Power plant dust removal system parameter optimization system and method based on reinforcement learning

Abstract

The invention relates to the technical field of power plant dust removal system parameter optimization systems, in particular to a power plant dust removal system parameter optimization system and method based on reinforcement learning, and a data acquisition and preprocessing module is used for collecting and preprocessing historical operation data of a power plant dust removal system; the reinforcement learning training module is used for training a control strategy of the dust removal system by adopting a reinforcement learning algorithm based on the preprocessed historical operation data; the online decision-making module is connected with the reinforcement learning training module and is used for receiving real-time operation data and making an action decision based on a trained strategy model; the safety guarantee module is connected with the online decision-making module and is used for ensuring the stability and the safety of the online decision-making module; and the feedback optimization module is used for collecting decision results and actual operation data and carrying out feedback evaluation and strategy optimization, so that the dust removal efficiency is improved by 1.5%, the energy consumption is reduced by 15.5%, the pollutant emission concentration is also remarkably reduced, and considerable economic benefits are brought.

Description

Technical Field

[0001] The present invention relates to the technical field of power plant dust removal system parameter optimization system, in particular to a power plant dust removal system parameter optimization system based on reinforcement learning and a method thereof. Background Art

[0002] With the increasingly stringent environmental protection requirements, the performance optimization of power plant dust removal systems has become the focus of the industry. Traditional power plant dust removal system parameter optimization methods mainly rely on experience debugging and fixed parameter control, which is difficult to adapt to complex and changeable working conditions. In recent years, with the development of artificial intelligence technology, some researchers have begun to try to apply intelligent algorithms to the parameter optimization of dust removal systems.

[0003] At present, the closest existing technologies mainly include optimization methods based on fuzzy control and neural networks. These methods have improved the performance of dust removal systems to a certain extent, but there are still many problems. First, although the fuzzy control method can handle the uncertainty of the system, the formulation of its control rules still relies on expert experience and is difficult to adapt to new working conditions. Secondly, although the neural network-based method has self-learning capabilities, it often requires a large amount of labeled data for training. In actual industrial environments, it is often difficult and costly to obtain high-quality labeled data.

[0004] In addition, existing intelligent optimization methods generally have problems with insufficient real-time performance and insufficient safety considerations. In complex industrial environments, adjustments to system parameters need to respond quickly, while also ensuring that radical adjustments do not cause system instability or safety hazards. Existing methods often perform unsatisfactorily in this regard and are unable to meet the needs of actual power plant operations. Summary of the invention

[0005] The present invention aims to solve the problems existing in the existing power plant dust removal system parameter optimization method, and proposes a power plant dust removal system parameter optimization system and method based on reinforcement learning. The system realizes intelligent, real-time and safe optimization of dust removal system parameters by introducing reinforcement learning algorithm, combining modules such as data collection and preprocessing, online decision-making, safety assurance and feedback optimization.

[0006] The present invention proposes a power plant dust removal system parameter optimization system based on reinforcement learning, including a data acquisition and preprocessing module, a reinforcement learning training module, an online decision-making module, a safety assurance module and a feedback optimization module, wherein:

[0007] The data acquisition and preprocessing module is used to collect and preprocess the historical operation data of the power plant dust removal system;

[0008] The reinforcement learning training module is connected to the data acquisition and preprocessing module, and is used to train the control strategy of the dust removal system using a reinforcement learning algorithm based on the preprocessed historical operation data;

[0009] The online decision-making module is connected to the reinforcement learning training module and is used to receive real-time operation data and make action decisions based on the trained strategy model;

[0010] The security assurance module is connected to the online decision-making module to ensure the stability and security of the online decision-making module;

[0011] The feedback optimization module is connected to the online decision-making module and the safety assurance module, and is used to collect decision-making results and actual operation data, and perform feedback evaluation and strategy optimization.

[0012] Preferably, the data acquisition and preprocessing module includes a data collection unit, a data cleaning unit, an outlier processing unit, a feature extraction unit and a data partitioning unit, wherein:

[0013] The data collection unit is used to collect historical operation data of the power plant dust removal system through sensors, databases and data cleaning tools;

[0014] The data cleaning unit is connected to the data collection unit and is used to clean and convert the historical operation data to ensure the quality and integrity of the data;

[0015] The outlier processing unit is connected to the data cleaning unit and is used to identify and process outliers using statistical methods and machine learning algorithms;

[0016] The feature extraction unit is connected to the abnormal value processing unit and is used to extract key features from the historical operation data, including real-time data and set values, as inputs to the reinforcement learning training module;

[0017] The data division unit is connected to the feature extraction unit and is used to divide the historical operation data into a learning data set and a test data set.

