A deep learning combustion optimization control method and system based on burner working conditions
The burner and boiler neural network model was constructed through deep learning methods, which solved the problem of fine regulation of boiler combustion optimization control system when coal quality fluctuations and load changes, achieved efficient combustion control and pollution reduction, and improved the stability of the boiler and Security.
Patent Information
- Application Number
- CN202410707850.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-03
- Publication Date
- 2024-12-17
- Estimated Expiration
- 2044-06-03
AI Technical Summary
The existing boiler combustion optimization control system is difficult to achieve fine regulation when facing coal quality fluctuations and load changes, and the data processing is insufficient, multi-objective optimization selection is one-sided, model training is time-consuming, system stability and robustness are insufficient, and integration is difficult. Lack of friendly human-computer interaction interface.
Using deep learning methods based on the working conditions of the burner, the burner and boiler neural network models are constructed, and the parameter optimization is optimized using genetic algorithms and stochastic gradient descent algorithms. Combined with multi-objective optimization algorithms, the combustion parameters are monitored and adjusted in real time, and integrated into the power plant DCS In the system, a friendly operating interface is provided.
The optimal combustion control of each burner is achieved, reducing coal and energy consumption, reducing pollution emissions, improving boiler stability and reliability, timely detecting and preventing temperature damage, and ensuring safe operation.
Smart Images
Figure CN118466427B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of thermal automation control technology, and in particular to a deep learning combustion optimization control method and system based on burner working conditions. Background Art
[0002] In modern power generation, coal-fired power plant boilers are one of the main power generation equipment. Their combustion efficiency and emission performance are of great significance to energy utilization and environmental protection. However, power plant boilers face the following challenges in actual operation:
[0003] 1. Challenges of coal quality changes. Due to the diversity of coal sources, power plants often face large fluctuations in coal quality (such as calorific value, moisture, ash and sulfur content). These fluctuations directly affect combustion stability and efficiency, and thus affect the economy and reliability of power plants. In order to adapt to changes in coal quality, power plants adopt a strategy of mixed coal use, but the mixing ratio and combination of different types of coal have a complex effect on the combustion process, making it difficult to achieve optimal combustion. Changes in coal quality require the boiler to make fine combustion adjustments, including adjustments to the air and fuel supply, to maintain efficient combustion. Different coal qualities place higher demands on the adaptability of combustion equipment, requiring specific burner designs or adjustment devices to adapt to different coal qualities.
[0004] 2. Power plant boilers face the challenges of deep peak load regulation and low-load operation during operation. Low-load operation may lead to a series of problems such as decreased boiler combustion stability, incomplete combustion, increased deviation of water-cooled wall and heating surface flue gas temperature, and increased pollutant emissions. For operators, low-load operation requires a higher level of operating skills and experience to ensure the safe and stable operation of the boiler. Accordingly, the control system also needs to be adjusted more finely and frequently, which puts higher requirements on the performance of the control system.
[0005] Since the burner is the core equipment of the boiler, the combustion conditions near it directly determine the thermal efficiency of the boiler and directly reflect the combustion stability and safety of the boiler. Existing monitoring methods often cannot provide real-time operating data for the combustion conditions of each burner. Therefore, the traditional control method often distributes fuel equally to each layer of burners, and in the control of the air supply system, it relies on empirical air distribution strategies and lacks fine-tuning for the specific conditions of individual burners. This control method lacks the ability of self-learning and adaptive adjustment, and it is difficult to achieve the optimal combustion state. It is even more powerless when facing coal quality fluctuations and rapid load changes.
[0006] At present, boiler combustion optimization control systems are usually used to solve the above problems. This control system has model self-learning capabilities and can automatically adjust combustion parameters according to real-time data to improve boiler efficiency, reduce NOx emissions, and ensure the safe operation of the boiler heating surface. Through real-time monitoring of the working conditions of the burners, fine-tuning the allocation of coal and air volume for each layer of burners, and determining the optimal air-coal ratio, the corresponding problems can be effectively solved, which is of great significance for improving the economy, environmental protection and operational safety of power plants. However, this control system still has the following problems that need to be solved:
[0007] 1. There are significant deficiencies in data requirements and processing. Current boiler combustion optimization methods often rely on operating parameter records in the DCS, or other limited off-furnace detection data, and do not pay attention to the operating data near the burner. In addition, the existing data may be interfered by a variety of external factors, resulting in low data quality. Current optimization control systems usually require a large amount of data for training and optimization, especially in boiler combustion optimization, where many variables and parameters need to be considered. Data acquisition, preprocessing, and cleaning can be complex and may be limited by data quality and real-time performance.
