AI flow field prediction-based self-adaptive regulation and control system for decomposing ammonia-doped rotational flow pulverized coal burner

Abstract

The invention discloses an AI flow field prediction-based self-adaptive regulation and control system for a decomposition ammonia-doped rotational flow pulverized coal burner, and belongs to the field of burner regulation and control. The invention discloses an AI flow field prediction-based self-adaptive regulation and control system for a decomposition ammonia-doped rotational flow pulverized coal burner. The system comprises a multi-mode sensing unit, a flow field prediction unit, a self-adaptive optimization unit and an execution mechanism unit. The problems that in the prior art, ammonia-doped combustion regulation is lagged, multiple targets are difficult to cooperate, and consequently efficiency and environmental protection performance cannot be considered are solved, the multi-mode sensing unit comprehensively collects multi-source operation data of the combustor, the AI model subjected to deep learning is used for accurately predicting changes of an outlet three-dimensional flow field, a temperature field and a component concentration field in advance, and the control accuracy is improved. On the basis, multi-target collaborative optimization of combustion efficiency, NOx emission, ammonia escape and flame stability is taken as a core, optimal operation parameters are dynamically solved, and an execution mechanism is driven to perform accurate adjustment, so that the unification of efficient combustion, ultralow emission and stable operation is effectively realized under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of burner control technology, specifically to an adaptive control system for ammonia-doped swirl pulverized coal burners based on AI flow field prediction. Background Technology

[0002] In the energy and environmental protection field, coal-fired power generation still occupies an important position, but it faces the dual challenges of improving combustion efficiency and reducing pollutant emissions. Ammonia-blended combustion technology, as a promising low-carbon combustion method, effectively reduces carbon dioxide emissions by mixing ammonia as a partial fuel with pulverized coal. However, the introduction of decomposed ammonia into a swirl pulverized coal burner makes the combustion process more complex, with strong interactions between flow field structure, temperature distribution, and component concentration fields, easily leading to problems such as combustion instability, decreased efficiency, increased nitrogen oxide emissions, and ammonia escape. Traditional burner control methods mostly rely on preset operating curves or feedback control based on wired sensors, which are insufficient to cope with the strong nonlinearity, large time delay, and multivariate coupling characteristics of the combustion system after ammonia blending, making it impossible to achieve real-time and accurate optimization of the combustion state. Furthermore, existing technologies lack the ability to accurately predict the three-dimensional flow field inside and at the burner outlet, resulting in delayed control response and limited optimization effects. Therefore, these methods do not meet current needs. To address this, we propose an adaptive control system for ammonia-blended swirl pulverized coal burner based on AI flow field prediction. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive control system for ammonia-blended swirl pulverized coal burner based on AI flow field prediction. The system uses a multi-modal sensing unit to collect real-time burner operating data. A flow field prediction unit jointly trains the AI ​​prediction model based on historical operating data and combustion fluid dynamics simulation data, accurately predicting the three-dimensional flow field structure, temperature field, and component concentration field at the burner outlet. An adaptive optimization unit incorporates a multi-objective optimization function for ammonia-blended combustion, aiming for maximum combustion efficiency, minimum NOx emissions, minimum ammonia escape, and optimal flame stability. The system solves the optimization function based on the output of the flow field prediction unit, generating a control instruction set which is then executed by the actuator unit, completing closed-loop control of the combustion state and solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an adaptive control system for an ammonia-doped swirl pulverized coal burner based on AI flow field prediction, comprising:

[0005] The multimodal sensing unit is configured to collect the operating data of the burner in real time. The operating data includes at least the primary air pulverized coal concentration and flow rate, the secondary air flow rate and swirl intensity, the injection location, flow rate and temperature of the decomposed ammonia, and the flame morphology image and spectral information of the burner outlet area.

[0006] The flow field prediction unit is communicatively connected to the multimodal sensing unit and is configured to jointly train a pre-built AI prediction model based on historical operating data and combustion fluid dynamics simulation data. It also receives the operating data of the multimodal sensing unit as input and outputs predicted data of the three-dimensional flow field structure, temperature field and component concentration field of the burner outlet at a specific time step in the future.

[0007] An adaptive optimization unit is communicatively connected to the flow field prediction unit and is configured to have a built-in multi-objective optimization function for ammonia-blended combustion. The optimization function aims to achieve the highest combustion efficiency, the lowest NOx emissions, the minimum ammonia escape, and the best flame stability. By receiving the output of the flow field prediction unit, the unit solves the optimization function and generates a set of control instructions for the primary air, secondary air, and decomposed ammonia supply systems.

[0008] The actuator unit, which is communicatively connected to the adaptive optimization unit, includes an independently adjustable primary air valve, a secondary air cyclone actuator, and decomposed ammonia gas nozzle flow regulating valves arranged in multiple axial positions. It is configured to execute the set of control instructions issued by the adaptive optimization unit to complete the closed-loop control of the combustion state.