[0018] Preferably, the reinforcement learning training module includes a model selection unit, a model training unit, a model evaluation unit and an iterative optimization unit, wherein:

[0019] The model selection unit is used to select an appropriate reinforcement learning model, including a Q-learning algorithm, and initialize it in an environment simulation;

[0020] The model training unit is connected to the model selection unit and is used to train the selected reinforcement learning model in the environment simulation;

[0021] The model evaluation unit is connected to the model training unit and is used to evaluate the trained reinforcement learning model on the test data set and calculate its performance index;

[0022] The iterative optimization unit is connected to the model evaluation unit and is used to adjust the parameters and hyperparameters of the model based on the evaluation result and perform iterative optimization.

[0023] Preferably, the online decision module includes a data receiving unit, a state space calculation unit, an action space calculation unit, a decision calculation unit and a result display unit, wherein:

[0024] The data receiving unit is used to receive real-time operation data of the power plant dust removal system;

[0025] The state space calculation unit is connected to the data receiving unit and is used to calculate the dimension of the state space according to the input requirements of the reinforcement learning model;

[0026] The action space calculation unit is connected to the state space calculation unit and is used to calculate the dimension of the action space according to the output requirements of the reinforcement learning model;

[0027] The decision calculation unit is connected to the action space calculation unit, and is used to input the received real-time operation data into the trained reinforcement learning model to obtain an action decision;

[0028] The result display unit is connected to the decision calculation unit and is used to display the decision result to the operator, including action decision, reward and accumulated reward.

[0029] Preferably, the security assurance module includes a security indicator setting unit, a security monitoring unit, an alarm generating unit and a security control unit, wherein:

[0030] The safety index setting unit is used to set the indexes related to the safety of the power plant dust removal system, including but not limited to limiting SO2 / NO x Operating time during which emissions exceed warning values;

[0031] The safety monitoring unit is connected to the safety index setting unit and is used to monitor the actual data of the power plant dust removal system in real time and track the change trend of the index;

[0032] The alarm generating unit is connected to the safety monitoring unit and is used to generate an alarm when the actual data exceeds a predetermined safety index, including but not limited to SO2 / NO x The operation time exceeding the warning value exceeds 5 minutes;

[0033] The safety control unit is connected to the alarm generating unit, and is used to take necessary safety control measures when receiving the alarm information, including but not limited to immediately stopping the dust removal system and notifying the management personnel.

[0034] Preferably, the feedback optimization module includes a data collection unit, a strategy evaluation unit and a strategy optimization unit, wherein:

[0035] The data collection unit is used to collect the results of the online decision-making module and the actual operation data of the power plant dust removal system, including action decisions, rewards, cumulative rewards, and safety indicators;

[0036] The strategy evaluation unit is connected to the data collection unit and is used to perform feedback evaluation based on the collected data;

[0037] The strategy optimization unit is connected to the strategy evaluation unit and is used to perform strategy optimization according to the feedback evaluation result, including adjusting the parameters and hyperparameters of the reinforcement learning model, improving the calculation method of the state space and action space, and repeating the training and evaluation process until the predetermined optimization goal is met.

[0038] A method for optimizing parameters of a power plant dust removal system based on reinforcement learning based on the system includes the following steps:

[0039] Through the data acquisition and preprocessing module, the historical operation data of the power plant dust removal system is collected and preprocessed;

[0040] Through the reinforcement learning training module, based on the pre-processed historical operation data, a reinforcement learning algorithm is used to train the control strategy of the dust removal system;

[0041] Receiving real-time operation data through an online decision-making module, and making action decisions based on the trained strategy model;

[0042] Through the security module, the stability and security of the online decision-making module are ensured;

[0043] Through the feedback optimization module, the decision results and actual operation data are collected to conduct feedback evaluation and strategy optimization.

[0044] Preferably, the step of collecting and preprocessing the historical operation data of the power plant dust removal system by the data acquisition and preprocessing module comprises:

[0045] Collect historical operating data of power plant dust removal systems through sensors, databases and data cleaning tools;

[0046] Clean and convert the historical operation data to ensure the quality and integrity of the data;

[0047] Use statistical methods and machine learning algorithms to identify and handle outliers;

[0048] Extracting key features from the historical operation data, including real-time data and set values, as input to the reinforcement learning training module;

[0049] The historical operation data is divided into a learning data set and a test data set.