[0008] 2. The problem of target optimization selection in multi-objective optimization. Most methods focus on a single or a few targets such as boiler combustion efficiency and NOx emissions, while ignoring other important indicators such as combustion stability and coal powder distribution rate in the same layer. This one-sided optimization strategy may cause problems such as biased burning and unstable combustion in the actual operation of the boiler, affecting the overall performance and safety of the boiler. In addition, the existing methods also lack attention to the temperature control of each burner in the same layer, and cannot effectively solve the temperature difference and biased burning between burners.
[0009] 3. The accuracy and stability of current optimization control systems may be affected by many factors, including data quality, model design, parameter adjustment, etc. In practical applications, data models may face problems such as overfitting, underfitting, or failure to converge, resulting in poor optimization results or failure to achieve stable control.
[0010] 4. In boiler combustion control, the combustion process needs to be monitored and adjusted in real time to adapt to the changing working conditions and environmental conditions. However, the training and optimization process of the model in the current optimization control system may be time-consuming, and the prediction and optimization results of the model may not meet the real-time requirements, resulting in slow or untimely control response.
[0011] 5. System stability and robustness guarantee issues. The boiler combustion process is disturbed and affected by a variety of external factors, such as changes in fuel composition and climate conditions. Therefore, the optimization control system needs to have sufficient system stability and robustness to cope with these external disturbances and maintain stable operation of the system. Although the current optimization control system can provide effective optimization control strategies, in practical applications, ensuring system stability and robustness is still a challenge.
[0012] 6. Difficulties in system integration and compatibility. The boiler combustion optimization control system needs to be integrated with the existing power plant control system to achieve comprehensive monitoring and control. Therefore, it is necessary to consider the compatibility and integration difficulty of the optimization system with the existing system. Although the current optimization control system has strong optimization capabilities, it may face problems such as interface compatibility and data format conversion when integrated into the existing system, resulting in increased integration difficulty and reduced system stability.
[0013] 7. Lack of friendly human-computer interaction interface. In practical applications, operators need to monitor and adjust the boiler combustion process through a friendly human-computer interaction interface. Therefore, the optimization control system needs to design a corresponding operation interface and human-computer interaction method so that operators can use and understand it easily. Although the current optimization control system can provide efficient optimization algorithms, the design of the operation interface and user experience still cannot meet user needs. Summary of the invention
[0014] The purpose of the present invention is to provide a deep learning combustion optimization control method and system based on the working conditions of the burner, so as to solve the above-mentioned problems existing in the prior art.
[0015] In order to achieve the above object, the technical solution adopted by the present invention is as follows:
[0016] A deep learning combustion optimization control method based on burner working conditions comprises the following steps:
[0017] S1, single burner control:
[0018] Based on the collected operating data of each burner in the boiler and the DCS operation data of the power plant, a burner training data set is constructed for the burner neural network model corresponding to each burner, and the training data set is input into the burner neural network model corresponding to each burner for training. During the training process, the genetic algorithm is used to tune the parameters, and the burner neural network model with the optimal parameters is obtained by evaluating the burner neural network model.
[0019] S2. Whole boiler control:
[0020] A boiler neural network model is constructed based on the burner neural network model with the best parameters. A boiler training data set is constructed based on the real-time acquired historical operation data of the boiler. The boiler training set is input into the boiler neural network model for training. During the training process, the stochastic gradient descent algorithm is used to optimize the parameters, and the boiler neural network model with the best parameters is obtained by evaluating the boiler neural network model.
[0021] Preferably, the burner operating condition data include the temperature near the burner, ignition distance, combustion stability, and coal powder distribution rate; the power plant DCS operation data include the primary wind speed of the pulverizer, the primary air temperature, the secondary air volume, the burnout air volume, and the coal supply amount.
[0022] Preferably, the burner training data set includes a plurality of training samples, and the training samples include input features and corresponding output parameters;
[0023] The input characteristics include but are not limited to coal quality information, unit load, primary air speed, primary air temperature, secondary air volume, burnout air volume, coal mill current, coal feed rate, burner swing angle, and induced draft fan current;
[0024] The output parameters include burner ignition point temperature, ignition distance, combustion stability, fly ash carbon content, superheated steam temperature, reheated steam temperature, exhaust gas temperature, flue gas oxygen content, NOx content, and heat exchange surface wall temperature;
[0025] Each training sample contains N input features and corresponding N output parameters.
[0026] Preferably, the burner neural network model is a neural network model having two fully connected layers;
[0027] The input layer of the model contains the same number of neurons as the input features, using ReLU as the activation function; the output layer of the model contains the same number of neurons as the output parameters;
[0028] During the training process of the Burner neural network model, the Adam optimizer is selected to adjust the weights and biases of the neural network model to minimize the mean square error as the loss function; the speed and effect of training are controlled by setting an appropriate learning rate, batch size, and number of training cycles.
