Flue gas denitration self-adaptive ammonia spraying device

Abstract

The invention discloses a self-adaptive ammonia spraying device for flue gas denitrification, relates to the technical field of flue gas denitrification environmental protection, and is used for improving the denitrification efficiency. In the device, a cooperative control framework of static hardware matching and dynamic software adjustment is constructed by the flue gas denitration self-adaptive ammonia spraying device, so that complex online mechanical adjustment is converted into external fluid control, and the problems of jamming and abrasion in a high-temperature and high-dust environment are solved by utilizing a fixed gap unit without a movable component; therefore, the long-term operation reliability of the flue gas denitration self-adaptive ammonia spraying device is improved. And meanwhile, by combining the fusion calculation of the AI model on the multi-source data and a double-level intelligent diagnosis early warning mechanism, the dual self-adaption of the ammonia spraying strategy on the space non-uniformity and the time change is realized, so that the denitration efficiency is improved on the premise of improving the stability of the flue gas denitration self-adaptive ammonia spraying device.

Description

Technical Field

[0001] This application relates to the field of flue gas denitrification environmental protection technology, and in particular to an adaptive ammonia injection device for flue gas denitrification. Background Technology

[0002] In industrial flue gas denitrification, selective catalytic reduction (SCR) technology is widely used due to its high denitrification efficiency and good stability. The core process of this technology involves injecting ammonia gas, a reducing agent, into the flue gas through an ammonia injection grid installed in the reactor inlet flue. After mixing with the flue gas, the ammonia reacts with nitrogen oxides (NOx) in the flue gas under the action of a catalyst. x A reduction reaction occurs, producing harmless nitrogen gas and water.

[0003] To address the uneven distribution of nitrogen oxide concentrations within the flue gas cross-section, related technologies typically employ continuously adjustable ammonia injection grid devices. This technology incorporates mechanical adjustment mechanisms (such as nested movable inner and outer sleeves or linked micro-adjusting valves) at each sub-unit or nozzle of the ammonia injection grid, operating online in conjunction with an external drive module and control system. During operation, the actuator drives the internal mechanical adjustment mechanism of the grid to generate continuous displacement or rotation. The relative sliding between moving parts alters the effective physical opening area of ​​each nozzle, thereby distributing different amounts of ammonia to flue gas regions with varying concentrations by changing the physical flow cross-section.

[0004] However, since the flue gas environment at the inlet of the denitrification reactor usually contains a large amount of high-concentration fly ash, accompanied by high temperature and corrosive volatile substances, the relevant technology relies on frequently moving internal mechanical components and tiny gaps to ensure movement during dynamic adjustment. Under harsh working conditions of long-term high-temperature baking and dust scouring, fly ash is prone to intrusion and hardening, depositing in the relative sliding gaps of moving sleeves or valves, resulting in reduced denitrification efficiency. Summary of the Invention

[0005] This application provides an adaptive ammonia injection device for flue gas denitrification, which is used to improve denitrification efficiency.

[0006] In a first aspect, an adaptive ammonia injection device for flue gas denitrification is provided, characterized by comprising: a data acquisition module for acquiring operating status data of the denitrification reactor, wake evaluation feedback data, and regional concentration statistical average values ​​for characterizing the spatial distribution differences in the flue; an AI calculation module for using the regional concentration statistical average values ​​as static spatial features and the operating status data as dynamic temporal features, inputting them into a preset deep neural network model for multi-source feature fusion calculation to obtain the target ammonia injection allocation corresponding to each fixed gap unit, wherein the fixed gap unit is mounted on a straight grid body and has no mechanical adjustment components with relative sliding inside, and the straight grid body is horizontally arranged in the inlet flue of the denitrification reactor along the flue gas flow direction; and a control and adjustment module for receiving the target ammonia injection allocation and converting the target ammonia injection allocation into a frequency converter for the total ammonia supply fluid pressure. The system includes control signals and zoned flow control signals for each independent flow regulation branch of the multi-path flow distributor. The multi-path flow distributor is equipped with multiple independent flow regulation branches, each of which is connected to a fixed void unit via a corrosion-resistant ammonia injection pipe. The ammonia injection execution module is used to adjust the total ammonia supply fluid pressure based on the frequency conversion control signal and drive the multi-path flow distributor according to the zoned flow control signal to guide ammonia gas to each fixed void unit for injection according to the target ammonia injection distribution amount. The data visualization module is used to correlate and display the execution status of each target ammonia injection distribution amount with the wake evaluation feedback data. When the wake evaluation feedback data is determined to produce a denitrification deviation that exceeds the preset performance degradation threshold, a hardware reconstruction warning is generated. The hardware reconstruction warning is used to indicate the reacquisition of the regional concentration statistical average value and the replacement of the pre-installed fixed void unit.

[0007] By adopting the above technical solution and constructing a collaborative control framework of static hardware matching and dynamic software adjustment, the complex online mechanical adjustment is transformed into external fluid control. The use of fixed-gap units without moving components avoids jamming and wear problems in high-temperature and high-dust environments, thereby improving the long-term operational reliability of the flue gas denitrification adaptive ammonia injection device. Simultaneously, by combining AI models with multi-source data fusion calculations and a two-level intelligent diagnostic and early warning mechanism, the ammonia injection strategy achieves dual adaptation to spatial unevenness and temporal variations, thus improving denitrification efficiency while enhancing the stability of the flue gas denitrification adaptive ammonia injection device.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, the data acquisition module specifically includes: a spatial concentration difference pre-detection unit, used to arrange a sensor array in the grid installation area in the inlet flue of the denitrification reactor, continuously collect regional nitrogen oxide concentration distribution data within a preset period, and calculate the regional concentration statistical average value corresponding to each detection point after performing outlier removal preprocessing on the regional nitrogen oxide concentration distribution data; a real-time operating condition acquisition unit, used to synchronously read the flue gas flow rate data, flue gas temperature data, flue gas pressure data and the actual ammonia injection amount of the current denitrification unit through a multi-dimensional sensor group, and fuse them to form the operating status data input to a preset deep neural network model; and a wake efficiency feedback unit, used to monitor the downstream outlet flue gas status after the catalyst reaction, obtain the real-time nitrogen oxide concentration deviation value and ammonia escape concentration value after denitrification, and extract the transient change rate of unit load and the fluctuation amplitude of the unit inlet operating condition parameters, and combine them to form wake evaluation feedback data.

[0009] By adopting the above technical solution, the data acquisition module subdivides the data acquisition function into three dimensions: spatial pre-detection, real-time operating conditions, and wake feedback, achieving a more comprehensive perception of static prior features, dynamic instantaneous states, and final control effects. This multi-dimensional, multi-timescale data input provides the AI ​​computing module with more accurate decision-making basis, thereby improving the accuracy and robustness of the flue gas denitrification adaptive ammonia injection device.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the matching of the fixed gap unit with the regional concentration statistical average value is achieved through preset hardware optimization rules, specifically including: a concentration gradient division unit, used to pre-establish multiple concentration gradient intervals from low to high, and assign cylindrical detachable structures with corresponding standard aperture sizes as candidate fixed gap units for each concentration gradient interval; a static mapping pairing unit, used to compare the regional concentration statistical average value corresponding to each detection point with the concentration gradient interval one by one, determine the target gradient interval to which it belongs, and extract the standard aperture size bound to the target gradient interval; and a hardware pre-installation guidance unit, used to output an assembly diagram containing the coordinate positions of each detection point and its corresponding standard aperture size, to indicate the offline screw-in installation of the fixed gap unit in the grid installation area.

[0011] By adopting the above technical solution and establishing a standardized concentration-aperture matching rule and offline installation process, the complex hardware selection problem is transformed into a clear and executable engineering procedure. This approach not only ensures that the aperture of the fixed-gap unit can more scientifically match the inherent concentration distribution of the flue, but also simplifies the difficulty of on-site installation and replacement work by generating assembly diagrams.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, a preset deep neural network model is included, specifically comprising: a multi-scale sequence decomposition unit, used to perform multi-scale decomposition of the time-series features of nitrogen oxide concentration in the operating state data using wavelet transform, to obtain a high-frequency fluctuation feature vector characterizing the flue gas mutation point; a time-dependent capture unit, used to input the high-frequency fluctuation feature vector and static spatial features into the gated recurrent unit branch built into the model, and calculate the evolution correlation features between the device load and flue gas parameters through a long short-term memory mechanism; and an autoregressive multi-step prediction unit, used to perform rolling iterative calculations based on the evolution correlation features, combined with the residual jump connection structure set by the model to prevent the deep network gradient from vanishing, and output the target ammonia injection allocation corresponding to each fixed gap unit at the next control moment.