[0050] Preferably, the step of using a reinforcement learning algorithm to train the control strategy of the dust removal system based on the pre-processed historical operation data through a reinforcement learning training module comprises:

[0051] Select an appropriate reinforcement learning model, including the Q-learning algorithm, and initialize it in the environment simulation;

[0052] In the environment simulation, training the selected reinforcement learning model;

[0053] Evaluate the trained reinforcement learning model on the test dataset and calculate its performance indicators;

[0054] Based on the evaluation results, the parameters and hyperparameters of the model are adjusted and iterative optimization is performed.

[0055] Preferably, the step of receiving real-time operation data through an online decision module and making an action decision based on the trained strategy model includes:

[0056] Receive real-time operating data of the power plant dust removal system;

[0057] Calculating the dimension of the state space according to the input requirements of the reinforcement learning model;

[0058] Calculating the dimension of the action space according to the output requirements of the reinforcement learning model;

[0059] Inputting the received real-time operation data into the trained reinforcement learning model to obtain an action decision;

[0060] The decision results are presented to the operator, including action decisions, rewards, and accumulated rewards.

[0061] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0062] Beneficial effects of the present invention

[0063] The power plant dust removal system parameter optimization system and method based on reinforcement learning proposed in the present invention have significant advantages in overall architecture design and module collaboration, effectively solve many problems existing in the prior art, and produce synergistic, complementary and superimposed effects in many aspects.

[0064] From a macro perspective, the present invention constructs a closed-loop intelligent optimization system, in which each module closely cooperates to form an adaptive, self-learning optimization mechanism. The data acquisition and preprocessing module provides a high-quality data basis for the entire system, the reinforcement learning training module continuously optimizes the control strategy based on these data, and the online decision-making module promptly applies the optimized strategy to actual control. The safety assurance module and the feedback optimization module supervise and improve the system from the perspectives of safety and long-term performance, respectively, to ensure the stability of the system and its continuous optimization capabilities.

[0065] Going deep into the microscopic level, the present invention has achieved technical breakthroughs in multiple key points:

[0066] 1. Adaptive learning capability: By introducing reinforcement learning algorithms, the system can continuously learn and optimize control strategies from interactions with the environment without the need for large amounts of labeled data, effectively solving the problem of traditional methods’ high reliance on expert experience.

[0067] 2. Real-time decision-making and control: The online decision-making module can quickly process real-time data and make decisions, significantly shortening the system response time and effectively responding to instantaneous changes in working conditions.

[0068] 3. Security assurance: A specially designed security assurance module ensures system security during parameter optimization through real-time monitoring and early warning mechanisms, solving the problem of insufficient security considerations in existing methods.

[0069] 4. Long-term optimization capability: The feedback optimization module achieves long-term optimization of the control strategy by continuously collecting operating data and evaluating system performance, enabling the system to adapt to slowly changing operating conditions and equipment performance degradation.

[0070] 5. Efficient data utilization: The data collection and preprocessing modules not only improve data quality, but also reduce data dimensions through techniques such as feature extraction, thereby improving the efficiency of subsequent learning and decision-making.

[0071] These micro-level advantages interact with each other to produce significant synergistic effects. For example, the combination of adaptive learning capabilities and real-time decision-making enables the system to quickly adapt to new working conditions; the combination of safety assurance and long-term optimization ensures that the system does not sacrifice stability and safety while pursuing high efficiency.

[0072] In practical applications, the system of the present invention has shown excellent performance. Compared with the traditional method, the dust removal efficiency has increased by 1.5 percentage points, the energy consumption has decreased by 15.5%, and the pollutant emission concentration has also decreased significantly. These improvements have not only brought considerable economic benefits, but also made important contributions to environmental protection.

[0073] In general, this invention has built an intelligent, efficient and safe power plant dust removal system parameter optimization scheme by integrating advanced technologies such as reinforcement learning, real-time control and safety assurance. This scheme not only solves many problems in the existing technology, but also achieves a qualitative leap in performance, efficiency and safety, providing a new technical path for the intelligent upgrade of power plant dust removal systems, which has important theoretical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is the overall logic block diagram of the system of the present invention.

[0075] Figure 2 It is a logic block diagram of the data acquisition and preprocessing module of the present invention.

[0076] Figure 3 It is a detailed logic block diagram of the reinforcement learning training module of the present invention.