[0029] Preferably, the training process of the burner neural network model is:
[0030] Each training sample of the training data set is input into the burner neural network model, and the back propagation algorithm is trained. The loss function is used to evaluate the performance of the burner neural network model in each training cycle, and the parameters of the burner neural network model are adjusted according to the feedback of the Adam optimizer. As the training progresses, the burner neural network model gradually learns the complex relationship between input features and output parameters to accurately predict the working state of a single burner. During the training process, the genetic algorithm is used to tune the parameters, and the training is stopped when the burner neural network model reaches a preset training cycle or the loss function reaches a satisfactory level. The burner neural network model in the training process is evaluated, and the generalization ability of the burner neural network model is verified using the verification data set, and finally a burner neural network model with optimal parameters is obtained.
[0031] Preferably, the burner neural network models with the best parameters corresponding to all burners in the boiler are connected to a main model to form a boiler neural network model; the burner neural network models with the best parameters adopt a parallel structure, and each burner neural network model with the best parameters serves as a branch of the main model to receive the input characteristics of the corresponding burner;
[0032] The real-time operation data of the boiler includes burner operating data, power plant DCS operation data, and boiler load data; during the model training process, key features reflecting the operation status of the boiler can be extracted from these data, including temperature uniformity in the same layer, uniformity of pulverized coal distribution rate, combustion efficiency, NOx emissions, and cooling water volume;
[0033] The output of the boiler neural network model includes burner gas supply, air volume regulation, and coal powder distribution.
[0034] Preferably, step S2 also includes monitoring and diagnosing abnormal conditions during boiler operation using a boiler neural network model with optimal parameters; the specific process is:
[0035] Collect various parameters of burner operation in real time, including but not limited to primary air velocity, primary air temperature, secondary air volume, burnout air volume, and coal supply; extract key features from the collected parameters, including but not limited to temperature uniformity in the same layer, uniformity of pulverized coal distribution rate, combustion efficiency, NOx emissions, and desuperheating water volume. These key features can reflect the operating status of the boiler;
[0036] Use historical data to train the boiler neural network model so that it learns the data pattern under normal operating conditions; input the real-time collected data into the trained boiler neural network model with optimal parameters, and determine whether the boiler has an abnormality by comparing the difference between the real-time data of the boiler and the model prediction results; if the difference between the real-time data of the boiler and the model prediction results exceeds the preset threshold, the diagnosis result is marked as abnormal, and for the diagnosed abnormality, the boiler neural network model further analyzes the cause and automatically adjusts the combustion parameters or provides alarms and operation suggestions to the operator to solve the abnormal situation and ensure the safe and stable operation of the boiler;
[0037] If the abnormal situation is caused by, but not limited to, changes in coal quality, burner failure, or uneven air volume, the solution includes, but is not limited to, adjusting the air volume, coal volume, and burner configuration to optimize combustion efficiency and uniformity; if the abnormal situation is caused by equipment failure, the corresponding equipment needs to be maintained or replaced.
[0038] Preferably, step S2 also includes connecting the boiler neural network model with optimal parameters to the power plant DCS operation system, and automatically adjusting the power plant DCS parameters according to the actual operation of the boiler to adjust the power generation load to avoid exceeding the safe operation range of the boiler and improve the boiler combustion efficiency.
[0039] Preferably, before constructing the training data set, the collected data needs to be preprocessed, including smoothing and stabilization, removing out-of-limit values, and interpolating missing points. The processed data is stored in a local database.
[0040] The present invention also aims to provide a deep learning combustion optimization control system based on the working condition of the burner, the system is used to implement any of the above methods, the system includes:
[0041] Burner operating data acquisition module: including a plurality of multi-spectral sensors corresponding to each burner; each multi-spectral sensor obtains the operating data of the corresponding burner by converting the optical signal into a digital signal, and stores it in the server database;
[0042] DCS data acquisition module: used to collect power plant DCS operation data;
[0043] Burner control module: It is used to construct a burner training data set for the burner neural network model corresponding to each burner based on the collected operating data of each burner in the boiler and the DCS operation data of the power plant, and input the training data set into the burner neural network model corresponding to each burner to train it. During the training process, the genetic algorithm is used to tune the parameters, and the burner neural network model with the optimal parameters is obtained by evaluating the burner neural network model;
[0044] Boiler control module: used to build a boiler neural network model based on the burner neural network model with the best parameters, build a boiler training data set based on the boiler operation data obtained in real time, input the boiler training set into the boiler neural network model and optimize it by stochastic gradient descent. During the training process, a multi-objective optimization algorithm is used to optimize the parameters, and the boiler neural network model with the best parameters is obtained by evaluating the boiler neural network model;
[0045] Automatic control module: It is used to obtain various parameters under the optimal state based on the boiler neural network model with optimal parameters, and use these optimal parameters to control the DCS operation of the power plant.