[0013] By adopting the above technical solution and constructing a deep neural network structure specifically designed for spatiotemporal feature fusion, the predictive performance of the AI ​​model is improved. Wavelet transform effectively captures abrupt changes in operating conditions, gated recurrent units excel at handling long-term dependencies in time series, while residual connections and autoregressive prediction mechanisms ensure the stability of deep network training and the continuity of prediction. This allows the AI ​​computing module to gain a deeper understanding of the complex dynamics of the flue gas denitrification process, thereby outputting a more accurate ammonia injection allocation strategy.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the control and regulation module specifically includes: a global pressure supply calculation unit, used to mathematically sum the target ammonia injection allocation corresponding to all independent flow regulation branches to obtain the total target ammonia injection amount, and to map the total target ammonia injection amount into a frequency conversion control signal according to a fluid dynamics conversion function; and an independent channel proportional analysis unit, used to calculate the dynamic allocation ratio of the target ammonia injection allocation of each fixed gap unit in the total target ammonia injection amount, and to convert the dynamic allocation ratio into a zoned flow regulation control signal for controlling the opening of the electric valve of the corresponding branch.

[0015] By adopting the above technical solutions, the global pressure calculation ensures that the total ammonia supply capacity can quickly respond to changes in the overall load, while the independent channel ratio analysis ensures that the fine allocation strategy calculated by AI can be executed more accurately, thereby improving the performance of the flue gas denitrification adaptive ammonia injection device.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, the AI ​​computing module further includes: a multi-source dimension elimination unit, used to call the Z-score normalization algorithm to normalize the running state data with different physical dimension properties, thereby eliminating the gradient shift caused by the extreme value differences between heterogeneous data to the multi-source feature fusion calculation process.

[0017] By adopting the above technical solution and standardizing the data before inputting it into the AI ​​model, the negative impact of differences in physical dimensions and numerical ranges such as flue gas flow, temperature, and pressure on model training and prediction is eliminated. This reduces the problem of features with larger numerical values ​​dominating the gradient descent direction, enabling the model to learn the importance of all input features fairly, thereby improving the stability and accuracy of multi-source feature fusion calculation.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, the ammonia injection execution module specifically includes: a global variable frequency pressurization unit, used to receive a variable frequency control signal and adjust the global pressure of the ammonia supply fluid from the ammonia storage tank through a variable frequency ammonia pump to meet the dynamic pressure requirements of the main pipe corresponding to the total target ammonia injection volume; a flow guiding and limiting unit, used to receive a zone flow regulation control signal and adjust the fluid cross-sectional area of ​​each independent flow regulation branch inside the multi-path flow distributor to split the pressurized ammonia gas and inject it into each corresponding fixed gap unit; and a forced constant flow injection unit, used to force the input ammonia gas to be injected through the fixed gap unit through a physically preset static aperture.

[0019] By adopting the above technical solutions, the global variable frequency pressurization unit ensures rapid response of the total volume, the flow guiding and limiting unit located outside the reactor achieves safe and convenient fine distribution, and the forced constant flow injection unit inside the reactor completes the final injection task with the simplest structure. This structural design externalizes the complex adjustment functions and simplifies the internal components, enabling control commands to be more accurately and stably converted into physical ammonia injection actions.

[0020] In conjunction with some embodiments of the first aspect, in some embodiments, the data visualization module specifically includes: an efficiency status association mapping unit, used to synchronously compare and display on the visualization interface the execution curve of the target ammonia injection allocation aligned by the time axis and the actual denitrification index curve in the wake evaluation feedback data; a software elastic fine-tuning trigger unit, which activates the model online learning mechanism when the denitrification deviation falls into the preset safety adjustment fuzzy domain, and corrects the target ammonia injection allocation by relying on the preset deep neural network model without changing the hardware configuration; and a hardware over-limit fuse alarm unit, which fuses the model online learning mechanism and forcibly triggers the issuance of a hardware reconstruction warning when the denitrification deviation exceeds the preset safety adjustment fuzzy domain and reaches the preset performance degradation threshold.

[0021] By adopting the above technical solutions, the performance status correlation mapping provides operators with intuitive full-chain monitoring of decision-making, execution, and effect. The software elastic fine-tuning mechanism enables the device to self-repair minor performance drifts, while the hardware over-limit fuse alarm enables predictive maintenance early warning for systemic failures, reducing the difficulty and cost of operation and maintenance.

[0022] In conjunction with some embodiments of the first aspect, in some embodiments, the triggering logic of the data visualization module is as follows: when it is determined that the positive or negative deviation of the real-time nitrogen oxide concentration after denitrification continuously deviates from the set value by more than the safety tolerance, or the ammonia escape concentration value exceeds the warning line, and the duration of the over-limit state exceeds the preset time window, it is determined that it falls into the preset safety adjustment fuzzy domain, and the software elastic fine-tuning triggering unit is triggered; when the average absolute error between the predicted ammonia injection amount and the actual ammonia injection amount of the model online learning mechanism continuously exceeds the preset generalization tolerance, or when it is determined that the transient change rate of unit load and the fluctuation amplitude of the inlet operating parameters in the wake assessment feedback data exceed the design response extreme value of the denitrification device, it is determined that the preset performance degradation threshold is reached, and the hardware over-limit fuse alarm unit is triggered.

[0023] By adopting the above technical solution, a composite judgment rule based on performance indicators, model errors, and the severity of operating conditions can more accurately distinguish between temporary fluctuations and fundamental degradation of device performance, enabling software adaptation and hardware reconfiguration early warning to be triggered at more appropriate times, thus reducing false alarms and missed alarms.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. The adaptive ammonia injection unit for flue gas denitrification (FGD) transforms complex online mechanical regulation into external fluid control by constructing a collaborative control framework of static hardware matching and dynamic software adjustment. It utilizes fixed, gapless units without moving parts to avoid jamming and wear issues in high-temperature, high-dust environments, thereby improving the long-term operational reliability of the FGD adaptive ammonia injection unit. Simultaneously, by combining AI models with multi-source data fusion calculations and a two-level intelligent diagnostic and early warning mechanism, it achieves dual adaptation of the ammonia injection strategy to spatial unevenness and temporal variations, thus improving denitrification efficiency while enhancing the stability of the FGD adaptive ammonia injection unit.

[0026] 2. By constructing a deep neural network structure specifically designed for spatiotemporal feature fusion, the predictive performance of the AI ​​model was improved. Wavelet transform effectively captures abrupt changes in operating conditions, gated recurrent units excel at handling long-term dependencies in time series, while residual connections and autoregressive prediction mechanisms ensure the stability of deep network training and the continuity of prediction. This allows the AI ​​computing module to gain a deeper understanding of the complex dynamics of the flue gas denitrification process, thereby outputting a more accurate ammonia injection allocation strategy.

[0027] 3. The flue gas denitrification adaptive ammonia injection device is equipped with a diagnostic and optimization mechanism that includes software adaptive fine-tuning and hardware reconfiguration early warning. For small fluctuations, the AI ​​model can learn and correct itself online. For systematic deviations caused by long-term drift in operating conditions, it can automatically identify and issue hardware reconfiguration early warnings, guiding maintenance personnel to perform targeted offline maintenance, thereby reducing the difficulty and cost of maintenance. Attached Figure Description

[0028] Figure 1 This is a module architecture diagram of an adaptive ammonia injection device for flue gas denitrification in an embodiment of this application.

[0029] Figure 2 This is a schematic diagram of the data acquisition module of an adaptive ammonia injection device for flue gas denitrification in an embodiment of this application.

[0030] Figure 3 This is a schematic diagram of the control and adjustment module of an adaptive ammonia injection device for flue gas denitrification in an embodiment of this application.

