Circulating fluidized bed air distribution optimized combustion method and system

Abstract

This invention relates to the field of fluidized bed combustion technology, and more particularly to a method and system for optimizing combustion in a circulating fluidized bed. This application involves real-time acquisition of bed pressure differential and temperature signals from multiple monitoring points above the air distribution plate. A multi-dimensional fluidization state feature vector is extracted for each monitoring point. Based on the feature vectors and spatial locations of each monitoring point, all monitoring points are dynamically clustered into one or more virtual zones. For each virtual zone, a prediction model is established using its feature vector, and a target airflow is calculated, thereby adjusting the airflow distribution of each virtual zone. This application transforms fixed physical zones into dynamic virtual zones, enabling real-time adaptation to changes in fuel type, load, and bed material characteristics. This alleviates problems of poor local fluidization or excessive airflow when deviating from design conditions, thereby improving airflow uniformity and combustion efficiency.

Description

Technical Field

[0001] This invention relates to the field of fluidized bed combustion technology, and in particular to a circulating fluidized bed air distribution optimization combustion method and system. Background Technology

[0002] A circulating fluidized bed boiler is a clean coal combustion device that uses gas-solid fluidized bed combustion technology. The air distribution plate is the core component of a circulating fluidized bed boiler, and its air distribution uniformity directly determines the fluidization quality of the bed, combustion efficiency, and operational safety.

[0003] In the operation of circulating fluidized bed boilers, the fluidization quality of the air distribution plate bed surface directly affects combustion efficiency, bed temperature uniformity, and operational safety. To achieve uniform fluidization, existing technologies have proposed a static zone-optimized air distribution scheme. Based on the general laws of primary air inlet direction and material particle distribution in the furnace, the air distribution plate bed surface is pre-divided into multiple fixed functional zones. Each zone is equipped with air caps with different resistance characteristics and pitches. By changing the orifice diameter of the air caps, the resistance of each zone's air caps is fixed to match the fluidizing air volume required by different zones, aiming to achieve simultaneous fluidization of the entire bed.

[0004] However, the static zoned air distribution scheme in the existing technology is based on experience design for common operating conditions. Once the air cap resistance is installed, it is fixed and cannot adapt to the dynamic changes in fuel type, particle size, moisture, boiler load and bed material characteristics during boiler operation in real time. This leads to poor fluidization or excessive air volume in local areas when deviating from the design operating conditions, which affects combustion efficiency and increases the risk of coking or wear. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a circulating fluidized bed air distribution optimization combustion method and system to alleviate the technical problem in the prior art that it cannot adapt in real time to the dynamic changes of fuel type, particle size, moisture, boiler load and bed material characteristics during boiler operation, resulting in poor fluidization or excessive air volume in local areas when deviating from the design conditions, thereby affecting combustion efficiency and increasing the risk of coking or wear.

[0006] In a first aspect, embodiments of the present invention provide a method for optimizing combustion in a circulating fluidized bed, specifically including the following steps: Real-time acquisition of bed pressure differential and temperature signals from multiple monitoring points above the air distribution plate.

[0007] For each monitoring point, feature extraction is performed on the bed pressure difference signal and temperature signal of the monitoring point to obtain the multidimensional fluidization state feature vector of the monitoring point.

[0008] Based on the multidimensional fluidized state feature vector and spatial location of each monitoring point, all monitoring points are clustered into one or more virtual partitions.

[0009] For each virtual partition, a prediction model is established by combining the multidimensional fluidization state feature vector of the monitoring point within the virtual partition, and the target air volume of the virtual partition is calculated by combining the preset rules.

[0010] Adjust the airflow allocation corresponding to each virtual partition according to the target airflow of each virtual partition.

[0011] In conjunction with the first aspect, the present invention provides a possible implementation of the first aspect, comprising the step of extracting features from the bed pressure difference signal and temperature signal at the monitoring point to obtain a multidimensional fluidization state feature vector of the monitoring point, including: The collected bed pressure difference signal is subjected to time-domain feature extraction to extract time-domain fluctuation features that characterize the local fluidization state.

[0012] The temperature deviation characteristics are calculated based on the difference between the temperature signal at the monitoring point and the average temperature of the bed.

[0013] The time-domain fluctuation characteristics and the temperature deviation characteristics are combined to form a multidimensional fluidization state feature vector.