[0077] Figure 4 It is a detailed logic block diagram of the online decision-making module of the present invention.

[0078] Figure 5 It is a detailed logic block diagram of the security protection module of the present invention.

[0079] Figure 6 It is a detailed logic block diagram of the feedback optimization module of the present invention.

[0080] Figure 7 Detailed flow chart of the method of the present invention. DETAILED DESCRIPTION

[0081] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects thereof are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0082] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0083] Example 1

[0084] See also Figure 1-7 The present invention discloses a power plant dust removal system parameter optimization system and method based on reinforcement learning. The specific implementation of the present invention will be described in detail below so that those skilled in the art can fully understand and implement the present invention.

[0085] First, the system proposed in the present invention includes a data acquisition and preprocessing module 1, a reinforcement learning training module 2, an online decision-making module 3, a safety assurance module 4 and a feedback optimization module 5. These modules work together to achieve intelligent optimization of the parameters of the power plant dust removal system. The data acquisition and preprocessing module 1 is responsible for collecting and preprocessing the historical operation data of the power plant dust removal system. The reinforcement learning training module 2 uses a reinforcement learning algorithm to train the control strategy of the dust removal system based on the preprocessed historical operation data. The online decision-making module 3 receives real-time operation data and makes action decisions based on the trained strategy model. The safety assurance module 4 ensures the stability and security of the online decision-making module 3. The feedback optimization module 5 collects decision results and actual operation data, and performs feedback evaluation and strategy optimization.

[0086] In one embodiment of the present invention, the data acquisition and preprocessing module 1 can collect real-time data such as dust concentration, wind speed, temperature, etc. through a sensor network arranged at key points of the dust removal system. After preprocessing, these data provide high-quality input for subsequent reinforcement learning training. This comprehensive data acquisition and preprocessing method can significantly improve the effect of model training, making the final control strategy more accurate and reliable.

[0087] Secondly, in one embodiment of the present invention, the data acquisition and preprocessing module 1 includes a data collection unit 11, a data cleaning unit 12, an outlier processing unit 13, a feature extraction unit 14 and a data partitioning unit 15. The data collection unit 11 collects the historical operation data of the power plant dust removal system through sensors, databases and data cleaning tools. The data cleaning unit 12 cleans and converts the historical operation data to ensure the quality and integrity of the data. The outlier processing unit 13 uses statistical methods and machine learning algorithms to identify and process outliers. The feature extraction unit 14 extracts key features from the historical operation data as input to the reinforcement learning training module 2. The data partitioning unit 15 divides the historical operation data into a learning data set and a test data set.

[0088] Preferably, the outlier processing unit 13 can use the 3σ principle or the isolation forest algorithm to detect outliers. For example, when a parameter value exceeds the range of plus or minus 3 standard deviations of the mean value, it can be regarded as an outlier. This method can effectively identify and process outliers in the data and improve the accuracy of subsequent model training.

[0089] Again, in one embodiment of the present invention, the reinforcement learning training module 2 includes a model selection unit 21, a model training unit 22, a model evaluation unit 23 and an iterative optimization unit 24. The model selection unit 21 selects an appropriate reinforcement learning model, including a Q-learning algorithm, and initializes it in an environmental simulation. The model training unit 22 trains the selected reinforcement learning model in an environmental simulation. The model evaluation unit 23 evaluates the trained reinforcement learning model on a test data set and calculates its performance index. The iterative optimization unit 24 adjusts the parameters and hyperparameters of the model based on the evaluation results and performs iterative optimization.

[0090] In one embodiment of the present invention, the Q-learning algorithm is selected based on its advantages in dealing with discrete action space problems. The core formula of the Q-learning algorithm is:

[0091] Q(s,a)←Q(s,a)+α[r+γ·max(Q(s',a'))-Q(s,a)]

[0092] Among them, s represents the current state, a represents the current action, r represents the immediate reward, s' represents the next state, α is the learning rate, and γ is the discount factor. By continuously updating the Q value, the algorithm can eventually learn the optimal action strategy.

[0093] Then, in one embodiment of the present invention, the online decision module 3 includes a data receiving unit 31, a state space calculation unit 32, an action space calculation unit 33, a decision calculation unit 34 and a result display unit 35. The data receiving unit 31 receives the real-time operation data of the dust removal system of the power plant. The state space calculation unit 32 calculates the dimension of the state space according to the input requirements of the reinforcement learning model. The action space calculation unit 33 calculates the dimension of the action space according to the output requirements of the reinforcement learning model. The decision calculation unit 34 inputs the received real-time operation data into the trained reinforcement learning model to obtain an action decision. The result display unit 35 displays the decision results to the operator, including action decisions, rewards and accumulated rewards.