[0031] Figure 4 This is a schematic diagram of the ammonia injection execution module of an adaptive ammonia injection device for flue gas denitrification in an embodiment of this application.

[0032] Figure 5 This is a schematic diagram of the data visualization module of an adaptive ammonia injection device for flue gas denitrification in an embodiment of this application. Detailed Implementation

[0033] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0034] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0035] This application provides an adaptive ammonia injection device for flue gas denitrification, which is used to improve denitrification efficiency.

[0036] This application provides an adaptive ammonia injection device for flue gas denitrification, mainly consisting of an ammonia injection grid installed inside the inlet flue of the denitrification reactor, and an ammonia injection execution and control unit located outside the reactor. Specifically, it includes: a straight grid body, which is typically made of high-temperature and corrosion-resistant metal material, forming a hollow frame structure, horizontally arranged in the inlet flue of the denitrification reactor along the flue gas flow direction. The grid body has multiple standardized mounting holes evenly distributed on it, providing a unified assembly interface for subsequent fixed-gap units. Multiple fixed-gap units, which are the core injection elements of this technical solution, are cylindrical and detachable structures. Each unit can be easily screwed into the corresponding mounting hole on the grid body via threads on its outer wall, achieving modular, rapid installation and replacement. To ensure sealing, high-temperature and corrosion-resistant fluororubber seals are specially embedded at the connection points between the fixed-gap units and the mounting holes, thereby preventing ammonia leakage from the connection points. Each fixed-gap unit has a pre-set static aperture on its sidewall, the aperture size of which is selected based on the statistical average concentration of the area corresponding to that installation location. An ammonia injection execution and control unit integrates all functions of ammonia supply, pressurization, distribution, and control, and is located entirely outside the reactor for easy maintenance. This unit includes a connected ammonia storage tank, a variable frequency ammonia pump, and a multi-path flow distributor. The multi-path flow distributor is key to achieving precise distribution; it has multiple independently adjustable flow branches, each connected one-to-one to the corresponding fixed-gap unit on the flue gas grating via an ammonia injection pipe made of corrosion-resistant material (such as stainless steel). A PLC controller is also integrated within this ammonia injection execution and control unit. As the hardware gateway and execution brain of the entire device, the PLC controller is responsible for receiving and parsing the frequency conversion control signals and zone flow regulation control signals issued by the upper-level AI computing module and control and regulation module. It then converts these high-level instructions into specific electrical control actions in real time to drive the frequency conversion ammonia pump to adjust its speed and drive the valves of each independent flow regulation branch in the multi-channel flow distributor to change their opening degree, thereby completing the entire closed-loop control.

[0037] Please see Figure 1 This is a module architecture diagram of the flue gas denitrification adaptive ammonia injection device in the embodiments of this application.

[0038] The flue gas denitrification adaptive ammonia injection device includes:

[0039] The data acquisition module 101 is used to acquire the operating status data of the denitrification reactor, the wake assessment feedback data, and the regional concentration statistical average value used to characterize the spatial distribution differences of the flue gas.

[0040] The denitrification reactor refers to the equipment used for selective catalytic reduction (SCR) reactions in the flue gas denitrification adaptive ammonia injection unit. Operating status data refers to the set of dynamic parameters characterizing the current operating conditions of the flue gas denitrification adaptive ammonia injection unit, such as flue gas flow rate, temperature, pressure, and the current total ammonia water or ammonia gas injection rate. These data are characterized by real-time changes. Wake-stream evaluation feedback data refers to downstream measurement data used to evaluate the denitrification effect and the unit's operating status, such as the deviation of the residual nitrogen oxide concentration in the flue gas after denitrification from the target value, the concentration of ammonia gas directly emitted without participating in the reaction (i.e., ammonia slip concentration), and the unit load change rate reflecting the stability of the operating conditions. The regional concentration statistical average refers to the time average of nitrogen oxide concentrations collected and statistically analyzed at multiple detection points over a relatively long time period at a specific cross-section of the flue gas inlet of the denitrification reactor. This value is considered a relatively stable static physical quantity used to characterize the inherent differences in concentration distribution at different locations within the flue gas duct.

[0041] Specifically, the data acquisition module 101 operates continuously during the operation of the flue gas denitrification adaptive ammonia injection unit. Its core function is to perceive the status of the flue gas denitrification adaptive ammonia injection unit from multiple dimensions. In the initial stage of the flue gas denitrification adaptive ammonia injection unit's commissioning or hardware configuration, this module activates its subordinate spatial concentration pre-detection unit to conduct continuous monitoring in the grid installation area for several days to obtain the regional concentration statistical average. During the normal operation of the flue gas denitrification adaptive ammonia injection unit, this module collects real-time and synchronous operating status data such as flue gas flow rate, temperature, and pressure through sensors deployed throughout the flue, forming a dynamic time-series data stream reflecting the instantaneous operating conditions of the unit. Simultaneously, it continuously monitors the flue gas composition at the denitrification reactor outlet, obtaining wake assessment feedback data including denitrification efficiency and ammonia slip level. These data from different sources and with different time scales (static spatial distribution and dynamic time series) collectively constitute the data foundation for the subsequent AI computing module's intelligent decision-making.

[0042] In some embodiments, the data acquisition module 101 can be implemented in several ways: Optionally, the module can be built based on a distributed I / O system. First, various sensors are physically installed at the flue gas inlet, outlet, and key nodes of the pipeline, such as using Pitot tubes or thermal mass flow meters to measure flue gas flow and using armored thermocouples to measure flue gas temperature. Second, these sensors are connected to distributed I / O substations via a fieldbus protocol. Finally, a central data acquisition server polls each I / O substation, converts the acquired signals into engineering data in a unified format, adds a timestamp, and stores it in a real-time database. Optionally, the module can also be implemented based on a wireless sensor network of the Internet of Things (IoT). First, intelligent sensor nodes integrating corresponding sensing elements, data processing units, and wireless transceiver modules are deployed at each monitoring point. Second, after each node independently completes data acquisition and preliminary processing, it wirelessly transmits the data to an IoT gateway deployed on-site via a low-power wide area network. Finally, the IoT gateway aggregates and parses the data and uploads it to a cloud platform database via Ethernet or a 5G network.

[0043] AI computing module 102 is used to input the average regional concentration as a static spatial feature and the operating status data as a dynamic temporal feature into a preset deep neural network model for multi-source feature fusion calculation to obtain the target ammonia injection allocation corresponding to each fixed gap unit. The fixed gap unit is assembled on the straight grid body and has no mechanical adjustment components with relative sliding inside. The straight grid body is horizontally arranged in the inlet flue of the denitrification reactor along the flue gas flow direction.

[0044] Among them, dynamic temporal characteristics refer to data sequences that change continuously over time, reflecting the real-time operating conditions of the flue gas denitrification adaptive ammonia injection unit; here, it refers to operational status data. The preset deep neural network model refers to a pre-trained artificial intelligence model with multiple nonlinear processing units. Multi-source feature fusion calculation represents the process of effectively combining and calculating static spatial characteristics and dynamic temporal characteristics—two different types of data—within the model. The target ammonia injection allocation refers to the amount of ammonia gas to be injected in the next control cycle for each independent fixed-gap unit, output by the model calculation. A fixed-gap unit refers to an ammonia injection element with a fixed physical structure, whose nozzle size is determined based on the NO₂ concentration at its location. x The average concentration was selected, and there were no mechanical adjustment components with relative sliding inside.

[0045] Specifically, the AI ​​calculation module 102 performs a complete prediction calculation once per control cycle. First, to eliminate differences in dimensions and numerical ranges between different physical quantities (such as flow rate in m³ / h and temperature in °C), the module normalizes the operating status data using a multi-source dimension elimination unit (e.g., by calling the Z-score normalization algorithm). Next, the normalized dynamic temporal features and static spatial features are fed into a pre-defined deep neural network model. The multi-scale sequence decomposition unit within the model uses wavelet transform to decompose the temporal data to capture abrupt changes in flue gas parameters (high-frequency fluctuation feature vectors). Subsequently, the temporal dependency capture unit utilizes structures such as gated recurrent units (GRUs) to fuse high-frequency fluctuation feature vectors and static spatial features, calculating the evolutionary correlation between dynamic parameters of the system through a long short-term memory mechanism. Finally, the autoregressive multi-step prediction unit, based on this evolutionary correlation feature and using a residual jump connection structure to prevent gradient vanishing, predicts the target ammonia injection allocation to each fixed-gap unit that best matches the current operating conditions and spatial concentration distribution at future moments.

