the specific embodiments of the present invention will be made in conjunction with the accompanying drawings of the specification.

[0057] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0058] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other from other embodiments.

[0059] Embodiment 1

[0060] This embodiment provides an Agent-driven closed-loop combustion optimization control method for pulverized coal boilers, as Figure 1As shown, the closed-loop control method for optimizing the combustion of a pulverized coal boiler driven by an Agent includes:

[0061] S101. Obtain the combustion data during the combustion process of the pulverized coal boiler;

[0062] S102. Combine the combustion data with a preset combustion task, and generate an initial combustion control strategy through the hierarchical cooperation of multiple agents in the power system;

[0063] S103. Optimize the initial combustion control strategy by predicting the combustion state of the boiler in a preset future period according to the combustion data and the initial combustion control strategy, obtain an optimized combustion strategy, and guide the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine.

[0064] Currently, the existing optimization control methods for the combustion process of gas turbines have problems such as single control mode and untimely response. In this embodiment, by combining power system agents and combustion state prediction, the combustion process is optimized in real time to achieve closed-loop control of the combustion optimization process. Using the hierarchical control method of multiple agents, diverse control is provided, complex tasks are decomposed, each agent makes independent decisions and cooperates with each other, improving the flexibility and adaptability of the system, being able to better handle variable combustion conditions, and at the same time improving the task response speed; according to the combustion state prediction, the control strategy is optimized to achieve closed-loop control, making the control more accurate and efficient.

[0065] In this embodiment, first, the combustion data of the pulverized coal boiler is collected. Corresponding sensors are set at multiple key parts of the boiler. For example, temperature sensors are installed at different heights in the furnace to collect the furnace temperature; gas composition sensors are installed in the flue to detect the contents of oxygen, carbon monoxide, nitrogen oxides, etc.; flow sensors are installed in the fuel and air pipelines to measure the fuel and air flow rates; pressure and water level sensors are installed in the steam drum to monitor the steam pressure and water level; the collected combustion data of the boiler provides a data basis for the analysis of the boiler combustion state.

[0066] Specifically, according to the combustion data and the preset combustion task, multiple power system agents are used to formulate corresponding combustion control strategies. Combustion tasks usually have complexity and multi-objectives, such as improving combustion efficiency, reducing pollutant emissions, maintaining stable output power, etc. The complex task is decomposed into multiple sub-tasks, and each agent is responsible for a corresponding sub-task. Among them, the high-level agent considers the task and the overall combustion condition from a global perspective and formulates a macro strategy; the low-level agent executes specific control operations based on the high-level strategy and local combustion data. Through this hierarchical cooperation, the advantages of each agent can be fully utilized, and the adaptability of the control strategy can be enhanced. Different agents can be flexibly adjusted according to the sub-tasks they are responsible for and the real-time data.

[0067] Among them, as Figure 2 , the power system agent is an integrated intelligent control system, whose main function is to receive different data and specific requirements from the power system, analyze them and output the processing results. The power system agent includes a data processing and classification layer and multiple data processing modules. Each data processing module is configured with different types of models, which can process the input data according to specific task requirements. The task of the data processing and classification layer in this architecture is to parse the input data of the model and input the data into the corresponding processing modules according to the requirements of different tasks, ensuring that the tasks can be effectively processed and accurate processing results can be obtained.

[0068] Specifically, according to the combustion data and the formulated combustion control strategy, the combustion state in the future period is predicted. By predicting the combustion state in the future period, potential problems and change trends can be discovered in advance, so as to optimize the initial combustion control strategy. The optimized strategy can better adapt to the future working condition changes, realize closed-loop control, continuously adjust the strategy, and realize the continuous optimization of the gas turbine combustion process, improve the combustion efficiency and reduce pollutant emissions.

[0069] Exemplarily, a gas turbine is used for power peak shaving. The preset combustion task is to keep the combustion efficiency higher than 90% and the nitrogen oxide emission lower than 50 ppm while meeting the rapid change of the grid load. Through the sensors installed in various parts of the gas turbine, data such as fuel flow rate, air flow rate, temperature and pressure in the combustion chamber, and pollutant content in the exhaust gas are collected in real time. The high-level agent formulates a macro strategy according to the collected data and the corresponding responsible tasks, including strategies such as increasing the supply ratio of fuel and air in advance during the load rising stage; the low-level agent obtains the implementation strategy of the specific implementation method according to the high-level strategy and the combustion data, including strategies such as adjusting the fuel supply amount and air flow rate.