[0094] In a specific embodiment of the present invention, the state space includes the current dust concentration (range 0-1000 mg / m 3 ), wind speed (range 0-20m / s), temperature (range 0-200℃) and other parameters. The action space includes adjustable parameters such as the dust collector's operating voltage (range 0-100kV), spray water volume (range 0-100L / min), etc. This sophisticated state and action space design enables the system to control the dust removal process more accurately.

[0095] Next, in one embodiment of the present invention, the safety assurance module 4 includes a safety indicator setting unit 41, a safety monitoring unit 42, an alarm generating unit 43 and a safety control unit 44. The safety indicator setting unit 41 sets indicators related to the safety of the power plant dust removal system, including but not limited to limiting SO2 / NO x The operating time when the emission exceeds the warning value. The safety monitoring unit 42 monitors the actual data of the power plant dust removal system in real time and tracks the changing trend of the indicators. The alarm generation unit 43 generates an alarm when the actual data exceeds the predetermined safety index. The safety control unit 44 takes necessary safety control measures when receiving the alarm information.

[0096] Preferably, the SO2 warning value is set to 35 mg / m 3 , NO x The warning value is set at 50 mg / m 3 These values ​​are determined based on the national environmental protection standards (GB13223-2011) and the specific conditions of the power plant. x If the operating time exceeds the warning value for more than 5 minutes, the system will generate an alarm. This 5-minute threshold is set based on the response time of the power plant dust removal system and the potential environmental impact.

[0097] Again, according to claim 6, the feedback optimization module 5 includes a data collection unit 51, a strategy evaluation unit 52 and a strategy optimization unit 53. The data collection unit 51 collects the results of the online decision module 3 and the actual operation data of the power plant dust removal system, including action decisions, rewards, cumulative rewards, and safety indicators. The strategy evaluation unit 52 performs feedback evaluation based on the collected data. The strategy optimization unit 53 performs strategy optimization based on the feedback evaluation results, including adjusting the parameters and hyperparameters of the reinforcement learning model, improving the calculation method of the state space and action space, and repeating the training and evaluation process until the predetermined optimization goal is met.

[0098] In one embodiment of the present invention, the policy optimization unit 53 adjusts the learning rate α and the discount factor γ of the Q-learning algorithm. For example, α=0.1 and γ=0.9 can be set initially, and then these parameters are gradually adjusted according to the model performance until the optimal parameter combination is found. This dynamic adjustment method enables the system to maintain optimal performance at different operating stages.

[0099] Next, in one embodiment of the present invention, the present invention also proposes a method for optimizing the parameters of a power plant dust removal system based on reinforcement learning, including the following steps: through a data acquisition and preprocessing module 1, collecting and preprocessing the historical operation data of the power plant dust removal system; through a reinforcement learning training module 2, based on the preprocessed historical operation data, using a reinforcement learning algorithm to train the control strategy of the dust removal system; through an online decision-making module 3, receiving real-time operation data, and making action decisions based on the trained strategy model; through a safety assurance module 4, ensuring the stability and security of the online decision-making module 3; through a feedback optimization module 5, collecting decision results and actual operation data, and performing feedback evaluation and strategy optimization.

[0100] The advantage of this method is that it can adaptively optimize the parameters of the dust removal system. Compared with traditional fixed parameters or simple feedback control methods, it can better cope with complex and changeable operating environments, improve dust removal efficiency and reduce energy consumption.

[0101] Then, according to one embodiment of the present invention, the steps of collecting and preprocessing the historical operation data of the power plant dust removal system through the data acquisition and preprocessing module 1 include: collecting the historical operation data of the power plant dust removal system through sensors, databases and data cleaning tools; cleaning and converting the historical operation data to ensure the quality and integrity of the data; using statistical methods and machine learning algorithms to identify and process outliers; extracting key features from the historical operation data as input to the reinforcement learning training module 2; and dividing the historical operation data into a learning data set and a test data set.

[0102] In a specific embodiment of the present invention, a principal component analysis (PCA) method can be used for feature extraction. PCA can effectively reduce the data dimension and retain the most important information, thereby improving the efficiency and effect of subsequent model training.