[0046] In some embodiments, the AI ​​computing module 102 can be implemented in several ways: Optionally, the module can be implemented using a sequence-to-sequence (Seq2Seq) model based on an encoder-decoder architecture. First, a sequence of historical operational state data over a period of time is input into an encoder composed of a recurrent neural network (such as LSTM), compressing it into a fixed-length context vector. Second, the regional concentration statistical average is used as the initial state of the decoder or concatenated with the context vector and input together. Finally, the decoder generates the target ammonia injection allocation amount corresponding to each fixed-gap unit based on these inputs. Optionally, the module can be implemented using a hybrid model of convolutional neural networks (CNN-Attention) based on an attention mechanism. First, a one-dimensional convolutional neural network (1D-CNN) is used to process the temporal sequence of operational state data to efficiently extract local patterns and short-term fluctuation features. Second, the spatial feature map composed of the regional concentration statistical average is concatenated with the temporal features extracted by the CNN. Third, a self-attention mechanism is introduced to allow the model to autonomously learn the importance weights of features at different time points and spatial locations. Finally, the target ammonia injection allocation of all fixed-gap units is output in parallel at one time through a fully connected layer based on the weighted fusion features.

[0047] In some embodiments, the preset deep neural network model needs to be trained offline before being deployed in online applications. The training process includes: collecting a large amount of multi-source data from the historical operation of the flue gas denitrification device as a training sample set; using the optimal ammonia injection allocation amount that achieves the standards for both outlet nitrogen oxide concentration and ammonia slip in manual debugging or historical records as a supervision label; using mean squared error as a loss function to measure the difference between the model's predicted allocation amount and the supervision label; calling the Adam or SGD optimizer; and continuously iteratively updating the network parameters inside the model (including wavelet transform weights, gated recurrent unit branches, and residual jump connection structures) through the backpropagation algorithm until the loss function converges to a preset accuracy range, thereby obtaining the trained preset deep neural network model.

[0048] The control and regulation module 103 is used to receive the target ammonia injection distribution amount, convert the target ammonia injection distribution amount into a frequency conversion control signal for the total ammonia supply fluid pressure, and a zone flow regulation control signal for each independent flow regulation branch of the multi-path flow distributor. The multi-path flow distributor is equipped with multiple independent flow regulation branches, and each independent flow regulation branch is connected to the fixed gap unit through a corrosion-resistant ammonia injection pipe.

[0049] The control and regulation module 103 represents the intermediate conversion layer connecting AI decision-making and physical execution. The total ammonia supply fluid pressure refers to the total power source driving the ammonia flow throughout the entire ammonia injection unit; its magnitude determines the unit's total ammonia supply capacity. The frequency converter control signal is an analog or digital signal used to instruct the frequency converter to adjust the motor speed. An independent flow regulation branch refers to a branch pipe that branches off from the total ammonia supply pipeline and supplies ammonia to a specific fixed-gap unit or group of units. The zoned flow regulation control signal refers to the control signal for the flow regulation valves on each independent flow regulation branch, typically used to set the valve opening percentage.

[0050] Specifically, after receiving a set of target ammonia injection allocations from the AI ​​calculation module, the control and regulation module 103 immediately begins calculation. This module first mathematically sums the target ammonia injection allocations for all independent flow regulation branches to obtain the total target ammonia injection amount required by the device within the current control cycle. Then, based on a preset fluid dynamics conversion function, this total target ammonia injection amount is mapped to the main pipe pressure required to generate that flow rate, and finally, a corresponding variable frequency control signal is generated and sent to the variable frequency ammonia pump. Simultaneously, the module calculates in parallel the proportion of each target ammonia injection allocation to the total target ammonia injection amount. Then, these dynamic allocation proportions are converted into specific opening commands for the electric valves of each independent flow regulation branch, i.e., zoned flow control signals, and sent to the actuator of the multi-path flow distributor.

[0051] In some embodiments, the function of the control and regulation module 103 can be implemented in several ways: Optionally, the module can be implemented by a dedicated programmable logic controller (PLC). First, a function block is written in the PLC's program logic to receive the target ammonia injection distribution array from the host computer. Second, an arithmetic instruction is written to sum the array, and a lookup table instruction is used to perform fluid dynamics conversion, outputting an analog signal to the AI ​​terminal of the frequency converter. Then, loop logic is written to perform division on each element in the array to obtain the distribution ratio, and then multiply it by the control range corresponding to the full stroke of the valve, driving each electric valve positioner through the PLC's analog output module. Optionally, the module can be integrated into the controller of a distributed control system (DCS). First, a control strategy is configured to receive the AI ​​calculation results. Second, the calculation block built into the DCS is used to complete the summation and ratio calculation. Finally, using the DCS's built-in PID control block or manual / automatic switching block, the calculated total flow is used as the setpoint of the PID controller, whose output directly controls the main pipe pressure; at the same time, the calculated distribution ratio of each branch is used as the remote setpoint of each downstream flow control loop.

[0052] The ammonia injection execution module 104 is used to adjust the total ammonia supply fluid pressure based on the frequency conversion control signal, and drive the multi-channel flow distributor according to the zone flow control signal to guide the ammonia gas to each fixed gap unit for injection according to the target ammonia injection distribution amount.

[0053] The multi-path flow distributor is a key piece of equipment. Located outside the denitrification reactor, it contains multiple parallel, independently adjustable flow control branches. Flow diversion refers to distributing the ammonia gas in the main pipe into various independent flow control branches according to a specific ratio. Injection refers to the high-speed ejection of ammonia gas under pressure through the orifices of a fixed-gap unit, mixing it with the flue gas.

[0054] Specifically, the ammonia injection execution module 104 executes instructions from the control and regulation module. Upon receiving the frequency conversion control signal, its internal global pressurization device (such as a variable frequency ammonia pump) immediately adjusts its operating state and regulates the main pipe pressure to meet the power requirements corresponding to the current total target ammonia injection volume. Simultaneously, the multi-path flow distributor located outside the reactor receives its respective zone flow control signals from multiple independent regulating valves. Each regulating valve adjusts its valve core position according to the signal instructions, changing the fluid flow cross-sectional area, thereby precisely splitting and guiding the pressurized ammonia in the main pipe according to the dynamic distribution ratio calculated by AI, to independent ammonia injection pipelines connected to different fixed-gap units. Finally, the precisely distributed ammonia enters its corresponding fixed-gap unit and is forcibly injected through its internal preset static orifice.

[0055] In some embodiments, the ammonia injection execution module 104 can be implemented in several ways: Optionally, the module can employ a combination of a variable frequency pump and an electric ball valve array. The global pressurization device is a screw pump linked to the variable frequency drive. The multi-path flow distributor consists of a set of parallel V-type open electric regulating ball valves with intelligent valve positioners. The zone flow control signal drives the valve positioner to control the rotation angle of the ball valve, thereby regulating the flow of each branch. Optionally, the module can employ a constant pressure ammonia supply + proportional integral valve group scheme. The global pressurization device maintains a constant mains pressure through a pressure closed-loop control circuit. The multi-path flow distributor employs a set of dynamic flow balancing valves or electric two-way regulating valves. The zone flow control signal directly sets the target opening degree of each valve.

[0056] The data visualization module 105 is used to associate and display the execution status of each target ammonia injection allocation with the wake evaluation feedback data. When it is determined that the wake evaluation feedback data produces a denitrification deviation that exceeds the preset performance degradation threshold, a hardware reconstruction warning is generated. The hardware reconstruction warning is used to indicate the reacquisition of the regional concentration statistical average value and the replacement of the pre-installed fixed gap unit.