[0014] In conjunction with the first aspect, the present invention provides one possible implementation of the first aspect, wherein the time-domain fluctuation characteristics include the standard deviation within a sliding time window calculated based on the bed pressure difference signal, and the dominant frequency of the spectrum extracted from the bed pressure difference signal by fast Fourier transform.

[0015] The step of extracting time-domain features from the acquired bed pressure difference signal to extract time-domain fluctuation features characterizing the local fluidization state includes: The standard deviation is calculated by performing a standard deviation calculation on the collected bed pressure difference signal within a sliding time window.

[0016] The collected bed pressure difference signal is subjected to a fast Fourier transform, and the frequency with the largest amplitude in the spectrum is extracted as the dominant frequency of the spectrum.

[0017] In conjunction with the first aspect, the present invention provides a possible implementation of the first aspect, comprising the step of clustering all the monitoring points into one or more virtual partitions based on the multidimensional fluidization state feature vectors and spatial locations of each monitoring point, including: For each monitoring point, the spatial distance between the monitoring point and each adjacent monitoring point is calculated, as well as the feature distance between the multidimensional fluidization state feature vector of the monitoring point and the multidimensional fluidization state feature vector of each adjacent monitoring point.

[0018] By combining the spatial distance and feature distance of each monitoring point, a density-based clustering algorithm is used to cluster all the monitoring points into one or more virtual partitions.

[0019] In conjunction with the first aspect, the present invention provides one possible implementation of the first aspect, comprising the steps of establishing a prediction model for each virtual partition based on the multidimensional fluidization state feature vector of the monitoring points within the virtual partition and calculating the target airflow of the virtual partition based on preset rules, including: For each virtual partition, the mean of the multidimensional fluidized state feature vectors of all monitoring points within the virtual partition is used as the partition state vector.

[0020] Based on the partition state vector, a simplified dynamic model of the partition is established.

[0021] The model predictive control algorithm is used to solve for the optimal airflow control value of the virtual partition based on the simplified dynamic model of the partition, and the optimal airflow control value is used as the target airflow of the virtual partition.

[0022] In conjunction with the first aspect, the present invention provides a possible implementation of the first aspect, wherein the step of employing a model predictive control algorithm to solve for the optimal airflow control quantity of the virtual zone based on the simplified dynamic model of the zone includes: An optimization objective function is constructed, which includes a deviation penalty term between the partition state vector of each virtual partition and the target state vector, and an incremental penalty term for the air volume control quantity.

[0023] Each virtual partition and each feature dimension uses an independently adjustable weight matrix.

[0024] In conjunction with the first aspect, the present invention provides one possible implementation of the first aspect, wherein the target state vector is dynamically set by the upper-level optimizer based on the boiler load and / or pollutant emission indicators.

[0025] In conjunction with the first aspect, the present invention provides one possible implementation of the first aspect, wherein the method performs closed-loop control in a rolling optimization manner: At the next sampling time, the bed pressure difference signal and temperature signal of each monitoring point are collected again, the division of the virtual partition is updated, and the air volume control quantity is re-solved and executed.

[0026] Secondly, this application provides a circulating fluidized bed air distribution optimized combustion system, specifically including: a detection mechanism and a controller.

[0027] The monitoring mechanism has multiple monitoring points above the air distribution plate, and the multiple monitoring points are arranged in a grid pattern. Each monitoring point includes a pressure monitoring element and a temperature monitoring element.

[0028] The controller has a feature extraction and fusion calculation module, a zoning decision module, an air volume setting module, an air cap execution array, and a bed fluidization state feedback calculation module.

[0029] The feature extraction and fusion calculation module is used to process the collected bed pressure difference signal to extract time-domain fluctuation features and temperature deviation features, and to combine the time-domain fluctuation features and temperature deviation features of each monitoring point to form a multidimensional fluidization state feature vector.

[0030] The time-domain fluctuation characteristics include the standard deviation within the sliding time window calculated based on the bed pressure difference signal and the dominant frequency of the spectrum extracted from the bed pressure difference signal by fast Fourier transform.

[0031] The partitioning decision module is used to combine the spatial coordinates of each monitoring point and the multidimensional fluidized state feature vector, and use a density-based clustering algorithm to dynamically cluster all the monitoring points to generate one or more virtual partitions.

[0032] The airflow setting module is used to establish a simplified dynamic model for each virtual partition and to use a model predictive control algorithm to solve for the optimal airflow control value for each virtual partition.

[0033] The wind cap execution array is used to receive the optimal airflow control value and adjust the airflow allocation of each of the virtual partitions.