[0103] Next, in one embodiment of the present invention, the steps of using a reinforcement learning algorithm to train a control strategy of a dust removal system based on preprocessed historical operating data through a reinforcement learning training module 2 include: selecting an appropriate reinforcement learning model, including a Q-learning algorithm, and initializing it in an environmental simulation; training the selected reinforcement learning model in the environmental simulation; evaluating the trained reinforcement learning model on a test data set and calculating its performance indicators; adjusting the parameters and hyperparameters of the model based on the evaluation results, and performing iterative optimization.

[0104] Preferably, in the model evaluation stage, indicators such as average reward, cumulative reward, and dust removal efficiency can be used to evaluate model performance. For example, the target dust removal efficiency can be set to 99%, and if the model can achieve this goal on the test data set, the training can be considered successful.

[0105] Finally, in one embodiment of the present invention, the online decision module 3 receives real-time operation data, and the steps of making action decisions based on the trained strategy model include: receiving real-time operation data of the power plant dust removal system; calculating the dimension of the state space according to the input requirements of the reinforcement learning model; calculating the dimension of the action space according to the output requirements of the reinforcement learning model; inputting the received real-time operation data into the trained reinforcement learning model to obtain action decisions; and displaying the decision results to the operator, including action decisions, rewards and accumulated rewards.

[0106] In one embodiment of the present invention, the system receives real-time data every 5 seconds and makes corresponding control decisions. This high-frequency decision-making capability enables the system to quickly respond to environmental changes and maintain the stability of the dust removal effect.

[0107] In summary, the power plant dust removal system parameter optimization system and method based on reinforcement learning proposed in the present invention can significantly improve the efficiency and stability of the dust removal system through intelligent data processing, model training and decision making, while ensuring the safe operation of the system. This method is not only applicable to power plant dust removal systems, but can also be extended to other similar industrial control scenarios, and has broad application prospects.

[0108] In order to verify the superiority of the power plant dust removal system parameter optimization method based on reinforcement learning of the present invention, we conducted comparative experiments of the embodiment and the comparative example. The following is a detailed experimental process and result analysis.

[0109] Example 1: A method for optimizing dust removal system parameters in a power plant based on reinforcement learning according to the present invention

[0110] We implemented the reinforcement learning-based dust removal system parameter optimization method of the present invention on a 1000MW unit of a thermal power plant. The system includes a data acquisition and preprocessing module, a reinforcement learning training module, an online decision module, a safety assurance module, and a feedback optimization module. We used the Q-learning algorithm as the core of reinforcement learning, with the initial learning rate α set to 0.1 and the discount factor γ set to 0.9. The system receives real-time data every 5 seconds and makes corresponding control decisions.

[0111] Comparative Example 1: Traditional PID Control Method

[0112] On the same 1000MW unit, we used the traditional PID (proportional-integral-differential) control method for comparison. The parameters of the PID controller were adjusted using the Ziegler-Nichols tuning method to obtain the relatively optimal control effect.

[0113] We conducted a 30-day continuous operation test on these two methods. The main test indicators included dust removal efficiency, energy consumption, SO2 emission concentration and NO x Emission concentration. Testing standards and methods are as follows:

[0114] 1. Dust removal efficiency: measured by weight method in accordance with GB / T16157-1996 standard.

[0115] 2. Energy consumption: The total power consumption of the dust removal system is recorded by an electric energy meter.

[0116] 3. SO2 emission concentration: Use the ultraviolet differential method to conduct online continuous monitoring in accordance with HJ57-2017 standard.

[0117] 4. NO x Emission concentration: chemiluminescence method is used for online continuous monitoring in accordance with HJ693-2014 standard.

[0118] The following is a comparison table of the test results:

[0119] index Example 1 Comparative Example 1 Average dust removal efficiency 99.7％ 98.2％ Average daily energy consumption 12500kWh 14800kWh <![CDATA[Average SO2 emission concentration]]> <![CDATA[28mg / m 3 ]]> <![CDATA[33mg / m 3 ]]> <![CDATA[NO x Average emission concentration]]> <![CDATA[42mg / m 3 ]]> <![CDATA[48mg / m 3 ]]> System response time Average 7 seconds Average 25 seconds Parameter adjustment frequency Every 5 seconds Fixed parameters

[0120] It can be seen from the test results that the method of the present invention is significantly better than the traditional PID control method in all indicators. The specific analysis is as follows:

[0121] 1. Dust removal efficiency: The average dust removal efficiency of the method of the present invention reaches 99.7%, which is 1.5 percentage points higher than that of traditional PID control. This improvement may seem small, but in the actual operation of large thermal power plants, it means that several tons of dust emissions can be reduced every day, which is of great significance to improving environmental quality.