[0009] Furthermore, the flow field prediction unit includes:

[0010] The data collection module is configured to receive burner operation data collected in real time by the multimodal sensing unit, and simultaneously receive historical operation data and combustion fluid dynamics simulation data. Through time series alignment, data cleaning, outlier removal and normalization processing, the multi-source heterogeneous data obtained above are integrated into a unified standardized time series dataset suitable for model training. The processes of time series alignment, data cleaning, outlier removal and normalization processing are existing technologies in this field and are not the inventive solutions of this application, and will not be described in detail here.

[0011] The feature extraction module is configured to extract features from the operational data received by the data collection module, including feature vectors of primary and secondary air velocity and swirl intensity, coal powder concentration distribution characteristics, injection parameter characteristics of decomposed ammonia, and key features in flame morphology images and spectral information (such as flame shape, color, intensity distribution, and intensity of specific bands in the spectrum).

[0012] The model training module is configured to build an AI prediction model based on deep learning algorithms and jointly train the AI ​​prediction model using a standardized time-series dataset integrated by the data fusion and processing module. By adjusting the model's hyperparameters (such as learning rate, batch size, number of network layers, etc.) and training strategies (such as early stopping, regularization, etc.), the performance of the AI ​​prediction model is optimized. At the same time, during the training process, cross-validation is used to evaluate the model's performance to ensure that the AI ​​prediction model can make stable and accurate predictions under different working conditions. The model building and model training processes are existing technologies in this field and are not the inventive solutions of this application, and will not be described in detail here.

[0013] The output module is configured to input the feature data extracted from the running data into the trained AI prediction model. Based on the input feature data, the AI ​​prediction model outputs the predicted data of the three-dimensional flow field structure, temperature field and component concentration field of the burner outlet at a specific time step in the future.

[0014] Furthermore, the AI ​​prediction model adopts an online incremental learning mechanism to compare the actual combustion state data subsequently collected by the multimodal sensing unit with the model prediction data, calculate the prediction error, and automatically trigger the fine-tuning update of the model parameters when the prediction error exceeds a preset threshold, so as to adapt to the dynamic characteristic changes caused by changes in the state of components such as burner coking and wear.

[0015] Furthermore, the AI ​​prediction model is a hybrid architecture that integrates graph neural networks and temporal convolutional networks. The graph neural network is used to model the unstructured spatial topology of the burner interior and outlet region to capture the local flow field structure. The temporal convolutional network is used to extract deep temporal features from the real-time data collected by the multimodal sensing unit. The fusion of graph neural networks and temporal convolutional networks is used to jointly achieve spatiotemporal prediction of future three-dimensional flow field, temperature field, and component concentration field.

[0016] Furthermore, the prediction results output by the AI ​​prediction model are post-processed, including:

[0017] Data decoding is used to convert the encoded data output by AI prediction models into interpretable physical quantities;

[0018] Data validation is used to verify the rationality of the prediction results, including verifying the physical rationality of the temperature field and the range of component concentrations;

[0019] Data output is used to output the prediction results in a standardized format for use by the adaptive optimization unit.

[0020] Furthermore, the adaptive optimization unit includes:

[0021] The data parsing module is configured to receive and parse the prediction data output by the flow field prediction unit, extract feature information related to the optimization target, including temperature distribution, component concentration (especially NOx and ammonia escape), and flow field structure features. At the same time, it receives real-time operating data from the multimodal sensing unit, including the current status of primary air, secondary air, and decomposed ammonia, to take the current operating conditions into account during the optimization process.

[0022] The target setting module is configured with a built-in multi-objective optimization function for ammonia-blended combustion. The optimization objectives are clearly defined as maximizing combustion efficiency, minimizing NOx emissions, minimizing ammonia slip, and optimizing flame stability. Weights are assigned to each optimization objective based on actual operational needs and priorities. In multi-objective optimization, conflicts may exist between objectives, requiring a trade-off mechanism to balance the relationships between them. For example, a penalty factor can be introduced or a multi-objective optimization algorithm (such as Pareto optimization) can be used to handle conflicts between objectives, ensuring that the optimization result achieves an optimal balance among multiple objectives.

[0023] The optimization algorithm configuration module is configured to select a genetic algorithm to solve multi-objective optimization functions. Based on the characteristics and complexity of the optimization problem, the parameters of the genetic algorithm are configured, including population size, crossover probability, and mutation probability. At the same time, the parameters of the genetic algorithm are dynamically adjusted in combination with the output of the AI ​​prediction model to improve optimization efficiency and accuracy. For example, during the optimization process, the exploration and development strategy of the algorithm is adjusted according to the uncertainty of the prediction data.

[0024] The instruction generation module is configured to generate a set of control instructions for the primary air, secondary air, and decomposed ammonia supply systems based on the solution results of the optimization algorithm. The instruction set includes the opening degree of the primary air valve, the adjustment angle of the secondary air cyclone separator, and the specific parameters of the decomposed ammonia nozzle flow rate. The generated control instruction set is then sent to the actuator unit.