[0046] The beneficial effects of the present invention are as follows: 1. The neural network model of the present invention can comprehensively consider the influence of coal quality data, boiler load and other operating parameters, provide the best combustion control setting value for each burner under each working condition, and ensure that the boiler operates in the best state. 2. The present invention uses genetic algorithm and multi-objective optimization algorithm to optimize the parameters of the corresponding model respectively, and can adjust various control parameters under the premise of ensuring combustion safety and stability to achieve the best combustion effect and energy utilization efficiency. 3. The present invention can reduce coal consumption and energy consumption, reduce operating costs, and improve the economic benefits of the power plant by optimizing the combustion control setting value. Reduce pollution emissions: The optimized combustion parameters can reduce the amount of emissions, reduce pollution to the environment, and improve the environmental protection performance of the power plant. 4. The present invention can quickly identify and adjust the working state of the burner with the help of a deep learning combustion optimization model, effectively reducing the situation of poor combustion and unstable combustion. By real-time monitoring and adjusting the combustion parameters, it is ensured that the burner operates in the best working state and improves the stability and reliability of the boiler. 5. The present invention can timely detect and prevent local temperature damage to the water-cooled wall. By real-time monitoring of combustion parameters and flame status, timely adjustment of combustion control parameters can be made to avoid damage caused by local overheating and safety issues such as pipe burst, thus ensuring the safe operation of the boiler. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a principle flow chart of the method in the embodiment of the present invention;
[0048] Figure 2 It is a structural diagram of the burner operating condition data acquisition module in an embodiment of the present invention. DETAILED DESCRIPTION
[0049] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation methods described herein are only used to explain the present invention and are not used to limit the present invention.
[0050] Embodiment 1
[0051] like Figure 1 As shown, in this embodiment, a deep learning combustion optimization control method based on the working condition of the burner is provided, which includes the following three parts:
[0052] 1. Data Acquisition
[0053] In this embodiment, relevant data contents such as burner operating data and power plant DCS operation data are obtained by using sensors or other technical means.
[0054] In this embodiment, the burner operating condition data includes the temperature near the burner, ignition distance, combustion stability, and coal powder distribution rate; the power plant DCS operation data includes the primary wind speed, primary air temperature, secondary air volume, burnout air volume, and coal supply of the pulverizer.
[0055] 2. Data Processing
[0056] In this embodiment, the collected relevant data needs to be smoothed and stabilized, out-of-limit values are removed, missing points are interpolated, and other related processing is performed, and then stored in the local database as samples to participate in the neural network model training later.
[0057] For a continuous signal f(t), use the following formula to transform it into F(ω):
[0058]
[0059] For a discrete signal f(n), use the following formula to transform it into F[k]:
[0060]
[0061] Where j is the imaginary unit, w is the angular frequency, N is the number of samples in the signal, and n and k are indices in the discrete time and frequency domains, respectively.
[0062] 3. Single burner control
[0063] Based on the collected operating condition data of each burner in the boiler and the DCS operation data of the power plant, a burner training data set is constructed for the burner neural network model corresponding to each burner. The training data set is input into the burner neural network model corresponding to each burner for training. During the training process, the genetic algorithm is used to tune the parameters, and the burner neural network model is evaluated to obtain the burner neural network model with the optimal parameters.
[0064] In this embodiment, the burner training data set includes a plurality of training samples, and the training samples include input features and corresponding output parameters;
[0065] The input characteristics include coal quality information, unit load, primary air speed, primary air temperature, secondary air volume, burnout air volume, coal mill current, coal feed rate, burner swing angle, and induced draft fan current.
[0066] The output parameters include burner ignition point temperature, ignition distance, combustion stability, fly ash carbon content, superheated steam temperature, reheated steam temperature, exhaust temperature, flue gas oxygen content, NOx content, and heat exchange surface wall temperature.
[0067] Collect a sufficient number of historical data samples, each sample containing N (e.g., 10) input features and corresponding N (e.g., 10) output parameters.
[0068] In this embodiment, the relevant important data in the sample are explained as follows:
[0069] (1) Temperature: The burner ignition point temperature, superheated steam temperature and reheated steam temperature are used as burner temperature indicators. These temperature parameters directly reflect the combustion conditions and thermal state inside the burner. By comparing with the set temperature range, it is possible to evaluate whether the burner temperature is in a suitable working state.
[0070] (2) Ignition distance: Ignition distance refers to the distance between the burner nozzle and the pulverized coal flame. This indicator directly affects the stability and efficiency of combustion. The ignition distance is predicted using the model, and the real-time measured data is compared with the predicted value to evaluate the ignition state of the burner. If the measured value deviates greatly from the predicted value, there is a poor ignition or unstable combustion.
