Intelligent ammonia injection control method and system suitable for kiln flue gas denitration big data prediction model

Abstract

The application discloses a kind of big data prediction model intelligent ammonia injection control method and system suitable for kiln flue gas denitration.It first collects the historical and real-time data of glass kiln fuel gas flow and other parameters, uses LSTM-GRU hybrid neural network modeling to predict the amount of NOx, calculates the amount of ammonia injection through adaptive control, and corrects the accuracy through two-stage correction. The calculated amount of ammonia injection is transmitted to the production main control system to adjust the ammonia supply valve. In case of failure, it automatically alarms and switches to manual control. The system covers data acquisition, intelligent prediction, adaptive control and other modules, and also has a kinetic catalytic reaction simulation test platform for optimization. The application realizes intelligent automatic control of big data model, accurately adjusts the amount of ammonia injection, improves denitration efficiency and ammonia utilization rate, reduces operating costs, and enhances system reliability and stability.

Description

Technical Field

[0001] This invention relates to the field of industrial pollution control, specifically to a big data prediction model-based intelligent ammonia injection control method and system applicable to kiln flue gas denitrification. Background Technology

[0002] The vast majority of nitrogen oxides in flue gas originate from thermal NOx. When the combustion temperature is below 1300℃, very little thermal NOx is generated; however, when the temperature exceeds 1300℃, the amount of NOx generated increases dramatically with increasing temperature. In glass melting furnaces, the fuel gas is premixed and heated in a regenerator before entering the furnace for combustion. Due to the melting process in glass furnaces, the temperature of the flue gas generated exceeds 1500℃, producing a large amount of NOx at this high temperature, with the NOx concentration in the glass furnace flue gas exceeding 2000 mg / Nm³. The operational characteristics of glass furnaces also lead to periodic fluctuations in flue gas NOx concentration. Under normal circumstances, to prevent overheating, the furnace typically requires flame reversal every 20-30 minutes, with a reversal interval of approximately half a minute. During the reversal, combustion needs to be stopped briefly, resulting in a rapid decrease in flue gas NOx concentration and flow rate. After the reversal, in order to maintain the temperature inside the kiln, the amount of fuel gas supplied was increased for about two minutes, resulting in continuous fluctuations in NOx in the flue gas. The NOx concentration trend in the glass kiln flue gas showed a periodic pattern, such as... Figure 1 Currently, glass furnace flue gas generally adopts medium- and high-temperature SCR denitrification technology. Due to the large fluctuation range of NOx value in the flue gas, it is impossible to automatically control the amount of ammonia added through PID proportional adjustment. Operators can only manually control the amount of ammonia added based on the hourly average value of NOx at the discharge port, which often leads to an increase in ammonia escape rate. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent ammonia injection control method based on a big data prediction model for denitrification of kiln flue gas, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a smart ammonia injection control method based on a big data prediction model for kiln flue gas denitrification, comprising the following steps:

[0005] Historical and real-time data on the reversing operation time, fuel gas flow rate, combustion air volume, flue gas flow rate, and NOx concentration of the glass furnace are collected. Data is acquired from the main control system via the Modbus TCP protocol and stored in the precision ammonia injection control system database.

[0006] The prediction model is trained using an LSTMGRU hybrid neural network to model the historical data of collected flue gas flow and NOx concentration. The prediction module uses a sliding window prediction to consider the impact of glass batching and melting process cycle on NOx emissions and predict the amount of NOx generated by the glass furnace in the future. The prediction frequency can be adjusted every 15 to 60 seconds.

[0007] An adaptive control system is adopted, which uses the predicted NOx emission value of the LSTMGRU hybrid neural network model as a benchmark, superimposes the adaptive control calculation, automatically adjusts the control parameters according to the process stage, calculates the corresponding ammonia injection amount, and corrects the lag time caused by slow instrument response. The response cycle of the adaptive control module does not exceed 15 seconds.

[0008] Based on the calculated ammonia injection rate, the system monitors the actual NOx emission value in real time and compares it with the predicted value. If there is a significant deviation, the prediction model is corrected once within the current cycle.

[0009] The NOx value of the preset discharge outlet is compared with the CEMS test data of the discharge outlet, and the average difference within a set period of time (not less than ten minutes) is calculated. The value is then corrected at regular intervals as a secondary correction to ensure the environmental protection requirements of the hourly average value of the discharge outlet are met.