[0070] Furthermore, using the pre-trained LSTM model, the combustion data in the next 10 minutes is predicted according to the control strategy and the combustion data, including output power, combustion efficiency, and pollutant emissions, etc. If it is predicted that the combustion efficiency will drop to 85% and the nitrogen oxide emission will rise to 60 ppm when the load rises rapidly, the initial combustion control strategy is optimized and adjusted in real time in combination with the predicted combustion data, the mixing ratio of fuel and air and the combustion time are adjusted to obtain an optimized combustion strategy, and the optimized combustion strategy is sent to the actuator to adjust the combustion process. At the same time, the combustion data is continuously monitored, and the above prediction and optimization process is repeated to realize the continuous optimization and stable operation of the gas turbine combustion.

[0071] This application optimizes the combustion control strategy through multi-agent hierarchical collaboration combined with combustion state prediction, realizes the closed-loop control of the combustion optimization process of pulverized coal boilers, can respond to different combustion tasks in a timely manner, automatically adjusts the control strategy according to the combustion tasks and the combustion state of the boiler, continuously improves the control performance, reduces manual intervention, improves production efficiency, realizes intelligent and adaptive control, enables the combustion state to always meet the task requirements, improves energy utilization rate, reduces fuel consumption, and ensures the stable operation of gas turbines.

[0072] Furthermore, according to the combustion data combined with the preset combustion tasks, an initial combustion control strategy is generated through the multi-agent hierarchical collaboration method of the power system, including:

[0073] S201. Decompose the preset combustion tasks, and construct a task network;

[0074] S202. According to the task network, find multiple agents in the preset multi-agent network that execute the corresponding tasks, and obtain an agent combination;

[0075] S203. According to the combustion data, each agent in the agent combination formulates the corresponding control strategy to obtain the initial combustion control strategy.

[0076] The control strategies formulated by the current combustion process control methods are not detailed enough, and the strategy formulation process takes a long time, resulting in untimely task responses; this embodiment combines task network construction and agent matching, decomposes the tasks, each agent is responsible for the corresponding subtasks, and the formulated control strategies are more detailed, and realizes the reasonable allocation of tasks and the efficient cooperation of agents, improving the overall performance of the system and the response ability to working condition changes.

[0077] In this embodiment, first, the input combustion tasks are decomposed. The combustion tasks are decomposed into multiple subtasks, simplifying complex problems. According to the correlation between the subtasks, a task network is constructed, clearly showing the dependency relationship, sequence, and logical connection between the subtasks, avoiding chaotic task execution, and improving work efficiency.

[0078] Specifically, according to the task network, a matching agent combination is found in the preset multi-agent network. Each agent in the preset multi-agent network has a corresponding responsible function, corresponding to the subtasks in the task network. For example, the fuel control agent is responsible for tasks such as fuel delivery and flow regulation; the air volume control agent focuses on the regulation of air flow and pressure; by matching the subtasks in the task network with the agents in the multi-agent network, agents suitable for executing each subtask can be found, forming an agent combination, realizing the reasonable division of labor of agents, and each agent is responsible for the tasks it is good at, improving the accuracy and efficiency of task execution.

[0079] Specifically, each agent in the agent combination formulates a corresponding control strategy according to the subtasks it is responsible for and the obtained combustion data, using its internal data processing module and decision-making model. Obtaining the corresponding control strategies for each agent can improve the pertinence of the control strategies. Each agent formulates strategies for the subtasks it is responsible for, which better meets the actual requirements.

[0080] Furthermore, the decomposition of the preset combustion task to construct a task network includes:

[0081] S301. Decompose the preset combustion task to obtain multiple subtasks;

[0082] S302. According to the refinement degree of the subtasks, perform hierarchical processing on the multiple subtasks to obtain multi-level subtasks;

[0083] S303. According to the dependency relationship between adjacent-level subtasks, connect the subtasks between adjacent levels to obtain a task network, where at least one subtask in the lower-level subtasks and the higher-level subtasks of the adjacent-level subtasks are connected.