[0071] (3) Combustion stability: Combustion stability refers to the continuity and uniformity of the flame during the combustion process. The stability of the burner is evaluated by real-time monitoring of the flame and combustion. The real-time collected data such as pressure and temperature are combined with parameters such as flue gas oxygen content predicted by the model to determine whether the burner is in a stable working state. The monitoring of pressure and temperature is a parameter directly related to the safety and thermal state of the burner, while the flue gas oxygen content is used to assist in judging the combustion efficiency and completeness of combustion. Therefore, in practical applications, these parameters need to be considered comprehensively to obtain a comprehensive assessment of the burner stability. If the flame shape is uneven or the combustion noise is abnormal, there may be a problem of unstable combustion.
[0072] Combustion stability is expressed using the following formula:
[0073]
[0074] Where M is the number of sampling points, Pj and Tj are the pressure and temperature of the jth sampling point, respectively, and P and T are the average pressure and temperature of all sampling points, respectively.
[0075] In this embodiment, the burner neural network model is a neural network model with two fully connected layers; the input layer of the model contains neurons with the same number of input features (e.g., 10), and uses ReLU as the activation function; the output layer of the model contains neurons with the same number of output parameters (e.g., 10).
[0076] During the training process of the Burner neural network model, the Adam optimizer is selected to adjust the weights and biases of the neural network model to minimize the mean square error (MSE) as the loss function; the speed and effect of training are controlled by setting an appropriate learning rate, batch size, and number of training cycles.
[0077] In this embodiment, each training sample of the training data set is input into the burner neural network model, and the back propagation algorithm is trained. In each training cycle, the performance of the model is evaluated by calculating the model evaluation index, and the parameters of the model are adjusted according to the feedback of the Adam optimizer. As the training progresses, the burner neural network model gradually learns the relationship between the input features and the output parameters, and accurately predicts the working state of the corresponding burner. During the training process, the genetic algorithm is used to tune the parameters, and the training is stopped when the burner neural network model reaches a preset training cycle or the loss function reaches a satisfactory level. The model in the training process is evaluated, and the validation set can be used to verify the generalization ability of the model to obtain a burner neural network model with optimal parameters.
[0078] In the process of parameter tuning by genetic algorithm, the burner neural network model with the best fitness, minimum mean square error in each generation and minimum complexity is taken as the burner neural network model with the best parameters.
[0079] In this embodiment, after the model training is completed, the real-time collected data can be used to perform frequency domain analysis on the burner operation data using Fourier transform to identify and remove noise components and extract key frequency features in the combustion process. These features will then be used as input to the neural network model to more accurately predict and evaluate the working status of the burner.
[0080] In this embodiment, the hierarchical structure of the neural network model is as follows:
[0081] (1) Dense layer formula:
[0082] The Dense layer is a fully connected layer used to build a neural network model. Its formula is:
[0083] output=act i vat i on(i nput×kerne l+bi as)
[0084] Among them, output is the output of the layer, action is the activation function, input is the input of the layer, kernel is the weight, and bias is the bias. Here, ReLU is used as the activation function.
[0085] (2) Dropout layer formula:
[0086] The Dropout layer is a method used to randomly drop neurons during training to reduce overfitting. Although there is no specific formula, it randomly sets a portion of neuron outputs to zero in each training iteration to reduce the complexity of the neural network.
[0087] (3) Adam optimizer formula:
[0088] The Adam optimizer is used to adjust the weights in a neural network to minimize the loss function. The formula for updating the weights is:
[0089] \text{new_we i ght}=\text{ol d_we i ght}-\text{l earn i ng_rate}\times
[0090] \frac{\text{grad i ent}}{\sqrt{\text{exp_avg_sq}+\eps il on}}
[0091] Where new_weight is the updated weight, old_weight is the old weight, length_rate is the learning rate, gradient is the gradient of the loss function, exp_avg_sq is the exponentially weighted average of the squared gradients, and epsilon is a small constant used to prevent division by zero errors.
[0092] 4. Whole boiler control
[0093] A boiler neural network model is constructed based on the burner neural network model with the best parameters. A boiler training data set is constructed based on the historical operation data of the boiler obtained in real time. The boiler training set is input into the boiler neural network model for training. During the training process, the stochastic gradient descent algorithm is used to optimize the parameters. The boiler neural network model with the best parameters is obtained by evaluating the boiler neural network model.
[0094] In this embodiment, the burner neural network models with the best parameters corresponding to all burners in the boiler are connected to a main model to form a boiler neural network model; the burner neural network models with the best parameters adopt a parallel structure, and each burner neural network model with the best parameters serves as a branch of the main model to receive the input characteristics of the corresponding burner;
[0095] The real-time operation data of the boiler include burner operating data, power plant DCS operation data and boiler load data; during the model training process, key features reflecting the operating status of the boiler can be extracted from these data, including temperature uniformity in the same layer, uniformity of coal powder distribution rate, combustion efficiency, NOx emissions, and cooling water volume; the output of the boiler neural network model includes burner air supply, air volume regulation, and coal powder distribution.