[0010] The calculated ammonia injection volume data is transmitted to the production main control system by the intelligent precision ammonia injection control system data transmission module. The production main control system then adjusts the ammonia supply valves. If the precision ammonia injection system malfunctions, an alarm is immediately triggered, and the system automatically switches to manual control mode.

[0011] Furthermore, the collected historical data on flue gas flow and NOx concentration from the glass furnace are preprocessed, including data cleaning and normalization, to improve the training effect of the LSTM-GRU hybrid neural network training prediction model.

[0012] Furthermore, the first correction specifically involves adjusting some parameters of the prediction model according to a preset correction formula when the deviation between the actual NOx emission value and the predicted value exceeds a preset first threshold. The correction formula is related to the magnitude of the deviation and the current process stage.

[0013] Furthermore, during the secondary correction, if the calculated average difference exceeds a preset second threshold, the correction coefficient for the ammonia injection amount is adjusted proportionally according to the sign and magnitude of the difference, thereby adjusting the total ammonia supply.

[0014] A big data predictive model-based intelligent ammonia injection control system suitable for kiln flue gas denitrification includes:

[0015] Data acquisition module: It interacts with the glass kiln control system to collect multi-dimensional data such as NOx concentration, flue gas temperature, flue gas flow rate, kiln pressure, and ammonia escape in the glass kiln flue gas in real time at a sampling frequency of once every 2 seconds. It uses an anomaly detection algorithm to remove outliers in the data and then stores them in the time series database of the intelligent precision ammonia injection control system.

[0016] Intelligent prediction module: The model is built using LSTMGRU hybrid neural network. The model evaluation index is RMSE root mean square error or MAPE mean absolute percentage error. It predicts the NOx value of the inlet flue gas after denitrification. The model supports online learning and is automatically updated every day.

[0017] Adaptive control module: The predicted NOx emission value of the LSTMGRU hybrid neural network model is used as a benchmark, and adaptive control calculation is superimposed to automatically adjust the control parameters according to the process stage to calculate the corresponding ammonia injection amount.

[0018] Calculation and correction module: Based on the calculated ammonia injection rate, the system monitors the actual NOx value at the denitrification inlet and the NOx data detected by the CEMS at the outlet in real time, and makes primary and secondary corrections to the ammonia supply.

[0019] Data transmission module: Transmits the final ammonia injection rate signal or fault signal to the denitrification main control system, which then directly controls the ammonia injection regulating valve to adjust the ammonia injection rate.

[0020] Fault diagnosis module: Real-time monitoring of system status; Based on anomaly detection algorithms, it identifies system fault modes such as sensor faults, prediction module faults, adaptive module faults, and network communication faults; When a fault occurs, it automatically switches to the backup control scheme.

[0021] Kinetic catalytic reaction simulation test platform: Based on the actual catalyst performance of the target denitrification unit, the platform simulates the actual production conditions by combining experimental data from catalyst manufacturers and operating data from glass furnaces. It uses a neural network intelligent algorithm to fit a multi-dimensional continuous reaction model platform, which simulates the actual production conditions and provides simulation testing for the big data model intelligent and precise ammonia injection control system, thereby optimizing system parameter settings.

[0022] Furthermore, the data acquisition module also has a data encryption function, which encrypts the collected multi-dimensional data on NOx concentration and flue gas temperature in the glass furnace during storage and transmission to ensure data security.

[0023] Furthermore, the intelligent prediction module dynamically adjusts the learning rate during the prediction process. When the loss value of the model training decreases less than a preset value in multiple consecutive training batches, the learning rate is reduced to improve the model convergence speed and prediction accuracy.

[0024] Furthermore, the adaptive control module sets up multiple control parameter groups corresponding to different process stages. Based on the real-time acquired process stage identifier, it quickly matches and calls the corresponding control parameter group to calculate the ammonia injection amount.

[0025] Furthermore, the kinetic catalytic reaction simulation test platform can also simulate sudden situations under different operating conditions, such as large fluctuations in fuel gas flow and catalyst poisoning in a short period of time, to test the system's response capability and control effect under extreme conditions.

[0026] Furthermore, the fault diagnosis module adopts a fusion of multiple anomaly detection algorithms to comprehensively determine whether the system has malfunctioned. Only when different detection algorithms detect anomalies is the system determined to have malfunctioned, thereby improving the accuracy of fault diagnosis and reducing false alarms.