[0122] 2. Energy consumption: The average daily energy consumption of the method of the present invention is 2300kWh lower than that of the traditional method, which is equivalent to saving about 15.5% of energy. This not only reduces operating costs, but also reduces carbon emissions, meeting the requirements of energy conservation and emission reduction.

[0123] 3. Emission concentration: SO2 and NO x The average emission concentration decreased by 5 mg / m 3 and 6 mg / m 3 Although the reduction may not seem large, considering the huge flue gas emissions from large thermal power plants, this reduction actually means that hundreds of tons of pollutant emissions can be reduced each year.

[0124] 4. System response time: The average response time of the method of the present invention is only 7 seconds, which is much faster than the 25 seconds of traditional PID control. This means that the system can adapt to changes in working conditions more quickly and effectively prevent the emission of transient pollutants exceeding the standard.

[0125] 5. Parameter adjustment frequency: The method of the present invention can adjust system parameters every 5 seconds, while traditional PID control uses fixed parameters. This dynamic adjustment capability enables the system to better cope with complex and changing working conditions and maintain the best operating state.

[0126] These test results fully demonstrate the superiority of the method of the present invention. The parameter optimization system based on reinforcement learning can adaptively learn the optimal control strategy and continuously optimize the operating parameters, thereby significantly reducing energy consumption and pollutant emissions while ensuring high dust removal efficiency.

[0127] It is worth mentioning that the advantages of the method of the present invention in terms of system response speed and parameter adjustment frequency reflect its intelligence and adaptability. This feature enables the dust removal system to quickly respond to the impact of factors such as coal quality changes and load fluctuations, and maintain a stable and efficient operating state.

[0128] In summary, the power plant dust removal system parameter optimization method based on reinforcement learning of the present invention is not only superior to the traditional method in various technical indicators, but also shows strong environmental and economic benefits. This intelligent control method provides a new technical path for the clean and efficient operation of thermal power plants and has broad application prospects.

[0129] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A power plant dust removal system parameter optimization system based on reinforcement learning, characterized in that: It includes data collection and preprocessing module, reinforcement learning training module, online decision-making module, security assurance module and feedback optimization module, among which: The data acquisition and preprocessing module is used to collect and preprocess the historical operation data of the power plant dust removal system; The reinforcement learning training module is connected to the data acquisition and preprocessing module, and is used to train the control strategy of the dust removal system using a reinforcement learning algorithm based on the preprocessed historical operation data; The online decision-making module is connected to the reinforcement learning training module and is used to receive real-time operation data and make action decisions based on the trained strategy model; The security assurance module is connected to the online decision-making module to ensure the stability and security of the online decision-making module; The feedback optimization module is connected to the online decision-making module and the safety assurance module, and is used to collect decision-making results and actual operation data, and perform feedback evaluation and strategy optimization.

2. The system according to claim 1, characterized in that The data acquisition and preprocessing module includes a data collection unit, a data cleaning unit, an outlier processing unit, a feature extraction unit and a data partitioning unit, wherein: The data collection unit is used to collect historical operation data of the power plant dust removal system through sensors, databases and data cleaning tools; The data cleaning unit is connected to the data collection unit and is used to clean and convert the historical operation data to ensure the quality and integrity of the data; The outlier processing unit is connected to the data cleaning unit and is used to identify and process outliers using statistical methods and machine learning algorithms; The feature extraction unit is connected to the abnormal value processing unit and is used to extract key features from the historical operation data, including real-time data and set values, as inputs to the reinforcement learning training module; The data division unit is connected to the feature extraction unit and is used to divide the historical operation data into a learning data set and a test data set.

3. The system according to claim 1, characterized in that The reinforcement learning training module includes a model selection unit, a model training unit, a model evaluation unit and an iterative optimization unit, wherein: The model selection unit is used to select an appropriate reinforcement learning model, including a Q-learning algorithm, and initialize it in an environment simulation; The model training unit is connected to the model selection unit and is used to train the selected reinforcement learning model in the environment simulation; The model evaluation unit is connected to the model training unit and is used to evaluate the trained reinforcement learning model on the test data set and calculate its performance index; The iterative optimization unit is connected to the model evaluation unit and is used to adjust the parameters and hyperparameters of the model based on the evaluation result and perform iterative optimization.