[0025] Furthermore, the optimization algorithm configuration module is also configured to adopt a dynamic constraint processing mechanism, which adaptively adjusts the constraints in the optimization process based on the real-time operating status of the burner and the prediction results of the flow field prediction unit. The constraints include, but are not limited to, the upper temperature limit for safe operation of the burner, the explosion-proof limit of ammonia injection, and the minimum swirl intensity to maintain flame stability. The dynamic constraint processing mechanism calculates the boundary margin of each constraint in real time and introduces an adaptive penalty function during the optimization process to ensure that the generated control commands are always within the feasible domain for safe and stable operation of the burner.

[0026] Furthermore, the generated control commands are verified and validated, including:

[0027] Check whether the format of the control commands conforms to the system's preset standards, including the command syntax, data type, and unit, specifically:

[0028] Verify that the parameters in the command (such as damper opening, nozzle flow rate, etc.) are within a reasonable range.

[0029] Check whether the format of the command conforms to the requirements of the communication protocol, such as whether it has the necessary header information, check bits, etc.

[0030] Ensure that all parameters included in the command have clearly defined units to avoid errors caused by inconsistent units.

[0031] Verify whether the parameters in the control commands are within the safe operating range of the equipment, specifically:

[0032] Compare the parameters in the control instructions with the safe operating parameter range provided by the equipment manufacturer. For example, whether the opening of the air valve is between 0% and 100%, and whether the flow rate of the decomposed ammonia nozzle is between the maximum and minimum flow rates allowed by the equipment. If the parameters exceed the safe range, an alarm will be issued and the parameters will be automatically adjusted to the safe range.

[0033] Check for logical contradictions among the parameters in the control commands, specifically:

[0034] Verify whether the adjustments of primary and secondary air are coordinated. For example, if the primary air is increased, does the secondary air need to be adjusted accordingly to maintain stable combustion?

[0035] Check whether the adjustment of the ammonia decomposition gas nozzle flow rate matches the current operating status of the burner (such as temperature and flow rate);

[0036] If a logical contradiction is found, an alarm will be issued and the control instructions will be recalculated;

[0037] The control commands are compared with the real-time operating data provided by the multimodal sensing unit to verify the rationality and feasibility of the commands. Specifically:

[0038] Compare the parameters in the control command with the actual parameters read by the sensor to check for significant differences. For example, if the control command requires an increase in airflow, but the sensor shows that the current airflow is already close to the upper limit, then the control command needs to be re-evaluated. If a difference is found, an alarm is issued and the control command is adjusted.

[0039] Furthermore, after the control command set is issued to the actuator unit, feedback information from the actuator unit is received, such as the status of command execution and changes in actual operating parameters. Based on the feedback information, the effect of the control command is evaluated. If it is found that the expected optimization effect is not achieved after the command is executed, the optimization strategy and control command are adjusted in a timely manner. Through the closed-loop feedback mechanism, it is ensured that the adaptive optimization unit can continuously optimize the combustion process and achieve the best balance between combustion efficiency, NOx emissions, ammonia slip and flame stability.

[0040] Furthermore, it also includes a safety protection unit, which is communicatively connected to the multimodal sensing unit and the adaptive optimization unit, and is configured as follows:

[0041] Real-time monitoring of safety parameters, including but not limited to combustion chamber pressure, extreme temperature values, and ammonia leakage concentration;

[0042] Based on real-time monitored safety parameters, if a dangerous working condition is detected, the control instructions of the adaptive optimization unit are exceeded, and a safety protection instruction is directly sent to the actuator unit.

[0043] It also provides multiple safety redundancy mechanisms, including hardware safety loops and software safety logic, to ensure that the burner can be safely shut down or switched to a safe operating mode in the event of a system failure.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] This invention constructs a closed-loop control system integrating a multimodal sensing unit, an AI flow field prediction unit, an adaptive optimization unit, and an actuator unit. This system enables real-time perception, forward prediction, and intelligent decision-making regarding the operating status of a decomposed ammonia-blended swirl pulverized coal burner. Utilizing an AI model jointly trained with historical data and fluid dynamics simulation, the system can accurately predict the three-dimensional flow field, temperature field, and component concentration field at a specific future time step. This overcomes the control lag problem caused by the large inertia and strong nonlinearity of traditional methods. Based on this, with multi-objective synergistic optimization of combustion efficiency, NOx emissions, ammonia escape, and flame stability as the core, the system automatically generates and executes optimal control commands. This significantly improves combustion efficiency and stability under complex and variable operating conditions, while simultaneously achieving effective synergistic control of nitrogen oxide and unburned ammonia escape emissions. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the adaptive control system for an ammonia-doped swirl pulverized coal burner based on AI flow field prediction, as described in this invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] To address the issues of lagging regulation and difficulty in coordinating multiple objectives in existing ammonia-blended combustion technologies, which result in a trade-off between efficiency and environmental performance, please refer to [link / reference needed]. Figure 1 This embodiment provides the following technical solution:

[0049] An adaptive control system for an ammonia-doped swirl pulverized coal burner based on AI flow field prediction includes:

[0050] The multimodal sensing unit is configured to collect the operating data of the burner in real time. The operating data includes at least the primary air pulverized coal concentration and flow rate, the secondary air flow rate and swirl intensity, the injection location, flow rate and temperature of the decomposed ammonia, and the flame morphology image and spectral information of the burner outlet area.