[0027] Compared with the prior art, the beneficial effects of the present invention are:

[0028] This invention utilizes an LSTM-GRU hybrid neural network prediction algorithm to deeply mine patterns in historical data and accurately predict the changing trend of NOx levels in the flue gas from the denitrification inlet. Based on the accurate prediction results, the system can precisely adjust the ammonia injection rate, achieving a better match between the ammonia injection rate and NOx generation. Compared to traditional manual adjustment methods, this effectively avoids the problems of low denitrification efficiency and ammonia escape caused by untimely or inaccurate adjustments, significantly improving denitrification efficiency and ammonia utilization, reducing the amount of denitrification agent used, and thus saving denitrification costs.

[0029] The adaptive control module can automatically adjust control parameters according to different stages of the production process, quickly responding to changes in the production process. Furthermore, the system is equipped with a two-level correction mechanism, monitoring actual NOx emissions in real time and comparing them with predicted values. Once a deviation is detected, the prediction model and ammonia injection rate can be corrected promptly, greatly improving the system's responsiveness to changes in actual operating conditions. This ensures the accuracy and stability of intelligent and precise ammonia injection, guaranteeing efficient and stable operation of the denitrification system regardless of the glass kiln's operating status.

[0030] The fault diagnosis module monitors the system status in real time. Based on advanced anomaly detection algorithms, it can accurately identify various fault modes, including sensor faults, predictive module faults, adaptive module faults, and network communication faults. Once a fault is detected, the system immediately triggers an alarm and automatically switches to manual control mode or a backup control scheme. This ensures that the denitrification system can maintain basic operation even in the event of a fault, avoiding denitrification interruptions and environmental pollution caused by system failures, and improving the system's reliability and safety.

[0031] By establishing a kinetic catalytic reaction simulation testing platform, and combining the actual catalyst performance of the target denitrification unit, experimental data from catalyst manufacturers, and operating data of the glass furnace, actual production conditions are simulated. This provides a comprehensive simulation testing environment for the intelligent and precise ammonia injection control system based on the big data model. It enables the optimization of system parameters before the system is put into actual production, allowing for the early detection and resolution of potential problems, and effectively improving the overall operational stability and performance of the system.

[0032] Big data AI intelligent models possess self-learning and self-driven optimization capabilities. As production data accumulates, the model can automatically optimize its own parameters, continuously improving the accuracy of prediction and control. This significantly increases the automation level of precision ammonia injection systems, reduces reliance on manual intervention, lowers labor costs, and enhances the intelligence level of the production process, providing strong support for the intelligent upgrading of industrial production. Attached Figure Description

[0033] Figure 1 A trend chart of NOx concentration in glass furnace flue gas;

[0034] Figure 2 This is a schematic diagram of the system flow of the present invention. Detailed Implementation

[0035] 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.

[0036] Example 1:

[0037] Please see Figure 1 —2. This invention provides a big data prediction model-based intelligent ammonia injection control method and system suitable for kiln flue gas denitrification. Its purpose is to achieve precise control of total ammonia regulation in the denitrification system and improve the performance of the denitrification ammonia injection system in terms of timeliness, accuracy, reliability and stability.

[0038] In a real industrial setting, a representative large-scale regenerative glass furnace is used as the implementation object. The glass furnace melts up to

[100] tons of glass per day and uses high-calorific-value natural gas as the main fuel. During the combustion process, the temperature inside the furnace reaches 1550℃-1650℃, which results in the thermal NOx concentration in the flue gas remaining at a high level for a long time, usually fluctuating drastically between 2500mg / Nm³-3800mg / Nm³, posing a great challenge to the denitrification work.

[0039] The data acquisition module is the foundation of the entire system, and its stable and accurate data acquisition capabilities are crucial. This module establishes a reliable data transmission link with the main control system (PLC or DCS) of the glass furnace via the Modbus TCP protocol. Every 2 seconds, the system automatically collects various key data during the operation of the glass furnace, including fuel gas flow rate, combustion air volume, flue gas flow rate, NOx concentration, flue gas temperature, furnace pressure, and ammonia slip, among other multi-dimensional information.