[0057] The "associated display" refers to the synchronized presentation of the AI ​​model's output commands and the actual operating effects of the flue gas denitrification adaptive ammonia injection device on a single interface, following a unified timeline. The performance degradation threshold is a pre-set critical value used to determine when the performance of the flue gas denitrification adaptive ammonia injection device has severely deteriorated and cannot be recovered through software adaptive adjustment. Denitrification deviation refers to the gap between the actual denitrification effect and the expected target. The hardware reconfiguration warning is the highest level of alert, explicitly indicating that the current fixed-aperture hardware configuration is no longer suitable for the operating conditions and requires physical adjustments.

[0058] Specifically, the data visualization module 105 provides users with a dynamic monitoring dashboard. Operators can see the set curve and actual flow execution curve for each ammonia injection rate, as well as the impact of these adjustments on the downstream outlet NO. x Real-time impact curves of concentration and ammonia slip concentration. The flue gas denitrification adaptive ammonia injection device incorporates a two-level intelligent diagnostic logic: when the detected denitrification deviation is small and falls within a safe adjustment fuzzy domain, the online learning mechanism of the AI ​​model is activated, automatically correcting the ammonia injection strategy by fine-tuning the model parameters without changing the hardware. However, when the denitrification deviation continues to expand and reaches the preset performance degradation threshold (e.g., even if the model has made every effort to adjust, the outlet NO...), x(If ammonia escape continues to exceed the standard for a long time), the flue gas denitrification adaptive ammonia injection device will immediately shut down the online learning function of the model, generate and push a prominent hardware reconstruction warning, and remind the operator that the current hardware configuration is no longer applicable, and the machine must be shut down for maintenance, the concentration field pre-inspection must be carried out again, and the fixed gap unit on the grid must be replaced according to the new test results.

[0059] The above describes the module architecture diagram of the flue gas denitrification adaptive ammonia injection device in the embodiments of this application. The modules are described below.

[0060] Please refer to the following: Figure 2 This is a schematic diagram of the data acquisition module of the flue gas denitrification adaptive ammonia injection device in the embodiments of this application.

[0061] The data acquisition module 101 includes:

[0062] The spatial concentration difference pre-detection unit 1011 is used to arrange a sensor array in the grid installation area in the inlet flue of the denitrification reactor, continuously collect regional nitrogen oxide concentration distribution data within a preset period, and calculate the statistical average value of regional concentration corresponding to each detection point after performing outlier removal preprocessing on the regional nitrogen oxide concentration distribution data.

[0063] The grille installation area refers to a specific cross-section within the flue where ammonia injection grilles are planned for installation. The sensor array is a collection of multiple independent nitrogen oxide sensors arranged in a matrix or linear pattern. The preset period typically refers to a timeframe sufficient to reflect stable operating conditions, such as 72 hours. Regional nitrogen oxide concentration distribution data refers to the set of raw concentration readings collected by the sensor array, encompassing both temporal and spatial dimensions. Outlier removal preprocessing is a data cleaning technique used to remove distorted data points caused by momentary sensor malfunctions. The regional concentration statistical average is the result obtained by arithmetically averaging all valid data for each detection point throughout the preset period after preprocessing.

[0064] Specifically, the workflow of this unit is divided into four stages: deployment, data acquisition, processing, and output. First, during the shutdown and maintenance of the flue gas denitrification adaptive ammonia injection unit, technicians will arrange the nitrogen oxide sensor array on the flue section where the ammonia injection grid will be installed, according to predetermined grid coordinates. After the flue gas denitrification adaptive ammonia injection unit is started and running stably, the unit begins to continuously record the concentration readings of each sensor within a preset period. After data acquisition, the unit's built-in algorithm preprocesses the dataset, for example, using the 3σ principle to remove isolated extreme values. Finally, the cleaned data is grouped by detection point, and the arithmetic mean of the concentration at each detection point within the period is calculated. This group represents the statistical average of the inherent concentration field distribution of the flue section, which will be output and used as the basis for subsequent matching of fixed-gap units with the regional concentration statistical average.

[0065] In some embodiments, the spatial concentration pre-detection unit 1011 can be implemented in several ways: Optionally, the unit can be implemented using a portable multi-point flue gas analyzer. First, multiple standard measurement point interfaces are reserved on the flue. Second, during pre-detection, multiple sampling probes of the analyzer are inserted into these interfaces respectively. Finally, after the acquisition cycle ends, a report of the average concentration of each measurement point is directly exported from the instrument. Optionally, the unit can be implemented using a removable online monitoring probe array. First, a permanent bracket containing multiple independently insertable and removable electrochemical sensor probes is installed. Second, all probes are inserted into place for measurement during pre-detection. Finally, data processing and average calculation are completed by host software.

[0066] The real-time operating condition acquisition unit 1012 is used to synchronously read the flue gas flow rate data, flue gas temperature data, flue gas pressure data, and the actual ammonia injection amount of the current denitrification unit through a multi-dimensional sensor group, and integrate them to form the operating status data input to the preset deep neural network model.

[0067] In this context, a multidimensional sensor array refers to a series of different types of sensors used to measure various physical quantities. Synchronous reading refers to acquiring all relevant data at the same point in time or within a very small time window to ensure data temporal consistency. Fusion to construct operational status data refers to combining multiple discrete physical quantity data points into a multidimensional vector or data frame, which serves as the input to an AI model at a specific moment.

[0068] Specifically, during the operation of the flue gas denitrification adaptive ammonia injection unit, this unit operates at a fixed sampling frequency (e.g., once per second). Through devices such as flue gas flow sensors, temperature sensors, pressure sensors, and ammonia flow meters deployed in the flue and ammonia pipelines, it simultaneously acquires the instantaneous values ​​of all key operating parameters. These raw electrical or digital signals are transmitted to the data processing unit, where, after signal conditioning, filtering, and engineering unit conversion, they are combined into a multi-dimensional data vector containing all real-time operating information. This vector is marked with a precise timestamp, forming a time-series data stream, which is continuously provided to the AI ​​computing module as the basis for its dynamic predictions.

[0069] In some embodiments, the real-time operational status acquisition unit 1012 can be implemented in several ways: Optionally, the unit can directly utilize the data interface of an existing distributed control system (DCS) on-site. First, a data communication task is configured in the DCS to read the real-time values ​​of various sensors connected to the system. Second, these data are provided in batches and at high speed to the upper-level AI computing server via the OPC (Open Platform Communication) protocol. Finally, the software on the AI ​​server side is responsible for combining the received data into the input format required by the model. Optionally, the unit can be built as an independent edge computing acquisition system. First, all relevant sensors are connected to an edge computing gateway via hardwiring or wireless means. Second, a data acquisition and preprocessing program is run on the gateway to synchronously read data from various sources in real time. Finally, the gateway sends the processed, structured operational status data packets directly to the AI ​​computing module via Ethernet.

[0070] The wake performance feedback unit 1013 is used to monitor the status of the downstream outlet flue gas after the catalyst reaction, obtain the real-time nitrogen oxide concentration deviation value and ammonia slip concentration value after denitrification, and extract the transient change rate of unit load and the fluctuation range of the inlet operating parameters of the unit, and combine them to form the wake evaluation feedback data.

[0071] Among them, the real-time nitrogen oxide concentration deviation value refers to the NO measured at the outlet. x The difference between the concentration and the environmental protection requirements or set target values. Ammonia slip concentration refers to the concentration of ammonia gas that is discharged with the flue gas without participating in the reaction. The transient rate of change in unit load and the fluctuation range of unit inlet operating parameters refer to derived indicators obtained from systems such as DCS that reflect the severity of operating conditions. Composite wake assessment feedback data refers to a comprehensive performance evaluation dataset formed by combining directly measured values ​​and calculated derived indicators.

[0072] Specifically, this unit uses a continuous online flue gas monitoring system (CEMS) installed on the outlet flue of the denitrification reactor to acquire real-time NOx and ammonia slip concentrations after denitrification. Simultaneously, it communicates with the power plant's DCS or SIS system to obtain parameters such as the unit's power generation load and boiler combustion conditions, and calculates their rate of change or fluctuation amplitude per unit time. This integrated data is used directly in the data visualization module for performance display; furthermore, it serves as the core basis for triggering software fine-tuning or hardware reconfiguration warnings. For example, when NO... x When the concentration deviation or ammonia slip value continues to exceed the standard, it indicates that the current control strategy has failed and an adjustment mechanism needs to be activated.