[0051] The flow field prediction unit is communicatively connected to the multimodal sensing unit and is configured to jointly train a pre-built AI prediction model based on historical operating data and combustion fluid dynamics simulation data. It also receives the operating data of the multimodal sensing unit as input and outputs predicted data of the three-dimensional flow field structure, temperature field and component concentration field of the burner outlet at a specific time step in the future.

[0052] An adaptive optimization unit is communicatively connected to the flow field prediction unit and is configured to have a built-in multi-objective optimization function for ammonia-blended combustion. The optimization function aims to achieve the highest combustion efficiency, the lowest NOx emissions, the minimum ammonia escape, and the best flame stability. By receiving the output of the flow field prediction unit, the unit solves the optimization function and generates a set of control instructions for the primary air, secondary air, and decomposed ammonia supply systems.

[0053] The actuator unit, which is communicatively connected to the adaptive optimization unit, includes an independently adjustable primary air valve, a secondary air cyclone actuator, and decomposed ammonia gas nozzle flow regulating valves arranged in multiple axial positions. It is configured to execute the set of control instructions issued by the adaptive optimization unit to complete the closed-loop control of the combustion state.

[0054] The technical effects of the above solution are as follows: By comprehensively and in real-time collecting burner operating data through a multimodal sensing unit, and utilizing an AI prediction model trained based on historical data and combustion fluid dynamics simulation data by a flow field prediction unit, the system can accurately output predicted data of the three-dimensional flow field structure, temperature field, and component concentration field at the burner outlet for a specific time step in the future. Then, an adaptive optimization unit optimizes the system based on the predicted data with multiple objectives, including maximizing combustion efficiency, minimizing NOx emissions, minimizing ammonia slip, and optimizing flame stability. This generates a precise control command set, which is then used by the actuator unit to achieve closed-loop control of the primary air, secondary air, and decomposed ammonia supply system. This effectively solves the control problem caused by the strong nonlinearity, large time delay, and multivariate coupling characteristics of ammonia-blended swirl pulverized coal burners, significantly improves combustion efficiency, greatly reduces NOx emissions and ammonia slip, ensures flame stability, and thus achieves highly efficient and low-pollution adaptive optimization operation of the burner.

[0055] The AI-based flow field prediction-based adaptive control system for ammonia-doped swirl pulverized coal burner also includes a safety protection unit, which is communicatively connected to the multimodal sensing unit and the adaptive optimization unit, and is configured as follows:

[0056] Real-time monitoring of safety parameters, including but not limited to combustion chamber pressure, extreme temperature values, and ammonia leakage concentration;

[0057] Based on real-time monitored safety parameters, if a dangerous working condition is detected, the control instructions of the adaptive optimization unit are exceeded, and a safety protection instruction is directly sent to the actuator unit.

[0058] It also provides multiple safety redundancy mechanisms, including hardware safety loops and software safety logic, to ensure that the burner can be safely shut down or switched to a safe operating mode in the event of a system failure.

[0059] The technical effects of the above solution are as follows: By introducing a safety protection unit, key safety parameters such as combustion chamber pressure, extreme temperature, and ammonia leakage concentration are monitored in real time. When a dangerous operating condition is detected, the unit can exceed the conventional control commands of the adaptive optimization unit and directly send safety protection commands to the actuator unit. At the same time, through a multi-layered safety redundancy mechanism composed of hardware safety loops and software safety logic, the burner can be immediately and safely shut down or smoothly switch to a safe operating mode when a system failure occurs. This effectively prevents safety accidents that may be caused by equipment failure, control failure, or sudden changes in operating conditions, significantly enhances the reliability and safety of the entire ammonia-blended combustion system under complex operating conditions, and provides a solid guarantee for the long-term stable operation of the burner.

[0060] The flow field prediction unit includes:

[0061] The data collection module is configured to receive burner operation data collected in real time by the multimodal sensing unit, and simultaneously receive historical operation data and combustion fluid dynamics simulation data. Through time series alignment, data cleaning, outlier removal and normalization processing, the multi-source heterogeneous data obtained above are integrated into a unified standardized time series dataset suitable for model training. The processes of time series alignment, data cleaning, outlier removal and normalization processing are existing technologies in this field and are not the inventive solutions of this application, and will not be described in detail here.

[0062] The feature extraction module is configured to extract features from the operational data received by the data collection module, including feature vectors of primary and secondary air velocity and swirl intensity, coal powder concentration distribution characteristics, injection parameter characteristics of decomposed ammonia, and key features in flame morphology images and spectral information (such as flame shape, color, intensity distribution, and intensity of specific bands in the spectrum).