[0040] Given the complexity of industrial production environments, data anomalies are frequent. To ensure the quality of collected data, an anomaly detection algorithm combining statistical analysis and machine learning is employed. On one hand, the 3σ rule is used for initial data screening, marking data points that deviate from the mean by more than three standard deviations. On the other hand, the Isolation Forest algorithm is introduced to identify data points that deviate from normal patterns within the data distribution structure. For example, if a maximum value appears in the collected flue gas flow data at a certain moment, the 3σ rule initially identifies it as an outlier, and the Isolation Forest algorithm further confirms that this data point is isolated in the data space, thus classifying it as an anomaly and removing it.

[0041] After outlier handling, a sliding window averaging method is used to optimize the data to make it smoother and more stable. Taking NOx concentration data as an example, the sliding window size is set to a fixed value. This means that the NOx concentration data at consecutive fixed time points is taken and the arithmetic mean is calculated, with the resulting average replacing the intermediate data points within the window. This effectively eliminates short-term fluctuations in the data, allowing it to better reflect the true trend of NOx concentration changes. Simultaneously, since data from different data sources may have time synchronization issues during transmission, a high-precision time synchronization device is used, with the main control system's clock as the reference, to precisely calibrate the data time of each data source, ensuring consistency in time across all data. This provides an accurate and reliable data foundation for subsequent data analysis and model training. The collected data is systematically stored in the time-series database of the intelligent precision ammonia injection control system for easy retrieval and in-depth analysis.

[0042] The intelligent prediction module, as a core component of the system, undertakes the crucial task of predicting future NOx emissions. This module employs an LSTM-GRU hybrid neural network for modeling. LSTM (Long Short-Term Memory) networks, with their unique gating mechanism, can effectively handle long-term dependencies in long-sequence data, offering significant advantages in capturing long-term trends in NOx emissions during glass furnace production. In this embodiment, the LSTM is configured with two layers, each containing 128 nodes, which can fully learn the complex features in historical data. The GRU (Gated Recurrent Unit), while maintaining model performance, simplifies the network structure and improves computational efficiency; it is also configured with two layers, each containing 64 nodes. The fully connected layers are configured with three layers, with 128, 64, and 1 node respectively. Their function is to integrate the features extracted by the LSTM and GRU and output the final prediction result.

[0043] Considering that the periodic changes in glass batching, melting, and other process stages have a significant impact on NOx emissions, the prediction frequency can be flexibly adjusted between 15 and 60 seconds based on actual production conditions. During the normal production phase of the glass furnace, the process is relatively stable, and NOx emission changes are relatively gradual. In this case, setting the prediction frequency to 30 seconds ensures prediction accuracy while reasonably balancing computational resource consumption. However, during special process stages such as fire changeovers in the glass furnace, NOx emissions fluctuate dramatically. To promptly capture these changes, the prediction time is shortened to 15 seconds to provide more real-time and accurate predictive data for ammonia injection control.

[0044] To evaluate the model's predictive performance, RMSE (Root Mean Square Error) and MAPE (Mean Absolute Percentage Error) were used as evaluation metrics. RMSE directly reflects the average error between predicted and actual values, while MAPE measures the accuracy of predictions from the perspective of relative error. During model training, hyperparameters such as learning rate and number of iterations were continuously adjusted to minimize RMSE and MAPE values, thereby improving the model's prediction accuracy. Furthermore, the model supports online learning. During off-peak hours after production ends each day, the system automatically collects new data generated during the day's production and incorporates it into the training dataset for retraining. In this way, the model can automatically learn new patterns and trends as production data accumulates, continuously optimizing its parameters to better adapt to the dynamic changes in the glass kiln production process.

[0045] The adaptive control module, based on the predicted NOx emission value using an LSTM-GRU hybrid neural network model and combined with an adaptive control algorithm, achieves accurate calculation and real-time adjustment of ammonia injection rate. This module can automatically adjust control parameters according to different process stages of the glass furnace. During the heating stage of the glass furnace, the fuel gas flow rate and combustion air volume increase significantly, leading to a significant increase in NOx generation. At this time, the adaptive control module increases the calculation coefficient of the ammonia injection rate accordingly, ensuring sufficient ammonia to react with NOx and achieve efficient denitrification. During the heat preservation stage of the glass furnace, NOx generation is relatively stable and low; the adaptive control module reduces the calculation coefficient of the ammonia injection rate to avoid excessive ammonia injection, thereby reducing ammonia escape.