4. The system according to claim 1, characterized in that The online decision module includes a data receiving unit, a state space calculation unit, an action space calculation unit, a decision calculation unit and a result display unit, wherein: The data receiving unit is used to receive real-time operation data of the power plant dust removal system; The state space calculation unit is connected to the data receiving unit and is used to calculate the dimension of the state space according to the input requirements of the reinforcement learning model; The action space calculation unit is connected to the state space calculation unit and is used to calculate the dimension of the action space according to the output requirements of the reinforcement learning model; The decision calculation unit is connected to the action space calculation unit, and is used to input the received real-time operation data into the trained reinforcement learning model to obtain an action decision; The result display unit is connected to the decision calculation unit and is used to display the decision result to the operator, including action decision, reward and accumulated reward.

5. The system according to claim 1, characterized in that The safety assurance module includes a safety indicator setting unit, a safety monitoring unit, an alarm generating unit and a safety control unit, wherein: The safety index setting unit is used to set the indexes related to the safety of the power plant dust removal system, including but not limited to limiting SO2 / NO x Operating time during which emissions exceed warning values; The safety monitoring unit is connected to the safety index setting unit and is used to monitor the actual data of the power plant dust removal system in real time and track the change trend of the index; The alarm generating unit is connected to the safety monitoring unit and is used to generate an alarm when the actual data exceeds a predetermined safety index, including but not limited to SO2 / NO x The operation time exceeding the warning value exceeds 5 minutes; The safety control unit is connected to the alarm generating unit, and is used to take necessary safety control measures when receiving the alarm information, including but not limited to immediately stopping the dust removal system and notifying the management personnel.

6. The system according to claim 1, characterized in that The feedback optimization module includes a data collection unit, a strategy evaluation unit and a strategy optimization unit, wherein: The data collection unit is used to collect the results of the online decision-making module and the actual operation data of the power plant dust removal system, including action decisions, rewards, cumulative rewards, and safety indicators; The strategy evaluation unit is connected to the data collection unit and is used to perform feedback evaluation based on the collected data; The strategy optimization unit is connected to the strategy evaluation unit and is used to perform strategy optimization according to the feedback evaluation result, including adjusting the parameters and hyperparameters of the reinforcement learning model, improving the calculation method of the state space and action space, and repeating the training and evaluation process until the predetermined optimization goal is met.

7. A method for optimizing parameters of a power plant dust removal system based on reinforcement learning based on the system according to any one of claims 1 to 6, characterized in that: The following steps are involved: Through the data acquisition and preprocessing module, the historical operation data of the power plant dust removal system is collected and preprocessed; Through the reinforcement learning training module, based on the pre-processed historical operation data, a reinforcement learning algorithm is used to train the control strategy of the dust removal system; Receiving real-time operation data through an online decision-making module, and making action decisions based on the trained strategy model; Through the security module, the stability and security of the online decision-making module are ensured; Through the feedback optimization module, the decision results and actual operation data are collected to conduct feedback evaluation and strategy optimization.

8. The method according to claim 7, characterized in that The steps of collecting and preprocessing the historical operation data of the power plant dust removal system through the data acquisition and preprocessing module include: Collect historical operating data of power plant dust removal systems through sensors, databases and data cleaning tools; Clean and convert the historical operation data to ensure the quality and integrity of the data; Use statistical methods and machine learning algorithms to identify and handle outliers; Extracting key features from the historical operation data, including real-time data and set values, as input to the reinforcement learning training module; The historical operation data is divided into a learning data set and a test data set.

9. The method according to claim 7, characterized in that: The step of using a reinforcement learning algorithm to train the control strategy of the dust removal system based on the pre-processed historical operation data through the reinforcement learning training module includes: Select an appropriate reinforcement learning model, including the Q-learning algorithm, and initialize it in the environment simulation; In the environment simulation, training the selected reinforcement learning model; Evaluate the trained reinforcement learning model on the test dataset and calculate its performance indicators; Based on the evaluation results, the parameters and hyperparameters of the model are adjusted and iterative optimization is performed.

10. The method according to claim 7, characterized in that The step of receiving real-time operation data through the online decision module and making an action decision based on the trained strategy model includes: Receive real-time operating data of the power plant dust removal system; Calculating the dimension of the state space according to the input requirements of the reinforcement learning model; Calculating the dimension of the action space according to the output requirements of the reinforcement learning model; Inputting the received real-time operation data into the trained reinforcement learning model to obtain an action decision; The decision results are presented to the operator, including action decisions, rewards, and accumulated rewards.

Images