[0096] Among them, the uniformity of temperature in the same layer and the uniformity of pulverized coal distribution rate directly reflect the biased burning situation and combustion center control situation inside the furnace, and are important indicators for the combustion control of the entire furnace; combustion efficiency and NOx emissions are commonly used indicators that reflect the combustion performance and environmental protection performance of the boiler. The formulas for each indicator are as follows:
[0097] (1) Temperature uniformity in the same layer
[0098]
[0099] Where N is the number of temperature sensors, Ti is the temperature measured by the i-th sensor, and T is the average temperature measured by all sensors.
[0100] (2) Uniformity of pulverized coal distribution rate
[0101]
[0102] Wherein, K is the number of pulverized coal supply points, Fk is the pulverized coal supply amount of the kth supply point, and F is the average pulverized coal supply amount of all supply points.
[0103] (3) NOx emissions
[0104]
[0105] Where L is the number of monitoring points, and NOx l is the NOx emission at the lth monitoring point.
[0106] In this embodiment, step S2 also includes monitoring and diagnosing abnormal conditions during boiler operation using the boiler neural network model with optimal parameters, specifically,
[0107] Collect various parameters of burner operation in real time, including but not limited to primary air velocity, primary air temperature, secondary air volume, burnout air volume, and coal supply; extract key features from the collected parameters, including but not limited to temperature uniformity in the same layer, uniformity of pulverized coal distribution rate, combustion efficiency, NOx emissions, and desuperheating water volume. These key features can reflect the operating status of the boiler;
[0108] Use historical data to train the boiler neural network model so that it can learn the data patterns under normal operating conditions; input the real-time collected data into the trained boiler neural network model with optimal parameters, and determine whether the boiler has abnormalities by comparing the difference between the real-time data of the boiler and the model prediction results; if the difference between the real-time data of the boiler and the model prediction results exceeds the preset threshold, the diagnosis result will be marked as abnormal. For the diagnosed abnormality, the boiler neural network model will further analyze the cause and automatically adjust the combustion parameters or provide alarms and operation suggestions to the operator to resolve the abnormal situation and ensure the safe and stable operation of the boiler.
[0109] In this embodiment, the solution to the abnormal situation is:
[0110] (1) If the abnormal situation is caused by, but not limited to, changes in coal quality, burner failure, or uneven air volume, the solution includes, but is not limited to, adjusting the air volume, coal volume, and burner configuration to optimize combustion efficiency and uniformity.
[0111] (2) If the abnormal situation is caused by equipment failure, the corresponding equipment needs to be maintained or replaced.
[0112] In this embodiment, the boiler neural network model with the optimal parameters can be connected to the DCS system (closed-loop control) to automatically adjust the power plant DCS parameters according to the actual operation of the boiler to adjust the power generation load to prevent it from exceeding the safe operating range of the boiler and improve the boiler combustion efficiency; or provide alarms and operation suggestions to operators (open-loop control) to resolve abnormal situations and ensure the safe and stable operation of the boiler.
[0113] For abnormal conditions diagnosed, solutions may include: adjusting the air volume, coal volume, burner configuration, etc. to optimize combustion efficiency and uniformity. If the diagnosis results show that the equipment is faulty, corresponding maintenance or component replacement is required. According to the actual operation of the boiler, the power generation load is adjusted to avoid exceeding the safe operating range of the boiler.
[0114] Embodiment 2
[0115] In this embodiment, a deep learning combustion optimization control system based on the working condition of the burner is provided. The system is used to implement the above-mentioned method. The system includes:
[0116] Burner operating data acquisition module: includes multiple multi-spectral sensors corresponding to each burner; each multi-spectral sensor obtains the operating data of the corresponding burner by converting the optical signal into a digital signal and stores it in the server database. Figure 2 .
[0117] DCS data acquisition module: used to collect power plant DCS operation data.
[0118] Burner control module: It is used to construct a burner training data set for the burner neural network model corresponding to each burner based on the collected operating data of each burner in the boiler and the DCS operation data of the power plant, and input the training data set into the burner neural network model corresponding to each burner for training. During the training process, the genetic algorithm is used to tune the parameters, and the burner neural network model is evaluated to obtain the burner neural network model with the optimal parameters.
[0119] Boiler control module: It is used to build a boiler neural network model based on the burner neural network model with the best parameters, build a boiler training data set based on the boiler operation data obtained in real time, input the boiler training set into the boiler neural network model and optimize it by stochastic gradient descent. During the training process, a multi-objective optimization algorithm is used to optimize the parameters, and the boiler neural network model with the best parameters is obtained by evaluating the boiler neural network.
[0120] Automatic control module: It is used to obtain various parameters under the optimal state based on the boiler neural network model with optimal parameters, and use these optimal parameters to control the DCS operation of the power plant.