[0063] The model training module is configured to build an AI prediction model based on deep learning algorithms and jointly train the AI ​​prediction model using a standardized time-series dataset integrated by the data fusion and processing module. By adjusting the model's hyperparameters (such as learning rate, batch size, number of network layers, etc.) and training strategies (such as early stopping, regularization, etc.), the performance of the AI ​​prediction model is optimized. At the same time, during the training process, cross-validation is used to evaluate the model's performance to ensure that the AI ​​prediction model can make stable and accurate predictions under different working conditions. The model building and model training processes are existing technologies in this field and are not the inventive solutions of this application, and will not be described in detail here.

[0064] The output module is configured to input the feature data extracted from the running data into the trained AI prediction model. Based on the input feature data, the AI ​​prediction model outputs the predicted data of the three-dimensional flow field structure, temperature field and component concentration field of the burner outlet at a specific time step in the future.

[0065] The technical effects of the above solution are as follows: the data collection module effectively integrates and standardizes real-time operating data, historical data, and combustion fluid dynamics simulation data to construct a high-quality time-series dataset. The feature extraction module accurately extracts multi-dimensional features related to flow field prediction from this dataset. The model training module then constructs and optimizes an AI prediction model based on deep learning algorithms, enabling the AI ​​prediction model to accurately capture the complex nonlinear dynamic characteristics of the combustion process. Finally, the result output module uses the trained AI prediction model to output accurate prediction data of the three-dimensional flow field structure, temperature field, and component concentration field for a specific time step in the future. This provides a reliable data foundation for subsequent adaptive optimization control, thereby improving the system's forward-looking perception and prediction capabilities of the combustion state.

[0066] The AI ​​prediction model adopts an online incremental learning mechanism, which compares the actual combustion state data collected by the multimodal sensing unit with the model prediction data, calculates the prediction error, and automatically triggers the fine-tuning and updating of the model parameters when the prediction error exceeds the preset threshold, so as to adapt to the dynamic characteristic changes brought about by the changes in the state of components such as burner coking and wear.

[0067] The triggering condition for the online incremental learning mechanism is: the prediction error exceeds the threshold for N consecutive sampling periods. N and the threshold are dynamically adjusted according to the stability requirements of the burner's operating conditions to avoid false triggering under transient combustion conditions.

[0068] The AI ​​prediction model is a hybrid architecture that integrates graph neural networks and temporal convolutional networks. The graph neural network is used to model the unstructured spatial topology of the burner interior and outlet region to capture the local flow field structure. The temporal convolutional network is used to extract deep temporal features from the real-time data collected by the multimodal sensing unit. The fusion of graph neural networks and temporal convolutional networks is used to jointly achieve spatiotemporal prediction of future three-dimensional flow field, temperature field and component concentration field.

[0069] The technical effects of the above solution are as follows: By adopting an online incremental learning mechanism, the AI ​​prediction model can compare the predicted data with the subsequently collected combustion state data and calculate the prediction error. When the error continuously exceeds the dynamic adjustment threshold, the model parameters are automatically fine-tuned and updated, thereby effectively adapting to the dynamic characteristics changes of the burner caused by changes in component states such as coking and wear. In addition, through the hybrid architecture of graph neural network and temporal convolutional network, the model fully utilizes the graph neural network's ability to model the unstructured spatial topology of the burner and the temporal convolutional network's ability to extract deep temporal features of multimodal sensor data, thus achieving accurate spatiotemporal joint prediction of the future three-dimensional flow field, temperature field and component concentration field, thereby improving the prediction accuracy and robustness of the model under complex and variable working conditions.

[0070] Post-processing of the prediction results output by the AI ​​prediction model includes:

[0071] Data decoding is used to convert the encoded data output by AI prediction models into interpretable physical quantities;

[0072] Data validation is used to verify the rationality of the prediction results, including verifying the physical rationality of the temperature field and the range of component concentrations;

[0073] Data output is used to output the prediction results in a standardized format for use by the adaptive optimization unit.

[0074] The technical effects of the above solution are as follows: by decoding the data, the prediction results have clear physical meaning and engineering application value; by verifying the physical rationality of the temperature field and the range of component concentrations through data validation, the system effectively identifies and filters out any possible abnormal prediction results, ensuring the reliability and rationality of the output data; and finally, by outputting the data in a standardized format, the system provides the adaptive optimization unit with high-quality and reliable prediction information that can be directly used, thereby ensuring the accuracy of the decision-making basis of the entire control system and the effectiveness of subsequent control commands.

[0075] Adaptive optimization unit, including:

[0076] The data parsing module is configured to receive and parse the prediction data output by the flow field prediction unit, extract feature information related to the optimization target, including temperature distribution, component concentration (especially NOx and ammonia escape), and flow field structure features. At the same time, it receives real-time operating data from the multimodal sensing unit, including the current status of primary air, secondary air, and decomposed ammonia, to take the current operating conditions into account during the optimization process.