[0034] The bed fluidization state feedback calculation module is used to feed back the updated bed pressure difference signal and temperature signal from the monitoring point to the feature extraction and fusion calculation module, so as to trigger the zoning decision module and the air volume setting module to perform rolling optimization calculation and realize closed-loop rolling optimization.

[0035] Furthermore, the wind cap actuation array includes a primary air branch pipe and an electrically adjustable valve.

[0036] There are multiple primary air branch pipes and multiple electric regulating valves. Each of the multiple electric regulating valves is installed on a corresponding primary air branch pipe and located on the air inlet side of the air cap. The control terminal of the electric regulating valve is electrically connected to the controller.

[0037] The embodiments of the present invention bring the following beneficial effects: The circulating fluidized bed air distribution optimization combustion method provided by this invention collects bed pressure difference and temperature signals from multiple monitoring points above the air distribution plate in real time. Based on the extracted multidimensional fluidization state feature vector and its spatial location, the monitoring points are dynamically clustered into virtual zones. Then, a prediction model is established for each virtual zone, and the target air volume is calculated for adjustment, realizing the transformation from fixed physical zones to dynamic virtual zones. When the fuel type, particle size, moisture content, boiler load, or bed material characteristics change, the method provided by this invention can sense the fluidization state shift of each local area in real time and adaptively adjust the division of virtual zones and the corresponding air volume distribution. This effectively alleviates the problem of poor local fluidization or excessive air volume that occurs when the existing static zoning scheme deviates from the design conditions, thereby reducing fluidization dead zones, preventing local coking, improving bed temperature uniformity, and reducing primary air fan power consumption while ensuring combustion efficiency.

[0038] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0039] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0040] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a flowchart illustrating Embodiment 1 of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] To facilitate understanding of this embodiment, the technical terms used in this application will be briefly introduced below.

[0044] Fluidized bed air distributor plate: The core component in a circulating fluidized bed boiler used to support the bed material and evenly distribute the primary air. Multiple air caps are installed on the air distributor plate. The primary air enters the furnace through the small holes of the air caps, so that the fuel particles and bed material are in a fluidized state.

[0045] Virtual partitions: Unlike fixed physical partitions pre-defined on the air distribution panel, virtual partitions are regions to be adjusted formed by dynamic clustering of fluidization state data collected in real time from each monitoring point using an algorithm. Their boundaries are adaptively adjusted as the operating conditions change.

[0046] Fluidization quality index: A quantitative indicator that comprehensively reflects the fluidization state of the bed in a local area. It is calculated by fusing multi-dimensional information such as the fluctuation characteristics of the bed pressure difference signal and temperature deviation. It is used to determine whether there are abnormal conditions such as poor fluidization or local high temperature in the area.

[0047] Model predictive control: an advanced control algorithm based on rolling optimization of a system dynamic model. It predicts the system state at multiple future moments within each control cycle and obtains the optimal control quantity at the current moment by solving the optimization problem.

[0048] After introducing the technical terms used in this application, the application scenarios and design concepts of the embodiments of this application will be briefly described below.

[0049] During the operation of a circulating fluidized bed boiler, factors such as changes in fuel type, load fluctuations, and bed material wear can all lead to poor or uneven fluidization in localized areas of the air distribution plate, thereby affecting combustion efficiency and even causing coking or accelerated wear. Existing static zoned air distribution schemes are based on empirical design, and the air cap resistance is fixed after installation, making it impossible to adapt to the aforementioned dynamic changes in real time. Therefore, this application provides a circulating fluidized bed air distribution optimization combustion method and system.

[0050] Example 1 Please combine Figure 1 The circulating fluidized bed air distribution optimization combustion method provided in this embodiment specifically includes the following steps: S110 collects bed pressure differential and temperature signals from multiple monitoring points above the air distribution plate in real time.

[0051] S120: For each monitoring point, feature extraction is performed on the bed pressure difference signal and temperature signal of the monitoring point to obtain the multidimensional fluidization state feature vector of the monitoring point.

[0052] S130, based on the multidimensional fluidized state feature vectors and spatial locations of each monitoring point, clusters all monitoring points into one or more virtual partitions.

[0053] S140: For each virtual zone, a prediction model is established by combining the multidimensional fluidization state feature vector of the monitoring points within the virtual zone, and the target air volume of the virtual zone is calculated by combining the preset rules.

[0054] S150 adjusts the airflow distribution of each virtual zone according to the target airflow of each virtual zone.