[0073] Please see Figure 3 This is a schematic diagram of the control and adjustment module of the flue gas denitrification adaptive ammonia injection device in the embodiments of this application.

[0074] The control and adjustment module 103 includes:

[0075] The global pressure supply calculation unit 1031 is used to mathematically sum the target ammonia injection allocation corresponding to all independent flow regulation branches to obtain the total target ammonia injection amount, and then map the total target ammonia injection amount into a frequency conversion control signal according to the fluid dynamics conversion function.

[0076] The total target ammonia injection rate is the sum of all zone target rates output by the AI ​​model, representing the theoretically required total ammonia volume under the current operating conditions. The fluid dynamics transformation function is a mathematical model or lookup table describing the pipeline characteristics of the ammonia injection unit, i.e., the relationship between the total flow rate and the main pipeline pressure (or pump speed / frequency) required to generate that flow rate. The frequency converter control signal is the command signal ultimately output to the frequency converter to adjust the ammonia pump speed.

[0077] Specifically, when the unit receives a set of target ammonia injection allocation values ​​(e.g., an array containing N elements) output by the AI ​​calculation module, it first performs a summation operation, adding the values ​​of all elements in the array to obtain a scalar value, i.e., the total target ammonia injection amount. Next, the unit uses this total target ammonia injection amount as input, substituting it into a pre-established fluid dynamics conversion function for calculation. This function can be a polynomial fitted based on pump performance curves and pipeline resistance formulas, or it can be a discrete lookup table. The result of the calculation or table lookup is the inverter operating frequency required to achieve this total flow rate. Finally, this frequency value is converted into a standard analog signal or sent to the inverter via bus communication, completing the generation of the total control command.

[0078] In some embodiments, the function of the global pressure supply calculation unit 1031 can be implemented in several ways: Optionally, the unit can be implemented using ladder diagrams or structured text logic written in a PLC. First, a data block is defined to store the target ammonia injection distribution array. Second, loop instructions and addition instructions are used to accumulate the array elements. Finally, a piecewise linear interpolation function block (or a series of comparison and calculation instructions) is used to map the fluid dynamics function and write the result to the analog output channel of the PLC. Optionally, the unit can be implemented using high-level language programming on an industrial control computer. First, data is received through a communication interface. Second, a numerical calculation library (such as NumPy) is called for summation. Finally, the mapping is completed by consulting a preset configuration file (such as a lookup table in JSON or XML format) or calling a function, and the frequency conversion control signal is output through the analog output port of the data acquisition card (DAQ).

[0079] The independent channel proportional analysis unit 1032 is used to calculate the dynamic distribution ratio of the target ammonia injection amount of each fixed gap unit in the total target ammonia injection amount, and convert the dynamic distribution ratio into a zone flow regulation control signal for controlling the opening degree of the corresponding branch electric valve.

[0080] The dynamic allocation ratio refers to the percentage of total flow that should be allocated to each branch, and this ratio changes in real time with the output of the AI ​​model. The zone flow control signal is a specific signal used to instruct the regulating valve of each branch to act, and its magnitude is usually proportional to the valve opening.

[0081] Specifically, the independent channel proportional analysis unit receives the target ammonia injection allocation array and obtains the total target ammonia injection amount calculated by the global supply pressure calculation unit. Then, this unit performs a division operation on each element in the array (i.e., the target ammonia injection allocation for each branch), with the divisor being the total target ammonia injection amount, thus obtaining the dynamic allocation ratio corresponding to that branch. Next, the unit multiplies this ratio value by a control range constant representing the full stroke of the valve (e.g., 16mA for a 4-20mA signal), and adds a reference value (e.g., 4mA), thereby generating a standard analog control signal linearly related to the allocation ratio. This process is repeated for all branches, ultimately generating a set of parallel zone flow control signals, which are sent to each electric valve in the multi-channel flow distributor.

[0082] In some embodiments, the function of the independent channel proportional analysis unit 1032 can be implemented in several ways: Optionally, this unit can be implemented in the same PLC program segment as the global supply pressure calculation unit. First, after calculating the total target ammonia injection amount, it is stored in a general-purpose register. Second, in a loop, the target allocation amount of each branch is read sequentially, and floating-point division is performed to obtain the ratio. Finally, the ratio is multiplied by the control range constant, and the result is sent to the corresponding analog output channel. Optionally, this unit can be implemented by a series of independent single-loop controllers. First, the target allocation amount of each branch output by the AI ​​is used as the remote setpoint (RSP) of each controller. Second, the total target ammonia injection amount is used as the common input of these controllers or broadcast via communication. Finally, each controller automatically completes the proportional calculation and outputs a control signal to drive the valve.

[0083] Please see Figure 4 This is a schematic diagram of the ammonia injection execution module of the flue gas denitrification adaptive ammonia injection device in the embodiments of this application.

[0084] The ammonia injection execution module 104 includes:

[0085] The global variable frequency pressurization unit 1041 is used to receive the variable frequency control signal and regulate the global pressure of the ammonia supply fluid from the ammonia storage tank through the variable frequency ammonia pump to meet the dynamic pressure requirements of the main pipeline corresponding to the total target ammonia injection volume.

[0086] Among them, the variable frequency ammonia pump is a pump whose motor speed is controlled by a frequency converter, which can flexibly adjust its outlet flow rate and pressure. An ammonia storage tank is a container for storing liquid ammonia or ammonia water. Main pipe dynamic pressure refers to the pressure exerted by the fluid due to flow in the main ammonia supply pipeline, which directly determines the total ammonia supply capacity of the unit.

[0087] Specifically, the core of this unit is the variable frequency ammonia pump. When its built-in frequency converter receives a frequency control signal from the control and regulation module, it adjusts the AC frequency output to the ammonia pump motor according to the signal magnitude. The motor speed changes accordingly, thereby adjusting the pumping capacity of the ammonia pump. The ammonia pump draws ammonia water or liquid ammonia from the ammonia storage tank, pressurizes and (if necessary) vaporizes it, and then sends it to the ammonia injection main pipe. The frequency converter's adjustment allows the pressure and flow rate of the main pipe to quickly and accurately match the total target ammonia injection amount calculated by the control system, providing a stable ammonia source that meets the total quantity requirements for subsequent refined allocation.

[0088] In some embodiments, the function of the global variable frequency pressurization unit 1041 can be implemented in several ways: Optionally, the unit may consist of a screw pump, a frequency converter, and necessary pipeline valves. First, the frequency converter receives a control signal. Second, the frequency converter drives the screw pump motor to rotate, drawing and pressurizing ammonia water from the storage tank. Finally, the pressurized ammonia water is transported through the main pipe. Optionally, the unit may employ a combination of a diaphragm metering pump and a frequency converter. First, the frequency converter controls the stroke frequency of the metering pump. Second, the diaphragm pump injects ammonia water into the device at a precise discharge rate. Finally, precise control of the total flow rate is achieved by adjusting the frequency.

[0089] The flow guiding and limiting unit 1042 is used to receive the zone flow regulation control signal, adjust the fluid cross-sectional area of ​​each independent flow regulation branch inside the multi-path flow distributor, and split the pressurized ammonia gas and inject it into the corresponding fixed gap unit.

[0090] The fluid cross-sectional area refers to the effective opening area inside the valve that allows fluid to pass through. Splitting and injecting refers to distributing the ammonia gas in the main pipe to various branch pipes according to different flow rates.

[0091] Specifically, the unit consists of a group of parallel-installed regulating valves with electric actuators, each forming an independent flow regulation branch. When the electric actuator of each regulating valve receives its corresponding zone flow control signal, the actuator drives the valve stem to move, changing the relative position between the valve core and the valve seat, thereby precisely adjusting the valve opening, i.e., changing the fluid cross-sectional area. Since the main pipe pressure is maintained at a level that meets the total flow requirement by the global frequency conversion pressurization unit, by adjusting the opening of each branch valve, the pressurized ammonia in the main pipe can be distributed to each independent branch pipe according to the proportion represented by the control signal. Each branch pipe is ultimately connected to one or a group of specific fixed-gap units, thus achieving the goal of injecting ammonia into different areas of the flue according to the distribution amount calculated by AI.