[0046] The response speed of the adaptive control module is crucial, with its response cycle strictly controlled to no more than 15 seconds. Taking the flame reversal operation of a glass furnace as an example, when a reversal occurs, the system can quickly detect the process change within 15 seconds, readjust the control parameters, and calculate the appropriate ammonia injection rate to ensure the continuity and stability of the denitrification process. To compensate for the lag time caused by slow instrument response, a time compensation algorithm is introduced into the adaptive control module. By accurately measuring and analyzing the instrument response time and combining it with actual data from the production process, a time compensation model is established. When calculating the ammonia injection rate, the output of the ammonia injection rate is adjusted in advance based on the instrument lag time, so that ammonia can be injected into the flue gas at the optimal time point to fully react with NOx, thereby improving denitrification efficiency.

[0047] Example 2:

[0048] The difference from Implementation 1 lies in the fact that the calculation correction module further improves the accuracy of ammonia injection through a two-level correction mechanism, ensuring that the denitrification effect always meets environmental protection requirements. During a correction process, the system continuously monitors the actual NOx value at the denitrification inlet in real time and dynamically compares it with the predicted value. When the deviation between the actual NOx emission value and the predicted value exceeds a preset first threshold, a correction procedure is immediately triggered. Assuming the preset first threshold is set to ±60 mg / Nm³, if the actual NOx value at the denitrification inlet is more than 60 mg / Nm³ higher than the predicted value at a certain moment, the system determines that there is a significant deviation. At this time, the prediction model is adjusted according to a preset correction formula. The correction formula comprehensively considers factors such as the magnitude of the deviation, the current process stage, and historical data from system operation. For example, the correction coefficient = deviation value / (predicted value × process stage correction factor), where the process stage correction factor is preset according to the degree of influence of different process stages on NOx emissions. By adjusting the correction coefficient, some parameters of the prediction model are optimized, such as adjusting the weights and biases in the LSTM-GRU hybrid neural network model, enabling the model to more accurately fit actual production data and improve prediction accuracy. After completing one correction to the prediction model, the system recalculates the ammonia injection amount based on the corrected model to ensure that the ammonia injection amount can more accurately match the actual NOx emissions.

[0049] The secondary correction is primarily to ensure that the hourly average NOx emission level at the outlet strictly complies with environmental standards. A target NOx value for the outlet is preset, determined based on local environmental regulations and the actual production conditions of the glass kiln, for example, set at 90 mg / Nm³. The system periodically compares the CEMS monitoring data from the outlet with the preset target value and calculates the average difference over that period. If the average difference exceeds a preset second threshold (e.g., ±15 mg / Nm³), the secondary correction process is initiated. The correction coefficient for ammonia injection is adjusted proportionally based on the sign and magnitude of the difference. If the average difference is positive, meaning the actual average NOx emission level is higher than the target value, it indicates insufficient ammonia injection, and the system appropriately increases the correction coefficient for ammonia injection; conversely, if the average difference is negative, the correction coefficient for ammonia injection is correspondingly decreased. In this way, the total ammonia supply is corrected twice to ensure that the hourly average NOx emission level remains consistently within the environmental requirements.

[0050] The data transmission module is responsible for accurately transmitting the final ammonia injection rate signal and fault signal to the denitrification main control system. During data transmission, multiple data checksums and redundant transmission technology are employed to ensure data integrity, accuracy, and reliability. Each time data is sent, the system simultaneously calculates multiple checksums, such as CRC checksums and parity checksums, and sends these checksums along with the data. Upon receiving the data, the denitrification main control system calculates and compares the different checksums. If all checksums match, the data transmission is considered correct; if any checksum does not match, the system determines that an error has occurred during transmission and immediately sends a retransmission request to the intelligent precision ammonia injection control system. Upon receiving the retransmission request, the intelligent precision ammonia injection control system retransmits the data until the data transmission is successful. Furthermore, to improve data transmission stability, redundant transmission links are used, i.e., data transmission is performed simultaneously through multiple network links. When one link fails, the system automatically switches to another normal link to ensure the continuity of data transmission.