[0055] In this embodiment, during the operation of the circulating fluidized bed boiler, multiple monitoring points arranged in a grid above the air distribution plate can collect the bed pressure difference signal and temperature signal at each monitoring point in real time. Specifically, each monitoring point is equipped with a pressure sensor and a thermocouple, and the sampling frequency is not less than 10Hz, so as to obtain the bed pressure difference signal Δ at the i-th monitoring point at time t in real time. (t) and temperature signal (t). Among them, the pressure sensor is used to monitor the differential pressure between the upper and lower parts of the air distribution plate to reflect the resistance of the material bed, and the thermocouple is used to monitor the temperature distribution of the bed in real time.

[0056] The monitoring points can also be equipped with non-contact particle concentration probes or velocity probes to obtain richer fluidization state information.

[0057] After acquiring the raw signals, feature extraction is performed on the bed pressure difference and temperature signals of each monitoring point. The raw signals are then transformed into quantitative indicators that can characterize the local fluidization state, resulting in a multidimensional fluidization state feature vector for that monitoring point, which provides a data foundation for subsequent dynamic partitioning.

[0058] After feature extraction, based on the multidimensional fluidized bed state feature vectors and spatial locations of each monitoring point, an algorithm is used to dynamically cluster all monitoring points into one or more virtual partitions in real time, replacing fixed physical partitions. Unlike the pre-divided fixed functional areas in existing technologies, the circulating fluidized bed air distribution optimization combustion method provided in this embodiment enables the partitions to adaptively adjust with changes in actual operating conditions through dynamic clustering of the algorithm.

[0059] After virtual zones are formed, a prediction model is established for each virtual zone by combining the multidimensional fluidization state feature vector of the monitoring points within the virtual zone. The prediction model is used to predict the future state, making the control decision forward-looking. At the same time, the target air volume of the virtual zone is calculated by combining the preset rules.

[0060] Finally, the airflow distribution for each virtual zone is adjusted according to the target airflow for each virtual zone. In this embodiment, the target airflow is achieved by controlling the electric regulating valve or adjustable vent cap on the primary air branch pipe.

[0061] The circulating fluidized bed air distribution optimization combustion method provided in this embodiment can sense changes in the bed fluidization state in real time and dynamically compensate for fluidization unevenness caused by changes in fuel type, particle size fluctuations, moisture changes, boiler load adjustments, and changes in bed material characteristics. For combustion of difficult-to-burn fuels or rapid load changes, the circulating fluidized bed air distribution optimization combustion method provided in this embodiment can maintain the bed in a near-optimal fluidization state, reduce fluidization dead zones, prevent local coking, and improve bed temperature uniformity, thereby improving combustion efficiency. At the same time, the system can automatically reduce the air supply to over-fluidized areas and accurately distribute the air volume to areas that need enhanced fluidization. Overall, it may achieve the same or even better fluidization effect with a smaller total primary air volume, which helps to reduce the power consumption of the primary air fan.

[0062] In this embodiment, step S120 includes: S121, extract time-domain features from the collected bed pressure difference signal to extract time-domain fluctuation features that characterize the local fluidization state.

[0063] S122, calculate the temperature deviation characteristics based on the difference between the temperature signal at the monitoring point and the average temperature of the bed.

[0064] S123 combines time-domain fluctuation characteristics and temperature deviation characteristics to form a multidimensional fluidization state feature vector.

[0065] Specifically, in this embodiment, the collected bed pressure difference signal is subjected to time-domain feature extraction to extract time-domain fluctuation characteristics characterizing the local fluidization state. The time-domain fluctuation characteristics include the standard deviation of the pressure difference signal within the sliding time window, which reflects the intensity of bubble activity. Bubble activity is an important mechanism for material mixing and heat transfer in a fluidized bed. A small standard deviation indicates weak bubble activity in that region, potentially indicating poor fluidization, while a large standard deviation may indicate airflow short-circuiting or channeling.

[0066] Next, based on the difference between the temperature signal at this monitoring point and the average bed temperature, the temperature deviation characteristic is calculated. In this embodiment, the temperature deviation characteristic is calculated as follows: the temperature at the current monitoring point... The difference between the temperature and the arithmetic mean of all monitoring points is calculated, i.e.

[0067] Where N represents the total number of monitoring points. A positive temperature deviation indicates localized high temperatures in the area, posing a risk of coking. A negative deviation may indicate material accumulation or insufficient fluidization.