[0084] In this embodiment, as Figure 3 , decompose the combustion task. Decompose the preset combustion task into multiple subtasks according to the task function. For example, decompose the task into a fuel supply subtask responsible for precisely controlling the fuel delivery volume and delivery speed; an air supply subtask to ensure the supply of an appropriate proportion of air for mixing with the fuel; a burner control subtask to adjust parameters such as the angle and flame shape of the burner to optimize the combustion effect; a pollutant emission control subtask to reduce the generation of pollutants such as nitrogen oxides by adjusting the combustion conditions. By decomposing the task, the complex combustion task is refined into simple subtasks, reducing the processing difficulty. Each subtask has clear work content, facilitating the allocation of resources and responsibilities and improving work efficiency.

[0085] Specifically, perform hierarchical processing on the decomposed multiple subtasks. The refinement degrees of different subtasks are different. For example, the subtask of "improving combustion efficiency" is a general task with a greater impact on the overall task, while the subtask of "adjusting the rotation speed of the fuel pump" is more specific and is a specific operation to implement the combustion strategy. Perform hierarchical classification according to the refinement degree of the subtasks. Classify the subtasks involving overall goal planning and strategy formulation as high-level; classify the subtasks involving specific equipment operation and parameter adjustment as low-level; classify those in between as middle-level; form a clear task hierarchy, facilitating task allocation and management, improving management efficiency. Subtasks at different levels can be independently adjusted and optimized to adapt to different combustion conditions and task requirements.

[0086] Specifically, each subtask in the combustion task does not exist in isolation, and there are corresponding dependencies between tasks. Analyze the logical relationships between adjacent-layer subtasks, determine which subtasks have dependencies, and connect the subtasks between adjacent layers with directed edges according to the corresponding dependencies to construct a task network, making the entire combustion control process form an organic whole. By connecting subtasks, the sequentiality and coordination of task execution are ensured. The goals of high-level subtasks are achieved through the specific operations of low-level subtasks, while the execution of low-level subtasks is guided and restricted by high-level subtasks.

[0087] Further, according to the task network, find multiple agents in the preset multi-agent network that execute the corresponding tasks to obtain an agent combination, including:

[0088] S401. Initialize the multi-agent combination according to the boiler combustion process to obtain a preset multi-agent network;

[0089] S402. Calculate the task adjacency matrix according to the topological structure of the task network;

[0090] S403. Extract features of each task in the task network through a preset task feature extraction model to obtain task features;

[0091] S404. Combine the task adjacency matrix and task features, and calculate the reward value of each agent corresponding to each task through a preset agent selection model;

[0092] S405. According to the reward value, select the agent with the highest reward value corresponding to each task from high to low in the preset multi-agent network according to the task network hierarchy relationship to obtain multiple agents;

[0093] S406. Connect the multiple agents accordingly according to the connection relationship of the subtasks in the task network to obtain an agent combination.

[0094] In this embodiment, as Figure 4 , the optimal-performance agents corresponding to each subtask in the task network are obtained according to the task network matching. The selected agents are combined according to the corresponding relationship to obtain an agent combination. Each agent in the agent combination collaborates according to the task dependency relationship. By directly selecting the optimal-performance agent corresponding to each task, the training process can be reduced, the agent selection and calculation efficiency can be improved, resource waste and task conflicts can be avoided, and the operation efficiency of the system can be improved.

[0095] Specifically, according to the combustion process of the pulverized coal boiler, the combustion process involves multiple complex links and functions, including fuel supply, air delivery, combustion reaction, heat transfer, steam generation, and pollutant emission control. Each link requires specific control and decision-making. By analyzing these links and functions, the agents are initialized and pre-trained. Based on multiple initialized agents, a preset multi-agent network is constructed to obtain an initial multi-agent combination. Each agent has corresponding implementation functions and implementation efficiencies, realizing distributed intelligent control and improving the flexibility and adaptability of the system. Based on the physical process of boiler combustion and combined with the types and functions of the agents, several corresponding candidate agents are set for each subtask to obtain an agent set.