[0051] The fault diagnosis module monitors the system's operational status in real time and accurately identifies fault modes through the fusion of multiple anomaly detection algorithms. It employs various techniques, including the 3σ rule based on statistical analysis, the Isolation Forest algorithm based on machine learning, and the autoencoder anomaly detection algorithm based on deep learning. A system fault is only determined when all different detection algorithms detect anomalies, thus effectively improving the accuracy of fault diagnosis and reducing false alarms. For example, when detecting sensor faults, if the 3σ rule detects that a sensor's data repeatedly exceeds the normal range, and both the Isolation Forest algorithm and the autoencoder anomaly detection algorithm identify this data point as an anomaly, then the sensor is considered potentially faulty. Once a fault is detected, the system immediately triggers a comprehensive alarm mechanism, including audible and visual alarms, SMS notifications, and email alerts, promptly informing operators. Simultaneously, the system automatically switches to a backup control scheme, such as switching to manual control mode or activating a pre-set emergency control strategy. In manual control mode, operators can manually adjust the ammonia injection rate based on real-time data and detailed alarm information provided by the system to ensure that the denitrification system maintains basic denitrification functions during a fault, preventing NOx emissions exceeding standards due to system failure.

[0052] Before the intelligent and precise ammonia injection system based on the big data model is put into actual production, it needs to undergo comprehensive and in-depth simulation testing through a kinetic catalytic reaction simulation testing platform. Accurate simulations are performed based on the actual catalyst performance parameters of the target denitrification unit, combined with experimental data provided by the catalyst manufacturer and extensive historical operating data from the glass furnace. A multi-dimensional continuous reaction model platform is constructed using neural network intelligent algorithms. During the simulation, not only normal production conditions are simulated, but also various unexpected situations are covered, such as rapid fluctuations in fuel gas flow (±40%), catalyst poisoning (simulating a 60% decrease in catalyst activity), and sudden power outages. Through simulation tests of these extreme conditions, the system's response capability and control effect under various complex conditions are comprehensively verified. Based on the test results, optimization algorithms are used to finely adjust the parameters of the intelligent and precise ammonia injection control system based on the big data model. This includes optimizing the structure and parameters of the prediction model, improving the control strategy of the adaptive control module, and refining the correction logic of the calculation correction module, ensuring that the system can cope with various complex and changing situations in actual operation and achieve stable and efficient denitrification control.

[0053] Example 3:

[0054] With the expansion of glass furnace production scale and the increase in process complexity, the lag in spectral analysis data acquisition speed can no longer meet the requirements of precise control. Therefore, based on existing data acquisition methods, rapid detection methods are added. In-situ analytical instruments with shorter response times are introduced to monitor changes in NOx, O2, and NH3 in flue gas in real time, obtaining more accurate NOx and NH3 concentration information, which is beneficial for establishing more accurate prediction models and for more precise correction and optimization.

[0055] In terms of data processing, multi-source data fusion technology is employed. Data from different sensors are correlated and analyzed to construct a data fusion model. For example, by combining fuel gas flow rate, combustion air volume, flue gas temperature, and data from in-situ analytical instruments, principal component analysis (PCA) is used to extract key features, reducing data dimensionality and improving data processing efficiency. Simultaneously, an autoencoder from deep learning is used to learn features from the fused data, uncovering potential complex relationships between data points and providing more valuable information for subsequent prediction and control.

[0056] In the intelligent prediction module, we continuously explore more advanced model architectures and optimization algorithms. An attention mechanism is introduced, enabling the model to pay closer attention to data features closely related to NOx emissions, thus improving prediction accuracy. For example, when processing data from different process stages in a glass furnace, the attention mechanism can automatically adjust its focus on features such as fuel gas flow rate and combustion air volume, thereby more accurately capturing the changing trends of NOx emissions.

[0057] The adaptive control module employs more refined control strategies. Based on factors such as different production loads of the glass furnace, product types, and ambient temperatures, the process is subdivided, and personalized control parameters are developed for each sub-stage. For example, when producing glass products of varying thicknesses, the furnace's heat requirements and combustion processes differ, leading to adjustments in the ammonia injection calculation coefficients and control parameters to achieve more precise ammonia injection control.

[0058] Simultaneously, the intelligent ammonia injection control system is optimized in synergy with other production stages of the glass furnace. Data exchange with the furnace temperature control system allows for advance adjustment of ammonia injection based on furnace temperature trends, preventing delays in denitrification caused by NOx generation fluctuations due to temperature variations. A linkage mechanism is established with the glass batching system to adjust the ammonia injection strategy in real-time based on changes in raw material composition, ensuring efficient denitrification under different raw material conditions.