[0092] In some embodiments, the function of the flow guiding and limiting unit 1042 can be implemented in several ways: Optionally, the unit can consist of a set of electrically operated V-type ball valves with intelligent valve positioners. First, each valve positioner receives a 4-20mA zoned flow control signal. Second, the microprocessor inside the positioner calculates the target opening degree based on the input signal and drives the motor to precisely rotate the ball valve. Finally, the V-shaped opening ensures good regulation characteristics even at small opening degrees. Optionally, the unit can employ a set of electrically operated sleeve regulating valves. First, the electric actuator receives a control signal. Second, the actuator drives the valve stem to move linearly, changing the overlapping area of ​​the sleeve window. Finally, linear regulation of the flow rate is achieved by controlling the valve stem displacement.

[0093] The forced constant flow injection unit 1043 is used to force the input ammonia gas to be injected through a physically preset static aperture inside the fixed gap unit.

[0094] Among them, the physically preset static orifice diameter refers to the nozzle size and shape that are determined during manufacturing and cannot be changed during operation. Forced injection refers to ammonia gas being forced to exit through a narrow orifice at a high velocity under the influence of upstream and downstream pressure differences.

[0095] Specifically, ammonia gas with a specific flow rate, allocated from the flow guiding and limiting unit, is transported to the reactor interior through corrosion-resistant pipes and enters its corresponding fixed-void unit. Since the pore size of the fixed-void unit is fixed, according to fluid mechanics principles, when a certain flow rate of fluid passes through a fixed cross-section, its velocity is determined. Therefore, the function of this unit is to convert the upstream regulated flow signal into a jet signal with specific momentum and diffusion characteristics. Under the pressure within the unit's internal space, the input ammonia gas is forced to eject at high speed from the static pores on the sidewalls or ends, forming one or more ammonia jets that mix with the flue gas flowing through the area, preparing for subsequent chemical reactions on the catalyst surface.

[0096] In some embodiments, the forced constant flow injection unit 1043 can be implemented in several ways: Optionally, the unit is a cylindrical nozzle with a single circular orifice on its sidewall. First, ammonia gas enters from one end of the nozzle. Second, under internal pressure, it is ejected from the static orifice on the sidewall. Finally, a jet is formed at a certain angle to the main pipe. Optionally, the unit is a structure with multiple micro-holes at the end (similar to a showerhead). First, ammonia gas fills the internal cavity of the nozzle. Second, ammonia gas is simultaneously and uniformly ejected from the multiple static micro-holes at the end. Finally, a conical jet with a wider coverage area is formed.

[0097] Please see Figure 5 This is a schematic diagram of the data visualization module of the flue gas denitrification adaptive ammonia injection device in the embodiments of this application.

[0098] Data visualization module 105 includes:

[0099] The performance status association mapping unit 1051 is used to synchronously compare and display the execution curve of the target ammonia injection allocation aligned with the time axis and the actual denitrification index curve in the wake evaluation feedback data on the visualization interface.

[0100] The visual interface refers to the graphical user interface (GUI) displayed on an industrial control computer or monitor. Time axis alignment means plotting data from different sources on a chart using a common time axis. The execution curve is the curve showing the change over time between the target value calculated by the AI ​​model and the actual values ​​fed back by the actuator (such as valve opening and actual flow rate). The actual denitrification index curve refers to the NO at the reactor outlet. x Curves showing the changes of key performance indicators such as concentration and ammonia slip concentration over time.

[0101] Specifically, the performance status correlation mapping unit reads from the real-time database the target ammonia injection allocation generated by the AI ​​calculation module, the actual flow of each branch fed back by the ammonia injection execution module, and the outlet NO collected by the wake performance feedback unit. x Concentration and ammonia slip concentration. In one or more trend chart controls, this unit plots this data in different colors or line styles, and all curves share the same horizontal axis—time. By observing these curves, operators can intuitively determine whether each adjustment action (change in target allocation) of the AI ​​is executed accurately (whether the execution curve follows), and whether these adjustments have brought about the expected results (whether the denitrification index curve is improving or remains stable).

[0102] The software elastic fine-tuning trigger unit 1052 activates the online learning mechanism of the model when the denitrification deviation falls into the preset safety adjustment fuzzy domain. Without changing the hardware configuration, it automatically corrects the target ammonia injection distribution amount by relying on the preset deep neural network model.

[0103] Among them, denitrification deviation refers to the difference between the actual denitrification index and the set target value. The preset safety adjustment fuzzy domain is a defined deviation range within which the flue gas denitrification adaptive ammonia injection device is considered to have a slight performance decrease, but still within the software's adjustable capabilities. Activating the model's online learning mechanism means sending a command to the AI ​​computing module to initiate a preset online optimization algorithm, using the latest data to make minor adjustments to the model parameters.

[0104] Specifically, this unit continuously monitors the denitrification deviation data provided by the wake performance feedback unit. It has clearly defined trigger conditions: when the real-time nitrogen oxide concentration after denitrification deviates continuously from the set value by a positive or negative deviation exceeding the safety tolerance (e.g., target 50 mg / m³), it triggers the denitrification process. 3 The measured value remained between 60-70 mg / m³ for 5 consecutive minutes. 3 When the ammonia escape concentration exceeds the warning threshold (e.g., 3 ppm) and the duration of the out-of-limit state exceeds the preset time window, the unit determines that the device's operating status has fallen into the safety adjustment fuzzy domain. At this time, it triggers a signal to activate the online learning function of the AI ​​computing module. The AI ​​model uses newly collected operating data and feedback data during this period to fine-tune its internal weights through algorithms such as backpropagation to better adapt to slight drifts in the current operating conditions. This allows it to attempt to bring the denitrification deviation back to the normal range by optimizing the ammonia injection strategy without replacing the hardware.

[0105] The hardware over-limit fuse alarm unit 1053, when the denitrification deviation exceeds the preset safety adjustment fuzzy domain and reaches the preset performance degradation threshold, triggers the fuse model online learning mechanism and forcibly issues a hardware reconfiguration warning.

[0106] Among these, the preset performance degradation threshold is a more stringent limit; once reached, it indicates that the device performance has severely deteriorated, and software adjustments are powerless to address it. The circuit breaker model online learning mechanism immediately stops and disables the online parameter update function of the AI ​​model to prevent erroneous learning under severely deviated operating conditions, which could lead to model performance collapse. The hardware reconfiguration warning is a clearly defined, highest-level alarm requiring manual intervention.

[0107] Specifically, the monitoring conditions of this unit are more stringent than those of the software fine-tuning trigger unit. The triggering conditions include: after the software fine-tuning mechanism is activated, if the mean absolute error (MAE) between the model's predicted ammonia injection rate and the actual ammonia injection rate continuously exceeds the preset generalization tolerance (e.g., continuously greater than 5%), it indicates that the model can no longer understand the current operating conditions. Alternatively, if the transient rate of change in unit load or the fluctuation amplitude of the unit inlet operating parameters in the wake assessment feedback data exceeds the design response extreme value of the denitrification unit (e.g., load change rate exceeding 20% / min), it indicates a fundamental and structural change in the operating conditions. Once either condition is met, the unit determines that the unit has reached the performance degradation threshold. It immediately performs two actions: first, it sends a circuit breaker signal to the AI ​​calculation module to prevent it from learning online; second, it generates a prominent and non-negligible hardware reconfiguration warning pop-up or audible and visual alarm on the data visualization module, along with clear instructions, such as recommending a re-detection of the inlet NOx concentration field and replacement of the fixed-gap unit if the unit's performance has severely degraded.