[0096] According to the topological structure in the task network, an adjacency matrix is constructed to quantify the dependency relationship between tasks in the task network. A parameter vector and a corresponding adjacency matrix are initialized for each agent. According to the static and dynamic features in the task scenario, feature extraction is performed through a preset task feature extraction model. The task feature extraction model can be a random forest model, a convolutional neural network model, or a recurrent neural network model. In this embodiment, the task feature extraction model is a random forest model. The random forest model is trained using a large amount of historical data to obtain a pre-trained task feature extraction model. Through the pre-trained task feature extraction model, task features are calculated to construct a task feature vector. Combining the task adjacency matrix and task features, through a preset agent selection model, the reward value of each agent corresponding to each task can be calculated. The agent selection model is specifically a context multi-armed bandit algorithm model. This model can dynamically select the agent with the optimal performance without pre-training according to the real-time task scenario features. The reward value of each agent is calculated through the agent selection model, and the calculation formula is as follows:

[0097]

[0098] In the formula, p i is the reward value of agent i, x is the task feature vector, θ i is the parameter vector of agent i, A i,kl is the adjacency matrix A i the element representing the connection relationship between node k and node l in it, β is the control exploration intensity (which can be set according to the calculation accuracy requirements), x k is the k-th element in the task feature vector x, x l is the l-th element in the task feature vector x.

[0099] According to the calculated reward values of each agent, starting from the highest layer in accordance with the task network structure, the agent with the highest reward value corresponding to the task is selected successively. The agent with a higher reward value has stronger performance in handling the task. By selecting the agent with a higher reward value, the agent selection process can be optimized and the task targeting can be improved. Starting from the highest layer and selecting downward successively the agents corresponding to each task, multiple agents are obtained. According to the connection relationship of the subtasks in the task network, the multiple agents are connected accordingly to obtain an agent combination, enabling the agents to orderly transmit information and execute tasks, and realizing efficient distributed control; after the high-level agent strategy is formulated, a message is transmitted to the corresponding connected low-level agent, triggering the low-level agent to execute the corresponding task.

[0100] Further, for each agent in the agent combination, according to the combustion data, corresponding control strategies are formulated to obtain an initial combustion control strategy, including:

[0101] S501. In the agent network, the high-level agent formulates a corresponding high-level control strategy according to the corresponding task.

[0102] S502. In the hierarchical structure of the agent network from high to low, the low-level agent between adjacent layers formulates a control strategy by combining the corresponding subtask according to the high-level control strategy of the corresponding connected high-level agent, obtaining a corresponding low-level control strategy.

[0103] S503. Until each agent in the agent network formulates a corresponding control strategy, obtaining an initial combustion control strategy.

[0104] Currently, the task response speed of the centralized decision-making method is slow and the adaptability is poor. In this embodiment, a hierarchical agent combination is used to perform real-time control on the boiler combustion process, improving the control rate, and being able to quickly adjust the control strategies of each agent according to different working conditions and task requirements, with strong flexibility and adaptability.

[0105] In this embodiment, according to the hierarchical structure, the agent combination first formulates an overall control strategy from the high-level agent, layer by layer downward. The corresponding low-level agent formulates a corresponding subtask control strategy according to the control strategy of the corresponding connected high-level agent until each agent obtains a corresponding control strategy.

[0106] Specifically, the control strategy is formulated by the highest-level agent. The highest-level agent is in a position of macro decision-making in the agent network and can control the entire boiler combustion process from a global perspective. The tasks it is responsible for are usually relatively macroscopic goals in the preset combustion tasks, such as meeting the grid load demand, achieving overall combustion efficiency improvement, reaching specific pollutant emission standards, etc. The highest-level agent formulates a guiding high-level control strategy through the internal data processing module based on the tasks it is responsible for and the combustion data obtained, providing direction and framework for the decision-making of the lower-level agents.

[0107] Specifically, the lower-level agents make further decisions and formulate strategies based on the macro strategies formulated by the highest-level agents, combined with the subtasks they are responsible for and the real-time combustion data. The lower-level agents receive the high-level control strategies sent by the corresponding connected highest-level agents, interpret them, and clarify the constraints and goals they need to follow. At the same time, they conduct a detailed analysis of the subtasks they are responsible for to determine the specific requirements and goals of the subtasks. For example, the air volume control agent is responsible for adjusting the air volume entering the furnace, and it needs to analyze factors such as the current combustion state and fuel supply to determine the appropriate air adjustment goal; combined with the high-level control strategy, subtask requirements, and real-time combustion data, it formulates the corresponding control strategy. The control strategy is formulated layer by layer downward. Finally, each agent formulates the corresponding control strategy according to its own tasks and the upper-level strategy, obtaining the initial combustion control strategy. Through hierarchical transmission and integration, the consistency and coordination among the control strategies of each agent are ensured, and strategy conflicts and contradictions are avoided.