[0059] The fault diagnosis module further incorporates fault prediction technology. By monitoring and analyzing system operation data in real time, it utilizes Long Short-Term Memory (LSTM) networks or Gated Recurrent Units (GRUs) from machine learning to predict the performance of key components such as sensors and actuators, thus identifying potential faults in advance. For example, by monitoring sensor signal fluctuations and drift trends, it can predict whether a sensor is about to fail, allowing for timely maintenance and replacement, and preventing control errors caused by sensor failures.

[0060] In terms of fault-tolerant control, in addition to switching to manual control mode, more intelligent fault-tolerant control algorithms are developed, and redundant designs are added. When a control module fails, the system automatically adjusts the control logic, utilizing the redundant functions of other normal modules to achieve partial control functions. For example, when part of the adaptive control module fails, basic ammonia injection control is maintained through preset backup control rules and calculation methods, combined with current production data, ensuring that the denitrification system can still maintain a certain level of operating efficiency during the failure period, reducing the impact on production.

[0061] To facilitate operators' intuitive understanding and monitoring of the system's operational status, a powerful visualization interface was developed. Real-time data visualization technology is used to display operating parameters of the glass furnace, NOx emission concentrations, ammonia injection rates, and system fault alarm information in the form of charts and graphs. Geographic Information System (GIS) technology is utilized to centrally display the denitrification systems of multiple glass furnaces, achieving unified monitoring of distributed production equipment.

[0062] Simultaneously, a remote monitoring and diagnostic platform has been established, supporting access from mobile terminals such as smartphones and tablets. Operators can view the system's operating status, receive alarm information, and perform remote operation and control anytime, anywhere via mobile devices. When a system malfunctions, remote experts can obtain system data in real time through the platform, conduct remote diagnosis, and provide guidance for repairs, improving the efficiency and timeliness of fault handling.

[0063] Regarding data privacy protection, the collected glass kiln production data and denitrification system operation data are stored and transmitted in encrypted form. Advanced Encryption Standard (AES) and other encryption algorithms are used to encrypt the data, ensuring its security during storage and transmission. Simultaneously, a data access audit mechanism is established to track and analyze data access records in detail, preventing data leakage and misuse.

Claims

1. A smart ammonia injection control method based on big data prediction model for kiln flue gas denitrification, characterized in that, Includes the following steps: Historical and real-time data on the reversing operation time, fuel gas flow rate, combustion air volume, flue gas flow rate, and NOx concentration of the glass furnace are collected. Data is acquired from the main control system via the Modbus TCP protocol and stored in the precision ammonia injection control system database. The prediction model is trained using an LSTMGRU hybrid neural network to model the historical data of collected flue gas flow and NOx concentration. The prediction module uses a sliding window prediction to consider the impact of glass batching and melting process cycle on NOx emissions and predict the amount of NOx generated by the glass furnace in the future. The prediction frequency can be adjusted every 15 to 60 seconds. An adaptive control system is adopted, which uses the predicted NOx emission value of the LSTMGRU hybrid neural network model as a benchmark, superimposes the adaptive control calculation, automatically adjusts the control parameters according to the process stage, calculates the corresponding ammonia injection amount, and corrects the lag time caused by slow instrument response. The response cycle of the adaptive control module does not exceed 15 seconds. Based on the calculated ammonia injection rate, the system monitors the actual NOx emission value in real time and compares it with the predicted value. If there is a significant deviation, the prediction model is corrected once within the current cycle. The NOx value of the preset discharge outlet is compared with the CEMS test data of the discharge outlet, and the average difference within a set period of time (not less than ten minutes) is calculated. The value is then corrected at regular intervals as a secondary correction to ensure the environmental protection requirements of the hourly average value of the discharge outlet are met. The calculated ammonia injection volume data is transmitted to the production main control system by the intelligent precision ammonia injection control system data transmission module. The production main control system then adjusts the ammonia supply valves. If the precision ammonia injection system malfunctions, an alarm is immediately triggered, and the system automatically switches to manual control mode.

2. The intelligent ammonia injection control method based on a big data prediction model for kiln flue gas denitrification as described in claim 1, characterized in that, The collected historical data on flue gas flow and NOx concentration from glass kilns are preprocessed, including data cleaning and normalization, to improve the training effect of the LSTM-GRU hybrid neural network training prediction model.