[0121] In this embodiment, the system also includes an operation interface for realizing human-computer interaction. The operation interface can display the current operating conditions of each burner of the boiler and the operating status of the power plant DCS, and the relevant parameter data during the operation process are displayed on the operation interface. There are automatic mode and manual mode on the operation interface. In automatic mode, the system automatically compares the relevant operating data currently acquired in real time according to the output results of the model, and adaptively adjusts the parameter value so that the boiler operation reaches the optimal state. In manual mode, the operator can adjust the parameter value through manual operation to make the boiler operation reach the corresponding state. At the same time, when the system detects an abnormal situation, the operation interface will promptly display alarms and operation suggestions to the operator so that the operator can solve the abnormal situation in time.
[0122] By adopting the above technical solution disclosed in the present invention, the following beneficial effects are obtained:
[0123] The present invention provides a deep learning combustion optimization control method and system based on the working condition of the burner. The neural network model of the present invention can comprehensively consider the influence of coal quality data, boiler load and other operating parameters, provide the best combustion control setting value for each burner under each working condition, and ensure that the boiler operates in the best state. The present invention uses genetic algorithm and multi-objective optimization algorithm to optimize the parameters of the corresponding model respectively, and can adjust various control parameters under the premise of ensuring combustion safety and stability to achieve the best combustion effect and energy utilization efficiency. By optimizing the combustion control setting value, the present invention can reduce coal consumption and energy consumption, reduce operating costs, and improve the economic benefits of the power plant. Reduce pollution emissions: The optimized combustion parameters can reduce the amount of emissions, reduce pollution to the environment, and improve the environmental protection performance of the power plant. With the help of the deep learning combustion optimization model, the present invention can quickly identify and adjust the working state of the burner, effectively reducing the situation of poor combustion and unstable combustion. By real-time monitoring and adjusting the combustion parameters, it is ensured that the burner operates in the best working state, and the stability and reliability of the boiler are improved. The present invention can timely detect and prevent local temperature damage to the water-cooled wall. By real-time monitoring of combustion parameters and flame status, timely adjustment of combustion control parameters can be made to avoid damage caused by local overheating and safety issues such as pipe burst, thus ensuring the safe operation of the boiler.
[0124] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be considered as the scope of protection of the present invention.
Claims
1. A deep learning combustion optimization control method based on burner working conditions, characterized in that: The following steps are included: S1, single burner control: Based on the collected operating data of each burner in the boiler and the DCS operation data of the power plant, a burner training data set is constructed for the burner neural network model corresponding to each burner, and the training data set is input into the burner neural network model corresponding to each burner for training. During the training process, the genetic algorithm is used to tune the parameters, and the burner neural network model with the optimal parameters is obtained by evaluating the burner neural network model. The burner operating data includes the temperature near the burner, ignition distance, combustion stability, and coal powder distribution rate; the power plant DCS operation data includes the primary wind speed, primary wind temperature, secondary air volume, burnout air volume, and coal supply of the coal mill; The burner training data set includes a plurality of training samples, and the training samples include input features and corresponding output parameters; The input characteristics include but are not limited to coal quality information, unit load, primary air speed, primary air temperature, secondary air volume, burnout air volume, coal mill current, coal feed rate, burner swing angle, and induced draft fan current; The output parameters include burner ignition point temperature, ignition distance, combustion stability, fly ash carbon content, superheated steam temperature, reheated steam temperature, exhaust gas temperature, flue gas oxygen content, NOx content, and heat exchange surface wall temperature; Each training sample contains N input features and corresponding N output parameters; Each burner corresponds to each multi-spectral sensor one by one, and each multi-spectral sensor obtains the operating data of the corresponding burner by converting the optical signal into a digital signal; S2. Whole boiler control: A boiler neural network model is constructed based on the burner neural network model with the best parameters. A boiler training data set is constructed based on the historical operation data of the boiler acquired in real time. The boiler training set is input into the boiler neural network model for training. During the training process, the stochastic gradient descent algorithm is used to optimize the parameters. The boiler neural network model with the best parameters is obtained by evaluating the boiler neural network model. The burner neural network models with the best parameters corresponding to all the burners in the boiler are connected to a main model to form a boiler neural network model; the burner neural network models with the best parameters adopt a parallel structure, and each burner neural network model with the best parameters serves as a branch of the main model to receive the input characteristics of the corresponding burner; Boiler real-time operation data includes burner operating data, power plant DCS operation data, and boiler load data; During the model training process, key features of the boiler operation status can be extracted from these data, including temperature uniformity in the same layer, uniformity of pulverized coal distribution rate, combustion efficiency, NOx emissions, and cooling water volume; The output of the boiler neural network model includes burner gas supply, air volume regulation, and coal powder distribution.