[0108] Furthermore, according to the combustion data and the initial combustion control strategy, by predicting the boiler combustion state in a preset future period, the initial combustion control strategy is optimized to obtain an optimized combustion strategy, which guides the boiler to perform combustion adjustment to achieve closed-loop control of the combustion optimization process of the gas turbine, including:

[0109] S601. According to the combustion data and the initial combustion control strategy, predict the boiler combustion data in a preset future period through a preset combustion prediction model, and analyze to obtain the predicted combustion state;

[0110] S602. Compare the predicted combustion state with the preset combustion task, optimize the initial combustion control strategy, and obtain the optimized combustion strategy;

[0111] S603. According to the optimized combustion strategy, control the combustion process of the pulverized coal boiler to achieve closed-loop control of the combustion optimization process of the gas turbine.

[0112] Furthermore, according to the initial combustion control strategy, the future combustion state of the boiler is predicted, and the control strategy is optimized in advance, overcoming the limitation that traditional control methods can only adjust according to the current state, improving the forward-looking and accuracy of control; realizing the closed-loop control of the combustion optimization process, effectively coping with various disturbances and working condition changes, maintaining the stability of the boiler operation, and reducing the probability of faults.

[0113] In this embodiment, the preset combustion prediction model is an LSTM model. The LSTM model is trained based on a large amount of historical combustion data and corresponding combustion state data to obtain a pre-trained LSTM model; the preprocessed current combustion data and the initial combustion control strategy are input into the trained combustion prediction model. The model calculates and infers according to the input data and outputs the predicted values of the boiler combustion data for a preset future period. The predicted values include key parameters such as furnace temperature, flue gas temperature, flue gas composition, and steam pressure. According to the predicted combustion data, the combustion state of the boiler is judged to determine the future combustion state of the boiler. The combustion state usually includes different categories such as good, normal, and abnormal (such as incomplete combustion, overheating, and overcooling).

[0114] Specifically, the predicted combustion state reflects the future operation of the boiler under the initial combustion control strategy. The predicted combustion state is compared with the preset combustion task to judge whether the initial combustion control strategy can meet the requirements of the preset combustion task. If there is a deviation, the initial combustion control strategy needs to be adjusted and optimized to make the future combustion state closer to the preset target; by adjusting the control strategy in a timely manner, the combustion process of the boiler is made closer to the preset combustion task target, improving the combustion efficiency and reducing pollutant emissions; the control strategy is flexibly adjusted according to different prediction results and actual situations, enhancing the adaptability and stability of the system.

[0115] Specifically, according to the optimized combustion control strategy, the combustion process of the boiler is controlled and adjusted in real time, and the optimization and adjustment process is continuously repeated to form a closed-loop control cycle, so that the combustion process of the boiler is always optimized towards the preset combustion task target, capable of timely correcting the deviation in the combustion process, maintaining the stability of the boiler operation, and reducing the fluctuation of parameters.

[0116] Furthermore, predicting the boiler combustion data for a preset future period according to the combustion data and the initial combustion control strategy through a preset combustion prediction model, and analyzing to obtain the predicted combustion state, including:

[0117] S701. According to the combustion data combined with the initial combustion control strategy, predict the boiler combustion data for a preset future period through a preset combustion prediction model to obtain predicted combustion data;

[0118] S702. Determine the combustion state of the boiler based on the predicted combustion data to obtain the predicted combustion state.

[0119] In this embodiment, the boiler combustion process has continuity and regularity. According to the current combustion data and the initial combustion control strategy, the changes in various parameters of the boiler in the future time period can be predicted. According to the pre-trained LSTM model, the predicted values of the boiler combustion data in the future time period (such as 5 minutes, 10 minutes, etc.) can be predicted. By predicting the future combustion data in advance, the execution effect of the initial combustion control strategy can be judged and adjusted.