3. The intelligent ammonia injection control method based on a big data prediction model for kiln flue gas denitrification as described in claim 1, characterized in that, The first correction specifically involves adjusting some parameters of the prediction model according to a preset correction formula when the deviation between the actual NOx emission value and the predicted value exceeds a preset first threshold. The correction formula is related to the magnitude of the deviation and the current process stage.

4. The intelligent ammonia injection control method based on a big data prediction model for kiln flue gas denitrification as described in claim 1, characterized in that, During the secondary correction, if the calculated average difference exceeds the preset second threshold, the correction coefficient of the ammonia injection amount is adjusted proportionally according to the sign and magnitude of the difference, thereby adjusting the total ammonia supply.

5. A big data prediction model-based intelligent ammonia injection control system for kiln flue gas denitrification, employing the method described in claim 1, characterized in that... include: Data acquisition module: It interacts with the glass kiln control system to collect multi-dimensional data such as NOx concentration, flue gas temperature, flue gas flow rate, kiln pressure, and ammonia escape in the glass kiln flue gas in real time at a sampling frequency of once every 2 seconds. It uses an anomaly detection algorithm to remove outliers in the data and then stores them in the time series database of the intelligent precision ammonia injection control system. Intelligent prediction module: The model is built using LSTMGRU hybrid neural network. The model evaluation index is RMSE root mean square error or MAPE mean absolute percentage error. It predicts the NOx value of the inlet flue gas after denitrification. The model supports online learning and is automatically updated every day. Adaptive control module: The predicted NOx emission value of the LSTMGRU hybrid neural network model is used as a benchmark, and adaptive control calculation is superimposed to automatically adjust the control parameters according to the process stage to calculate the corresponding ammonia injection amount. Calculation and correction module: Based on the calculated ammonia injection rate, the system monitors the actual NOx value at the denitrification inlet and the NOx data detected by the CEMS at the outlet in real time, and makes primary and secondary corrections to the ammonia supply. Data transmission module: Transmits the final ammonia injection rate signal or fault signal to the denitrification main control system, which then directly controls the ammonia injection regulating valve to adjust the ammonia injection rate. Fault diagnosis module: Real-time monitoring of system status; Based on anomaly detection algorithms, it identifies system fault modes such as sensor faults, prediction module faults, adaptive module faults, and network communication faults; When a fault occurs, it automatically switches to the backup control scheme. Kinetic catalytic reaction simulation test platform: Based on the actual catalyst performance of the target denitrification unit, the platform simulates the actual production conditions by combining experimental data from catalyst manufacturers and operating data from glass furnaces. It uses a neural network intelligent algorithm to fit a multi-dimensional continuous reaction model platform, which simulates the actual production conditions and provides simulation testing for the big data model intelligent and precise ammonia injection control system, thereby optimizing system parameter settings.

6. The intelligent ammonia injection control system based on a big data prediction model for kiln flue gas denitrification as described in claim 5, characterized in that, The data acquisition module also has a data encryption function, which encrypts the collected multi-dimensional data on NOx concentration and flue gas temperature in the glass furnace flue gas during storage and transmission to ensure data security.

7. The intelligent ammonia injection control system based on a big data prediction model for kiln flue gas denitrification as described in claim 5, characterized in that, During the prediction process, the intelligent prediction module dynamically adjusts the learning rate. When the loss value of the model training decreases less than a preset value in multiple consecutive training batches, the learning rate is reduced to improve the model convergence speed and prediction accuracy.

8. The intelligent ammonia injection control system based on a big data prediction model for kiln flue gas denitrification as described in claim 5, characterized in that, The adaptive control module sets up multiple control parameter groups corresponding to different process stages. Based on the real-time acquired process stage identifier, it quickly matches and calls the corresponding control parameter group to calculate the ammonia injection amount.

9. The intelligent ammonia injection control system based on a big data prediction model for kiln flue gas denitrification as described in claim 5, characterized in that, The kinetic catalytic reaction simulation test platform can also simulate sudden situations under different operating conditions, such as large fluctuations in fuel gas flow and catalyst poisoning in a short period of time, to test the system's response capability and control effect under extreme conditions.

10. The intelligent ammonia injection control system based on a big data prediction model for kiln flue gas denitrification as described in claim 5, characterized in that, The fault diagnosis module employs a fusion of multiple anomaly detection algorithms to comprehensively determine whether a system fault has occurred. Only when all different detection algorithms detect anomalies is a fault determined to have occurred, thereby improving the accuracy of fault diagnosis and reducing false alarms.

Images