[0077] The target setting module is configured with a built-in multi-objective optimization function for ammonia-blended combustion. The optimization objectives are clearly defined as maximizing combustion efficiency, minimizing NOx emissions, minimizing ammonia slip, and optimizing flame stability. Weights are assigned to each optimization objective based on actual operational needs and priorities. In multi-objective optimization, conflicts may exist between objectives, requiring a trade-off mechanism to balance the relationships between them. For example, a penalty factor can be introduced or a multi-objective optimization algorithm (such as Pareto optimization) can be used to handle conflicts between objectives, ensuring that the optimization result achieves an optimal balance among multiple objectives.

[0078] The optimization algorithm configuration module is configured to select a genetic algorithm to solve multi-objective optimization functions. Based on the characteristics and complexity of the optimization problem, the parameters of the genetic algorithm are configured, including population size, crossover probability, and mutation probability. At the same time, the parameters of the genetic algorithm are dynamically adjusted in combination with the output of the AI ​​prediction model to improve optimization efficiency and accuracy. For example, during the optimization process, the exploration and development strategy of the algorithm is adjusted according to the uncertainty of the prediction data.

[0079] The instruction generation module is configured to generate a set of control instructions for the primary air, secondary air, and decomposed ammonia supply systems based on the solution results of the optimization algorithm. The instruction set includes the opening degree of the primary air valve, the adjustment angle of the secondary air cyclone separator, and the specific parameters of the decomposed ammonia nozzle flow rate. The generated control instruction set is then sent to the actuator unit.

[0080] The technical effects of the above solution are as follows: the data analysis module accurately extracts key feature information from the flow field prediction data and real-time operation data, providing comprehensive and accurate input for target optimization. The multi-objective optimization function of ammonia-blended combustion built into the target setting module clarifies the optimization targets of combustion efficiency, NOx emissions, ammonia slip, and flame stability. By introducing a trade-off mechanism to balance the conflicting relationships between multiple objectives, the optimization algorithm configuration module uses a genetic algorithm and dynamically adjusts its parameters to achieve efficient and accurate optimization solutions. Finally, the instruction generation module generates a set of specific executable control instructions based on the optimization results, thereby realizing multi-objective collaborative optimization and precise closed-loop control of the combustion process under complex and variable operating conditions.

[0081] The optimization algorithm configuration module is also configured to adopt a dynamic constraint processing mechanism. Based on the real-time operating status of the burner and the prediction results of the flow field prediction unit, the constraints in the optimization process are adaptively adjusted. The constraints include, but are not limited to, the upper temperature limit for safe operation of the burner, the explosion-proof limit of ammonia injection, and the minimum swirl intensity to maintain flame stability. The dynamic constraint processing mechanism calculates the boundary margin of each constraint in real time and introduces an adaptive penalty function during the optimization process to ensure that the generated control commands are always within the feasible region for safe and stable operation of the burner.

[0082] The technical effects of the above-mentioned technical solution are as follows: The dynamic constraint processing mechanism can adaptively adjust the constraint conditions according to the real-time operating status of the burner and the flow field prediction results. By calculating the boundary margin of key constraint conditions such as the upper limit of temperature, the explosion limit of ammonia gas, and the minimum swirl intensity in real time, and introducing an adaptive penalty function in the optimization process, it effectively ensures that the multi-objective optimization solution process is always strictly limited to the feasible domain of safe and stable operation of the burner. Thus, while pursuing the improvement of combustion efficiency and the reduction of pollutant emissions, it avoids the equipment safety hazards and operational instability risks that may be caused by the optimization command going out of bounds.

[0083] The generated control commands are verified and validated, including:

[0084] Check whether the format of the control commands conforms to the system's preset standards, including the command syntax, data type, and unit, specifically:

[0085] Verify that the parameters in the command (such as damper opening, nozzle flow rate, etc.) are within a reasonable range.

[0086] Check whether the format of the command conforms to the requirements of the communication protocol, such as whether it has the necessary header information, check bits, etc.

[0087] Ensure that all parameters included in the command have clearly defined units to avoid errors caused by inconsistent units.

[0088] Verify whether the parameters in the control commands are within the safe operating range of the equipment, specifically:

[0089] Compare the parameters in the control instructions with the safe operating parameter range provided by the equipment manufacturer. For example, whether the opening of the air valve is between 0% and 100%, and whether the flow rate of the decomposed ammonia nozzle is between the maximum and minimum flow rates allowed by the equipment. If the parameters exceed the safe range, an alarm will be issued and the parameters will be automatically adjusted to the safe range.

[0090] Check for logical contradictions among the parameters in the control commands, specifically:

[0091] Verify whether the adjustments of primary and secondary air are coordinated. For example, if the primary air is increased, does the secondary air need to be adjusted accordingly to maintain stable combustion?

[0092] Check whether the adjustment of the ammonia decomposition gas nozzle flow rate matches the current operating status of the burner (such as temperature and flow rate);

[0093] If a logical contradiction is found, an alarm will be issued and the control instructions will be recalculated;

[0094] The control commands are compared with the real-time operating data provided by the multimodal sensing unit to verify the rationality and feasibility of the commands. Specifically:

[0095] Compare the parameters in the control command with the actual parameters read by the sensor to check for significant differences. For example, if the control command requires an increase in airflow, but the sensor shows that the current airflow is already close to the upper limit, then the control command needs to be re-evaluated. If a difference is found, an alarm is issued and the control command is adjusted.