[0068] Then, the aforementioned time-domain fluctuation characteristics and temperature deviation characteristics are combined to form a multidimensional fluidization state feature vector. In this embodiment, the form of this feature vector is:

[0069] This feature vector integrates the dynamic changes in differential pressure and the static distribution of temperature, reflecting the local fluidization state from two different dimensions, thus providing a more comprehensive basis for subsequent clustering and partitioning. By using the above feature extraction method, the original sensor signal is transformed into a standardized and comparable feature vector, which facilitates the accurate execution of subsequent clustering algorithms, enabling timely perception and response to subtle changes in the fluidization state.

[0070] In this embodiment, the time-domain fluctuation characteristics include the standard deviation within the sliding time window calculated based on the bed pressure difference signal, and the dominant frequency of the spectrum extracted from the bed pressure difference signal by fast Fourier transform.

[0071] Step S121 includes: S1211 calculates the standard deviation of the collected bed pressure difference signal within the sliding time window.

[0072] S1212 performs a fast Fourier transform on the acquired bed pressure difference signal and extracts the frequency with the largest amplitude in the spectrum as the dominant frequency of the spectrum.

[0073] Specifically, in this embodiment, the process of calculating the standard deviation of the acquired bed pressure difference signal within the sliding time window is as follows: Let the length of the sliding time window be τ, and calculate the standard deviation of the bed pressure difference signal Δ within the window. Standard deviation of (t) Its expression is:

[0074] in, This represents the average pressure difference signal within the window. The magnitude of this standard deviation reflects the intensity of bubble formation and movement near the monitoring point.

[0075] The acquired bed pressure difference signal was subjected to a Fast Fourier Transform, and the frequency with the largest amplitude in the spectrum was extracted as the dominant frequency of the spectrum. The specific process (t) is as follows: a fast Fourier transform is performed on the time-domain sequence of the pressure difference signal to obtain the spectral distribution, and the frequency component with the largest amplitude is identified from the spectrum. This dominant frequency is related to the characteristic frequency of bubble generation and movement. The frequency characteristics of bubbles differ under different fluidization states. During normal fluidization, the dominant frequency is usually concentrated in a certain range, while during poor fluidization or channeling, the dominant frequency will shift.

[0076] Standard deviation and dominant frequency of the spectrum characterize the dynamic features of the differential pressure signal from the time domain and frequency domain, respectively. The two complement each other and can more comprehensively reflect the changes in the local fluidization state, providing richer discrimination criteria for subsequent clustering and partitioning.

[0077] In this embodiment, step S130 includes: S131, For each monitoring point, calculate the spatial distance between the monitoring point and each adjacent monitoring point, and the feature distance between the multidimensional fluidization state feature vector of the monitoring point and the multidimensional fluidization state feature vector of each adjacent monitoring point.

[0078] S132 combines the spatial distance and feature distance of each monitoring point, and uses a density-based clustering algorithm to cluster all monitoring points into one or more virtual partitions.

[0079] Spatial distance reflects the physical proximity between monitoring points, while characteristic distance reflects the similarity of fluidization states. Specifically, in this embodiment, the spatial coordinates of the monitoring points are introduced. =( , Define the weighted distance metric between any two monitoring points i and j as:

[0080] in and These are the spatial coordinates of the two monitoring points. and Let be the multidimensional fluidized state feature vectors of the two monitoring points, and α and β be the weighting coefficients satisfying α + β = 1. By adjusting the weighting coefficients, a balance can be achieved between spatial continuity and feature similarity.

[0081] Then, combining the spatial distance and feature distance of each monitoring point, a density-based clustering algorithm is used to cluster all monitoring points into one or more virtual partitions. This clustering algorithm can automatically form partitions based on the density of data distribution without needing to pre-specify the number of partitions, ensuring that the monitoring points within each virtual partition are both spatially adjacent and similar in fluidization state. When the fluidization state characteristics of several adjacent areas deviate from the set value and show a consistent trend, the algorithm dynamically clusters them into a "virtual partition to be adjusted," enabling the system to accurately locate the specific area that needs adjustment, avoiding the limitations of fixed partitions in existing technologies that cannot adapt to changes in operating conditions.

[0082] In this embodiment, step S140 includes: S141, for each virtual partition, the mean of the multidimensional fluidized state feature vectors of all monitoring points within the virtual partition is used as the partition state vector.

[0083] S142, Based on the partition state vector, establish a simplified dynamic model of partition.