[0108] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0109] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0110] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A flue gas denitrification adaptive ammonia injection device, characterized in that, include: The data acquisition module is used to acquire the operating status data of the denitrification reactor, the wake assessment feedback data, and the regional concentration statistical average value used to characterize the spatial distribution differences in the flue gas duct. The AI ​​calculation module is used to input the average concentration of the region as a static spatial feature and the operating status data as a dynamic temporal feature into a preset deep neural network model for multi-source feature fusion calculation to obtain the target ammonia injection allocation corresponding to each fixed gap unit. The fixed gap unit is assembled on the straight grid body and has no mechanical adjustment components with relative sliding inside. The straight grid body is horizontally arranged in the inlet flue of the denitrification reactor along the flue gas flow direction. The control and adjustment module is used to receive the target ammonia injection distribution amount, convert the target ammonia injection distribution amount into a frequency conversion control signal for the total ammonia supply fluid pressure, and a zone flow adjustment control signal for each independent flow adjustment branch of the multi-channel flow distributor. The multi-channel flow distributor is configured with multiple independent flow adjustment branches, and each independent flow adjustment branch is connected to the fixed gap unit through an ammonia injection pipe made of corrosion-resistant material. The ammonia injection execution module is used to adjust the total ammonia supply fluid pressure based on the frequency conversion control signal, and drive the multi-channel flow distributor according to the zone flow control signal to guide the ammonia gas to each of the fixed gap units for injection according to the target ammonia injection distribution amount. The data visualization module is used to associate and display the execution status of each target ammonia injection allocation with the wake evaluation feedback data. When it is determined that the wake evaluation feedback data produces a denitrification deviation that exceeds the preset performance degradation threshold, a hardware reconstruction warning is generated. The hardware reconstruction warning is used to instruct the reacquisition of the regional concentration statistical average value and the replacement of the pre-installed fixed gap unit.

2. The flue gas denitrification adaptive ammonia injection device according to claim 1, characterized in that, The data acquisition module specifically includes: The spatial concentration difference pre-detection unit is used to arrange a sensor array in the grid installation area in the inlet flue of the denitrification reactor, continuously collect regional nitrogen oxide concentration distribution data within a preset period, and calculate the statistical average value of the regional concentration corresponding to each detection point after performing outlier removal preprocessing on the regional nitrogen oxide concentration distribution data. The real-time operating condition acquisition unit is used to synchronously read the flue gas flow rate data, flue gas temperature data, flue gas pressure data, and the actual ammonia injection amount of the current denitrification unit through the multi-dimensional sensor group, and integrate them to form the operating status data input to the preset deep neural network model; The wake performance feedback unit is used to monitor the status of the downstream outlet flue gas after the catalyst reaction, obtain the real-time nitrogen oxide concentration deviation value and ammonia slip concentration value after denitrification, and extract the transient change rate of unit load and the fluctuation range of the inlet operating parameters of the device, which are combined to form the wake evaluation feedback data.

3. The flue gas denitrification adaptive ammonia injection device according to claim 2, characterized in that, The matching of the fixed-gap unit with the statistical average concentration of the region is achieved through a preset hardware optimization rule, specifically including: A concentration gradient division unit is used to pre-establish multiple concentration gradient intervals from low to high, and to assign a cylindrical detachable structure with a corresponding standard aperture size to each concentration gradient interval as an alternative fixed gap unit. The static mapping pairing unit is used to compare the statistical average concentration of the region corresponding to each detection point with the concentration gradient interval one by one, determine the target gradient interval to which it belongs, and extract the standard pore size bound to the target gradient interval. The hardware pre-assembly guidance unit is used to output an assembly diagram containing the coordinate positions of each detection point and their corresponding standard aperture size, so as to indicate the offline screw-in installation of the fixed gap unit in the grid installation area.

4. The flue gas denitrification adaptive ammonia injection device according to claim 1, characterized in that, The preset deep neural network model specifically includes: The multi-scale sequence decomposition unit is used to perform multi-scale decomposition of the nitrogen oxide concentration temporal features in the operating status data using wavelet transform, so as to obtain a high-frequency fluctuation feature vector characterizing the flue gas abrupt change point. The time-dependent capture unit is used to input the high-frequency fluctuation feature vector and the static spatial features into the gated loop unit branch built into the model, and calculate the evolution correlation features between device load and flue gas parameters through the long short-term memory mechanism. The autoregressive multi-step prediction unit is used to perform rolling iterative calculations based on the evolutionary correlation features and the residual jump connection structure set by the model to prevent the gradient vanishing of deep networks, and output the target ammonia injection allocation amount corresponding to each fixed gap unit at the next control time.

5. The flue gas denitrification adaptive ammonia injection device according to claim 1, characterized in that, The control and adjustment module specifically includes: The global pressure supply calculation unit is used to mathematically sum the target ammonia injection allocation corresponding to all independent flow regulation branches to obtain the total target ammonia injection amount, and to map the total target ammonia injection amount into the frequency conversion control signal according to the fluid dynamics conversion function; An independent channel proportional analysis unit is used to calculate the dynamic distribution ratio of the target ammonia injection amount of each of the fixed gap units in the total target ammonia injection amount, and convert the dynamic distribution ratio into the zonal flow regulation control signal for controlling the opening degree of the corresponding branch electric valve.

6. The flue gas denitrification adaptive ammonia injection device according to claim 1, characterized in that, The AI ​​computing module also includes: The multi-source dimension elimination unit is used to call the Z-score normalization algorithm to normalize the running state data with different physical dimension properties, thereby eliminating the gradient shift caused by the extreme value differences between heterogeneous data in the multi-source feature fusion calculation process.

7. The flue gas denitrification adaptive ammonia injection device according to claim 5, characterized in that, The ammonia injection execution module specifically includes: A global variable frequency pressurization unit is used to receive the variable frequency control signal and regulate the global pressure of the ammonia supply fluid from the ammonia storage tank through a variable frequency ammonia pump to meet the dynamic pressure requirements of the main pipe corresponding to the total target ammonia injection volume. The flow guiding and limiting unit is used to receive the partition flow regulation control signal, adjust the fluid cross-sectional area of ​​each independent flow regulation branch inside the multi-path flow distributor, and split the pressurized ammonia gas and inject it into the corresponding fixed gap unit. The forced constant flow injection unit is used to force the input ammonia gas to be injected through a physically preset static aperture inside the fixed gap unit.

8. The flue gas denitrification adaptive ammonia injection device according to claim 1, characterized in that, The data visualization module specifically includes: The performance status association mapping unit is used to synchronously compare and display the execution curve of the target ammonia injection allocation amount aligned with the time axis and the actual denitrification index curve in the wake evaluation feedback data on the visualization interface. The software elastic fine-tuning trigger unit activates the online learning mechanism of the model when the denitrification deviation falls into the preset safety adjustment fuzzy domain. Without changing the hardware configuration, it automatically corrects the target ammonia injection distribution amount by relying on the preset deep neural network model. The hardware over-limit fuse alarm unit, when the denitrification deviation exceeds the preset safety adjustment fuzzy domain and reaches the preset performance degradation threshold, fuses the online learning mechanism of the model and forcibly triggers the issuance of the hardware reconstruction warning.

9. The flue gas denitrification adaptive ammonia injection device according to claim 8, characterized in that, The triggering logic for the data visualization module is as follows: When it is determined that the positive or negative deviation of the real-time nitrogen oxide concentration after denitrification continuously deviates from the set value by more than the safety tolerance, or the ammonia escape concentration value exceeds the warning line, and the duration of the over-limit state exceeds the preset time window, it is determined that it falls into the preset safety adjustment fuzzy domain and triggers the software elastic fine-tuning trigger unit. When the average absolute error between the predicted ammonia injection amount and the actual ammonia injection amount of the online learning mechanism of the model continues to exceed the preset generalization tolerance, or when it is determined that the transient change rate of unit load and the fluctuation amplitude of the inlet operating parameters in the wake assessment feedback data exceed the design response extreme value of the denitrification device, it is determined that the preset performance degradation threshold has been reached, and the hardware over-limit fuse alarm unit is triggered.

10. The flue gas denitrification adaptive ammonia injection device according to claim 1, characterized in that, The linear grid body has a plurality of mounting holes evenly distributed for assembling the fixed gap unit. The fixed gap unit is a cylindrical detachable structure and is screwed into the corresponding mounting hole through an external wall thread connection. Each fixed gap unit has a physically preset static aperture on its side wall and is fitted with a fluororubber seal at the mating connection with the mounting hole to prevent ammonia leakage.

Images