[0096] The technical effect of the above solution is as follows: by performing multi-level rigorous verification and validation on the generated control commands, potential erroneous commands can be effectively identified and corrected, ensuring that each issued control command is safe, reasonable and executable, thus guaranteeing the reliability of the burner control process and the safety of system operation.

[0097] After the control command set is issued to the actuator unit, the system receives feedback information from the actuator unit, such as the status of command execution and changes in actual operating parameters. Based on the feedback information, the system evaluates the effect of the control command. If it is found that the expected optimization effect has not been achieved after the command is executed, the optimization strategy and control command are adjusted in a timely manner. Through the closed-loop feedback mechanism, the system ensures that the adaptive optimization unit can continuously optimize the combustion process and achieve the best balance between combustion efficiency, NOx emissions, ammonia slip and flame stability.

[0098] The technical effects of the above-mentioned technical solution are as follows: By establishing a closed-loop feedback mechanism, after the control command set is issued to the actuator unit, it can receive feedback information on the command execution status and actual operating parameter changes in real time, and accurately evaluate the actual effect of the control command accordingly. Once it is found that the execution result does not meet the expected optimization target, the optimization strategy can be adjusted in time and the control command can be regenerated, thereby forming a dynamic adjustment cycle of continuous optimization, ensuring that the combustion process always approaches the multi-objective balance direction of the highest combustion efficiency, the lowest NOx emission, the minimum ammonia escape, and the best flame stability.

[0099] Working Principle: The multi-modal sensing unit collects burner operating data in real time. The flow field prediction unit, based on an AI model jointly trained with historical data and fluid dynamics simulation, performs high-precision spatiotemporal joint predictions of the three-dimensional flow field, temperature field, and component field for a specific future time step. The adaptive optimization unit, based on this prediction data, utilizes a built-in multi-objective optimization function for ammonia-blended combustion, solving for optimal control commands by considering combustion efficiency, NOx emissions, ammonia slip, and flame stability. Finally, the actuator unit precisely adjusts the supply parameters of primary air, secondary air, and decomposed ammonia, achieving dynamic optimization of the combustion state. By constructing a closed-loop control system integrating real-time sensing, intelligent prediction, optimization decision-making, and precise execution, a shift from passive response to active intervention is achieved, significantly improving combustion efficiency, drastically reducing NOx emissions and ammonia slip, and ensuring high flame stability.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. An AI flow field prediction-based self-adaptive control system for a decomposed ammonia-doped swirl coal-pulverized burner, characterized in that, The application relates to a combustion state closed-loop control system based on AI prediction and adaptive optimization, comprising the following parts: a multi-modal sensing unit configured to collect real-time operation data of a burner, wherein the operation data at least includes primary air coal powder concentration and flow rate, secondary air flow rate and rotational flow intensity, decomposition ammonia injection position, flow rate and temperature, and flame shape image and spectrum information of a burner outlet area; a flow field prediction unit in communication connection with the multi-modal sensing unit, configured to jointly train a pre-constructed AI prediction model based on historical operation data and combustion fluid dynamics simulation data, receive operation data of the multi-modal sensing unit as input, and output prediction data of a burner outlet three-dimensional flow field structure, temperature field and component concentration field at a future specific time step; an adaptive optimization unit in communication connection with the flow field prediction unit, configured to have a multi-objective optimization function of ammonia combustion built-in, the optimization function taking the highest combustion efficiency, the lowest NOx emission, the smallest ammonia escape amount and the best flame stability as optimization objectives, receiving output of the flow field prediction unit, solving the optimization function, and generating a set of control instructions for primary air, secondary air and decomposition ammonia supply systems; an actuator unit in communication connection with the adaptive optimization unit, comprising independently adjustable primary air valves, secondary air rotational flow actuators and decomposition ammonia injection flow regulating valves arranged at multiple axial positions, configured to execute the control instruction set issued by the adaptive optimization unit to complete closed-loop control of the combustion state.

2. The AI flow field prediction-based decomposition ammonia-doped swirl coal burner self-adaptive regulation system according to claim 1, characterized in that, The flow field prediction unit comprises: a data collection module configured to receive burner operation data collected by the multi-modal sensing unit in real time, receive historical operation data and combustion fluid dynamics simulation data, and integrate the obtained multi-source heterogeneous data into a unified, standardized time series data set suitable for model training through time series alignment, data cleaning, outlier elimination and normalization processing; a feature extraction module configured to extract features from the operation data received by the data collection module, extract features related to the burner flow field prediction, including feature vectors of the flow rate and rotational flow intensity of the primary air and the secondary air, distribution characteristics of the coal powder concentration, injection parameter characteristics of the decomposition ammonia, and key features in the flame shape image and spectrum information; a model training module configured to construct an AI prediction model based on a deep learning algorithm, jointly train the AI prediction model based on the standardized time series data set integrated by the data fusion and processing module, optimize the performance of the AI prediction model by adjusting the hyperparameters and training strategies of the model, and evaluate the performance of the model by using a cross-validation method during the training process; a result output module configured to input the feature data extracted from the operation data into the trained AI prediction model, and output prediction data of the burner outlet three-dimensional flow field structure, temperature field and component concentration field at a future specific time step based on the input feature data.

3. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 2, characterized in that, The AI prediction model adopts an online incremental learning mechanism, compares the combustion state data actually collected by the multi-modal sensing unit subsequently with the model prediction data, calculates the prediction error, and automatically triggers the fine tuning update of the model parameters when the prediction error exceeds a preset threshold.

4. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 2, characterized in that, The AI prediction model is a hybrid architecture of a graph neural network and a time series convolution network, wherein the graph neural network is used to model the unstructured spatial topological relationship of the internal and outlet regions of the burner, capture the local flow field structure, and the time series convolution network is used to extract deep time series features in the real-time data collected by the multi-modal sensing unit. The fusion of the graph neural network and the time series convolution network is used to jointly realize the spatio-temporal prediction of the future three-dimensional flow field, temperature field and component concentration field.

5. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 2, characterized in that, The prediction results output by the AI prediction model are post-processed, including: Data decoding is used to convert the encoded data output by the AI prediction model into interpretable physical quantities; Data verification is used to verify the reasonableness of the prediction results, including verifying the physical reasonableness of the temperature field and the range of the component concentration; Data output is used to output the prediction results in a standardized format for use by the adaptive optimization unit.

6. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 1, characterized in that, The adaptive optimization unit includes: A data analysis module configured to receive and analyze the prediction data output by the flow field prediction unit, extract feature information related to the optimization target, including temperature distribution, component concentration and flow field structure characteristics, and simultaneously receive real-time operating data of the multi-modal sensing unit, including the current state of the primary air, secondary air and decomposed ammonia, for considering the current operating conditions during optimization; A target setting module configured to internally set an ammonia-doped combustion multi-objective optimization function, clearly define the optimization targets as the highest combustion efficiency, the lowest NOx emission, the smallest ammonia escape amount and the best flame stability, and set the weights of each optimization target according to actual operating requirements and priorities; An optimization algorithm configuration module configured to select a genetic algorithm to solve the multi-objective optimization function, and configure the parameters of the genetic algorithm, including population size, crossover probability and mutation probability, according to the characteristics and complexity of the optimization problem, and dynamically adjust the parameters of the genetic algorithm in combination with the output of the AI prediction model; An instruction generation module configured to generate a set of control instructions for the primary air, secondary air and decomposed ammonia supply system according to the solution of the optimization algorithm, including the opening of the primary air valve, the adjustment angle of the secondary air swirler and the specific parameters of the decomposed ammonia injection flow, and issue the generated control instruction set to the execution mechanism unit.

7. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 6, characterized in that, The optimization algorithm configuration module is also configured to adopt a dynamic constraint processing mechanism to adaptively adjust the constraint conditions in the optimization process according to the real-time operating state of the burner and the prediction results of the flow field prediction unit, wherein the constraint conditions include but are not limited to the upper limit of the safe operating temperature of the burner, the explosion-proof limit of the ammonia injection amount, and the minimum swirl intensity to maintain flame stability.

8. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 6, characterized in that, The generated control instructions are verified and checked, including: Check whether the format of the control instructions conforms to the system preset standards, including the syntax, data type and unit of the instructions; Verify whether the parameters in the control instructions are within the safe operating range of the equipment; Check if there is a logical contradiction between the parameters in the regulation instruction; Compare the regulation instruction with the real-time operation data provided by the multi-modal sensing unit to verify the rationality and feasibility of the instruction.

9. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 6, characterized in that, After the regulation instruction set is issued to the execution mechanism unit, the feedback information of the execution mechanism unit is received, including the state of instruction execution and the change of actual operation parameters. According to the feedback information, the effect of the regulation instruction is evaluated. If it is found that the instruction execution cannot achieve the expected optimization effect, the optimization strategy and regulation instruction are adjusted in time.

10. The AI flow field prediction based decomposition ammonia-doped swirl pulverized coal burner self-adaptive regulation and control system according to claim 1, characterized in that, It also includes a safety protection unit, which is in communication connection with the multi-modal sensing unit and the adaptive optimization unit, and is configured to: Real-time monitoring of safety parameters, including but not limited to combustion chamber pressure, temperature extreme value and ammonia leakage concentration; Based on the real-time monitoring of safety parameters, if a dangerous working condition is detected, override the regulation instruction of the adaptive optimization unit and send a safety protection instruction to the execution mechanism unit; At the same time, multiple safety redundancy mechanisms are provided, including hardware safety loop and software safety logic, to ensure that the burner can be safely shut down or switched to a safe operation mode in case of system failure.

Images