[0120] Specifically, based on the predicted combustion data, the combustion state of the boiler is judged to determine the future combustion state of the boiler. The combustion state usually includes different categories such as good, normal, abnormal (such as insufficient combustion, overheating, overcooling, etc.). By judging the future combustion state of the boiler, the combustion problems that will occur in the future can be discovered in time, which is convenient for adjusting and optimizing the combustion process in advance to avoid the deterioration of the problems.

[0121] Further, the step of comparing the predicted combustion state with the preset combustion task and optimizing the initial combustion control strategy to obtain the optimized combustion strategy includes:

[0122] S801. Compare the predicted combustion state with the preset combustion task to obtain the differential combustion data among the combustion data;

[0123] S802. Update the combustion data sample pool according to the differential combustion data to obtain the updated sample pool;

[0124] S803. Update the control strategy of the corresponding agent according to the updated sample pool to obtain the updated control strategy;

[0125] S804. Optimize the control strategy of the corresponding agent in the initial combustion control strategy according to the updated control strategy to obtain the optimized control strategy.

[0126] In this embodiment, the sample pool is updated according to the differential data obtained by comparing the predicted state with the preset task, and the agent is optimized for the control strategy, realizing dynamic optimization based on real-time data. The control strategy of the corresponding agent is adjusted according to the differential data, improving the adaptability and coordination of the entire control strategy system, rather than the traditional single global adjustment, which is more accurate and efficient.

[0127] Specifically, by comparing the preset combustion task with the predicted combustion state, the gap between the actual prediction and the expected target is obtained. These gaps are reflected in the form of differential combustion data, which reflects in which aspects the boiler combustion process fails to meet the requirements of the preset task under the current control strategy. For example, if the preset combustion efficiency target is 90% and the predicted combustion efficiency is 85%, the difference in this indicator is -5%. Organizing the differences of all indicators into differential combustion data can quickly and accurately identify the deviations from the preset task in the boiler combustion process.

[0128] Specifically, put the differential combustion data into the sample pool, integrate it with the original data in the sample pool, clean the integrated data, remove the existing noise, outliers and duplicate data. At the same time, perform preprocessing operations such as normalization and standardization on the data to improve the quality and usability of the data, and obtain an updated sample pool. According to the updated sample pool, retrain the intelligent agent related to the corresponding differential data or adjust its control parameters so that it can learn new knowledge in the updated sample pool and obtain an updated control strategy more suitable for the current combustion condition. Replacing the control strategy of the corresponding intelligent agent in the initial combustion control strategy with the updated control strategy can ensure that the entire control strategy system can better respond to the current combustion condition, thereby realizing the optimization of the initial combustion control strategy.

[0129] Furthermore, according to the updated sample pool, update the control strategy of the corresponding intelligent agent to obtain an updated control strategy, including:

[0130] S901: According to the newly added differential combustion data in the updated sample pool, find the intelligent agent corresponding to the combustion task in the intelligent agent combination to obtain the intelligent agent to be updated;

[0131] S902: Use the updated sample pool combined with the corresponding task to update the control strategy of the intelligent agent to be updated to obtain an updated intelligent agent control strategy;

[0132] S903: According to the updated intelligent agent control strategy, perform corresponding control strategy updates on the intelligent agents connected to the intelligent agent to be updated in the intelligent agent combination to obtain an updated control strategy.

[0133] In this embodiment, each agent in the agent combination is responsible for a specific combustion task. The newly added differential combustion data in the sample pool is updated to reflect the deviation of the combustion state. Different deviations correspond to problems in different combustion task links. By analyzing the combustion task corresponding to the newly added differential combustion data, the agent that needs to update the control strategy, that is, the agent to be updated, can be accurately located. This can avoid unnecessary updates to the control strategies of all agents, improve the pertinence and efficiency of the update. For example, when the newly added differential data shows that the fuel combustion is insufficient, the corresponding combustion tasks are fuel supply, air ratio, etc. The control strategy of the agent is updated specifically, making the update process more efficient and enabling a faster response to changes in the combustion conditions.

[0134] Specifically, the agent to be updated is updated according to the newly added combustion data and differential information in the update sample pool, so that the agent to be updated can learn new data patterns and rules, thereby adjusting its own control strategy to obtain the updated agent control strategy. By continuously using the update sample pool for updating, the continuous optimization of the agent control strategy is realized, and the overall performance of the combustion process is improved.