2. The deep learning combustion optimization control method based on the burner working condition according to claim 1 is characterized in that: The burner neural network model is a neural network model with two fully connected layers; The input layer of the model contains the same number of neurons as the input features, using ReLU as the activation function; the output layer of the model contains the same number of neurons as the output parameters; During the training process of the Burner neural network model, the Adam optimizer is selected to adjust the weights and biases of the neural network model to minimize the mean square error as the loss function; the speed and effect of training are controlled by setting an appropriate learning rate, batch size, and number of training cycles.
3. The deep learning combustion optimization control method based on the burner working condition according to claim 2 is characterized in that: The training process of the burner neural network model is: Each training sample of the training data set is input into the burner neural network model, and the back propagation algorithm is trained. The loss function is used to evaluate the performance of the burner neural network model in each training cycle, and the parameters of the burner neural network model are adjusted according to the feedback of the Adam optimizer. As the training progresses, the burner neural network model gradually learns the complex relationship between input features and output parameters to accurately predict the working state of a single burner. During the training process, the genetic algorithm is used to tune the parameters, and the training is stopped when the burner neural network model reaches a preset training cycle or the loss function reaches a satisfactory level. The burner neural network model in the training process is evaluated, and the generalization ability of the burner neural network model is verified using the verification data set, and finally a burner neural network model with optimal parameters is obtained.
4. The deep learning combustion optimization control method based on burner working conditions according to claim 1 is characterized in that: Step S2 also includes monitoring and diagnosing abnormal conditions during boiler operation using the boiler neural network model with optimal parameters; the specific process is: Collect various parameters of burner operation in real time, including but not limited to primary air velocity, primary air temperature, secondary air volume, burnout air volume, and coal supply; extract key features from the collected parameters, including but not limited to temperature uniformity in the same layer, uniformity of pulverized coal distribution rate, combustion efficiency, NOx emissions, and desuperheating water volume. These key features can reflect the operating status of the boiler; Use historical data to train the boiler neural network model so that it learns the data patterns under normal operating conditions; input the real-time collected data into the trained boiler neural network model with optimal parameters, and determine whether the boiler is abnormal by comparing the difference between the real-time data of the boiler and the prediction results of the model; If the difference between the boiler's real-time data and the model's prediction results exceeds the preset threshold, the diagnosis result will be marked as abnormal. For the diagnosed abnormality, the boiler neural network model will further analyze the cause and automatically adjust the combustion parameters or provide alarms and operational suggestions to the operator to resolve the abnormal situation and ensure the safe and stable operation of the boiler. If the abnormal situation is caused by, but not limited to, changes in coal quality, burner failure, or uneven air volume, the solution includes, but is not limited to, adjusting the air volume, coal volume, and burner configuration to optimize combustion efficiency and uniformity; if the abnormal situation is caused by equipment failure, the corresponding equipment needs to be maintained or replaced.
5. The deep learning combustion optimization control method based on burner working conditions according to claim 1 is characterized in that: Step S2 also includes connecting the boiler neural network model with optimal parameters to the power plant DCS operation system, and automatically adjusting the power plant DCS parameters according to the actual operation of the boiler to adjust the power generation load to avoid exceeding the safe operation range of the boiler and improve the boiler combustion efficiency.
6. The deep learning combustion optimization control method based on burner working conditions according to claim 1 is characterized in that: Before constructing the training data set, the collected data needs to be preprocessed, including smoothing and stabilization, removing out-of-limit values, and interpolating missing points. After data processing, it is stored in the local database.
7. A deep learning combustion optimization control system based on the working condition of the burner, characterized in that: The system is used to implement the method described in any one of claims 1 to 6, and the system includes: Burner operating data acquisition module: including a plurality of multi-spectral sensors corresponding to each burner; each multi-spectral sensor obtains the operating data of the corresponding burner by converting the optical signal into a digital signal, and stores it in the server database; DCS data acquisition module: used to collect power plant DCS operation data; Burner control module: It is used to construct a burner training data set for the burner neural network model corresponding to each burner based on the collected operating data of each burner in the boiler and the DCS operation data of the power plant, and input the training data set into the burner neural network model corresponding to each burner to train it. During the training process, the genetic algorithm is used to tune the parameters, and the burner neural network model with the optimal parameters is obtained by evaluating the burner neural network model; Boiler control module: used to build a boiler neural network model based on the burner neural network model with the best parameters, build a boiler training data set based on the boiler operation data obtained in real time, input the boiler training set into the boiler neural network model and optimize it by stochastic gradient descent. During the training process, a multi-objective optimization algorithm is used to optimize the parameters, and the boiler neural network model with the best parameters is obtained by evaluating the boiler neural network model; Automatic control module: It is used to obtain various parameters under the optimal state based on the boiler neural network model with optimal parameters, and use these optimal parameters to control the DCS operation of the power plant.