[0135] Specifically, the change in the control strategy of the agent to be updated will affect the working environment and task requirements of the agents connected to it. According to the updated agent control strategy, the control strategies of the connected agents are adjusted accordingly to ensure the collaborative working effect of the entire agent combination, enable the entire agent control system to operate more stably, and improve the reliability and safety of the combustion process.

[0136] Embodiment 2

[0137] In this embodiment, as Figure 5 , an Agent-driven closed-loop control system for optimizing the combustion of pulverized coal boilers is provided, which is used to implement the Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers, including:

[0138] A data acquisition module that acquires combustion data during the combustion process of the pulverized coal boiler;

[0139] A control strategy formulation module that generates an initial combustion control strategy by means of multi-agent hierarchical collaboration according to the combustion data in combination with preset combustion tasks;

[0140] A control process optimization module that optimizes the initial combustion control strategy by predicting the combustion state of the boiler in a preset future period according to the combustion data and the initial combustion control strategy, obtains an optimized combustion strategy, and guides the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine.

[0141] In this embodiment, the data acquisition module includes a sensor group, a data acquisition unit, a data transmission unit, and a data storage unit, which can collect various key data during the combustion process of the pulverized coal boiler in real time and accurately, and transmit and store them. By continuously acquiring combustion data, the system can timely understand the operating state of the boiler, discover potential problems and change trends, so as to realize the effective monitoring and optimization of the combustion process; the control strategy formulation module includes a task decomposition unit, a task network construction unit, an agent management unit, an agent matching unit, and a strategy generation unit. According to the preset combustion task and real-time combustion data of the pulverized coal boiler, an initial combustion control strategy is generated through a multi-agent hierarchical cooperation method. This module decomposes the complex combustion task into multiple subtasks and matches the corresponding agents, giving full play to the advantages of each agent to achieve refined control of the combustion process. The generated initial combustion control strategy provides a guiding direction for the combustion process and is an important link to achieve combustion optimization.

[0142] Specifically, the control process optimization module includes a combustion prediction unit, a comparison and analysis unit, a sample pool update unit, an agent strategy update unit, and a control execution unit, which optimizes the initial combustion control strategy to achieve closed-loop control of the combustion process of the pulverized coal boiler. This module continuously adjusts the combustion control strategy through operations such as predicting the future combustion state, comparing with the preset task, updating the sample pool and the agent control strategy, so that the combustion process of the boiler can better meet the requirements of the preset task. Through closed-loop control, it can respond to changes in the combustion condition in real time, continuously optimize the combustion process, improve the combustion efficiency, reduce pollutant emissions, and ensure the safe, stable, and efficient operation of the gas turbine.

[0143] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers, characterized in that, Including: Obtain combustion data during the combustion process of a pulverized coal boiler; According to the combustion data and in combination with a preset combustion task, generate an initial combustion control strategy through the hierarchical cooperation of multiple agents in the power system; According to the combustion data and the initial combustion control strategy, optimize the initial combustion control strategy by predicting the boiler combustion state in a preset future time period, and obtain an optimized combustion strategy to guide the boiler to perform combustion adjustment, so as to perform closed-loop control on the combustion optimization process of the gas turbine.

2. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 1, characterized in that, According to the combustion data and in combination with a preset combustion task, generate an initial combustion control strategy through the hierarchical cooperation of multiple agents in the power system, including: Decompose the preset combustion task, and construct a task network; According to the task network, find multiple agents that execute corresponding tasks in a preset multi-agent network to obtain an agent combination; According to the combustion data, each agent in the agent combination formulates a corresponding control strategy to obtain an initial combustion control strategy.

3. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 2, characterized in that The decomposition of the preset combustion task to construct a task network includes: Decompose the preset combustion task to obtain multiple subtasks; According to the refinement degree of the subtasks, perform hierarchical processing on the multiple subtasks to obtain multi-layer subtasks; According to the dependency relationship between adjacent-layer subtasks, connect the subtasks between adjacent layers to obtain a task network, where at least one subtask in the lower-layer subtasks is connected to a subtask in the upper-layer subtasks among the adjacent-layer subtasks.

4. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 2, wherein According to the task network, finding multiple agents that execute corresponding tasks in a preset multi-agent network to obtain an agent combination includes: According to the boiler combustion process, initialize the multi-agent combination to obtain a preset multi-agent network; Calculate a task adjacency matrix according to the topological structure of the task network; Through a preset task feature extraction model, extract features of each task in the task network to obtain task features; Combined with the task adjacency matrix and task features, calculate the reward value of each agent corresponding to each task through a preset agent selection model; According to the reward value, select the agent with the highest reward value corresponding to each task from high to low in the preset multi-agent network according to the task network hierarchy relationship to obtain multiple agents; According to the connection relationship of the subtasks in the task network, connect the multiple agents accordingly to obtain an agent combination.

5. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 2, characterized in that, The formulation of the corresponding control strategy by each agent in the agent combination according to the combustion data to obtain an initial combustion control strategy includes: In the agent network, the highest-layer agent formulates a corresponding high-level control strategy according to the corresponding task; From high to low in the agent network hierarchy, the lower-layer agents between adjacent layers formulate control strategies in combination with the corresponding subtasks according to the high-level control strategies of the corresponding connected high-level agents to obtain corresponding low-level control strategies; Until each agent in the agent network formulates a corresponding control strategy to obtain an initial combustion control strategy.

6. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 1, characterized in that, According to the combustion data and the initial combustion control strategy, by predicting the boiler combustion state in a preset future period, the initial combustion control strategy is optimized to obtain an optimized combustion strategy, which guides the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine, including: According to the combustion data and the initial combustion control strategy, the boiler combustion data in a preset future period is predicted through a preset combustion prediction model, and the predicted combustion state is analyzed and obtained; The predicted combustion state is compared with a preset combustion task, and the initial combustion control strategy is optimized to obtain an optimized combustion strategy; According to the optimized combustion strategy, the combustion process of the pulverized coal boiler is controlled to achieve closed-loop control of the combustion optimization process of the gas turbine.

7. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 6, characterized in that The step of, according to the combustion data and the initial combustion control strategy, predicting the boiler combustion data in a preset future period through a preset combustion prediction model and analyzing and obtaining the predicted combustion state includes: According to the combustion data combined with the initial combustion control strategy, the boiler combustion data in a preset future period is predicted through a preset combustion prediction model to obtain predicted combustion data; According to the predicted combustion data, the combustion state of the boiler is judged to obtain the predicted combustion state.

8. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 6, characterized in that, The step of comparing the predicted combustion state with a preset combustion task, optimizing the initial combustion control strategy, and obtaining an optimized combustion strategy includes: The predicted combustion state is compared with a preset combustion task to obtain differential combustion data among the combustion data; The combustion data sample pool is updated according to the differential combustion data to obtain an updated sample pool; According to the updated sample pool, the control strategy of the corresponding agent is updated to obtain an updated control strategy; According to the updated control strategy, the control strategy of the corresponding agent in the initial combustion control strategy is optimized to obtain an optimized control strategy.

9. The Agent-driven closed-loop control method for optimizing the combustion of pulverized coal boilers according to claim 8, characterized in that, The step of, according to the updated sample pool, updating the control strategy of the corresponding agent to obtain an updated control strategy includes: According to the newly added differential combustion data in the updated sample pool, the agent corresponding to the combustion task in the agent combination is found to obtain the agent to be updated; The control strategy of the agent to be updated is updated by using the updated sample pool combined with the corresponding task to obtain an updated agent control strategy; According to the updated agent control strategy, the control strategies of the agents connected to the agent to be updated in the agent combination are updated accordingly to obtain an updated control strategy.

10. A closed-loop control system for optimizing the combustion of pulverized coal boilers driven by an agent, characterized in that, It is used to implement the Agent-driven closed-loop control method for combustion optimization of pulverized coal boilers according to any one of claims 1 to 9, including: A data acquisition module that acquires combustion data during the combustion process of the pulverized coal boiler; A control strategy formulation module that generates an initial combustion control strategy by means of multi-agent hierarchical cooperation according to the combustion data combined with a preset combustion task; A control process optimization module that optimizes the initial combustion control strategy by predicting the boiler combustion state in a preset future period according to the combustion data and the initial combustion control strategy to obtain an optimized combustion strategy, which guides the boiler to perform combustion adjustment to perform closed-loop control on the combustion optimization process of the gas turbine.

Images