[0084] S143 employs a model predictive control algorithm, based on a simplified dynamic model of the zoning, to solve for the optimal airflow control quantity of the virtual zone, and uses the optimal airflow control quantity as the target airflow of the virtual zone.

[0085] In this application, let virtual partition k contain m monitoring points, then the partition state vector The arithmetic mean of the feature vectors of these m monitoring points is used to represent the overall fluidization state of the partition.

[0086] Based on the partition state vector, a simplified dynamic model of the partition is established. In this embodiment, the model takes the following form:

[0087] in It is the state transition matrix (diagonal matrix). For the input matrix, For air volume control, Measurable disturbances include global variables such as changes in total primary air volume and coal feed rate, which have a measurable impact on the fluidization state of each zone. Model parameters can be initially identified through an offline system and updated online using recursive least squares during operation, enabling the model to adapt to system characteristic drift caused by equipment wear and changes in operating conditions.

[0088] Finally, a model predictive control algorithm is employed, based on a simplified dynamic model of the virtual partition, to solve for the optimal airflow control value for that virtual partition. This optimal airflow control value is then used as the target airflow for that virtual partition. Through this method, a targeted dynamic model is established for each virtual partition, enabling control decisions to consider the differences and coupling between partitions, thus achieving precise airflow control for each partition.

[0089] Step S143, employing a model predictive control algorithm based on a simplified dynamic model of the zoning, solves for the optimal airflow control quantity for the virtual zoning, specifically including: S1431, construct the optimization objective function, which includes a deviation penalty term between the partition state vector of each virtual partition and the target state vector, as well as an incremental penalty term for the air volume control quantity.

[0090] Each virtual partition and each feature dimension uses an independently adjustable weight matrix.

[0091] In this embodiment, when using the model predictive control algorithm to solve for the optimal airflow control quantity of the virtual partition, in each control cycle, starting from the current time t, the future airflow is predicted. The state evolution of the step (predicting the time domain) is calculated. The optimal control sequence for the step (control time domain).

[0092] The optimization objective function in this embodiment includes two penalty terms. The first term is a penalty term for the deviation between the partition state vector of each virtual partition and the target state vector, which is used to drive the fluidization state of each partition closer to its target value. The second term is a penalty term for the increment of the airflow control quantity, which is used to suppress frequent or large fluctuations in the control quantity and avoid excessive action of the actuator.

[0093] Specifically, in this embodiment, the optimization objective function takes the following form:

[0094] in, Let be the target state vector of the k-th partition; This represents the weighted Euclidean norm. It is a positive definite diagonal weight matrix, used to balance the importance of different feature dimensions; To control the increment, This is the incremental weighting coefficient.

[0095] In this embodiment, each virtual partition and each feature dimension employs an independently adjustable weight matrix, allowing different partitions to be assigned different weights based on their importance within the boiler. Furthermore, different weight ratios can be allocated among different feature dimensions, thus flexibly balancing various control requirements. Through the aforementioned optimized objective function, the control algorithm can pursue optimal fluidization while also ensuring the smoothness of actuator regulation, thereby extending the actuator's lifespan and reducing system energy consumption.

[0096] In this embodiment, the target state vector is dynamically set by the upper-level optimizer based on the boiler load and pollutant emission indicators.

[0097] In this application, the upper-layer optimizer constitutes a hierarchical control architecture. The bottom layer is a model prediction controller, responsible for stabilizing the streaming state of each virtual partition. The upper layer is a multi-objective optimizer, which dynamically adjusts the target setpoints of each partition based on global variables. .

[0098] The focus of optimization objectives differs under different operating conditions. At low loads, stable fluidization is the primary objective, and the target state vectors for each zone are set to ensure uniform fluidization of the entire bed of material and avoid dead zones. At high loads, while ensuring stable fluidization, the airflow distribution can be optimized to actively shape the gas-solid flow and mixing patterns within the furnace, consciously enhancing airflow penetration in the central area of ​​the furnace to alleviate central oxygen deficiency.

[0099] Pollutant emission targets are also one of the bases for target setting. When NOx emissions are too high, the upper-level optimizer can adjust the target state vector of the relevant zones, and by optimizing the primary air distribution method, it provides a more flexible means to cooperate with low-NOx combustion technologies such as secondary air regulation or fuel staging. Through the above-mentioned hierarchical architecture and dynamic target setting method, this method, based on fluidization stability, takes into account both economy and environmental protection while ensuring the fluidization quality of the lower layer, and achieves multi-objective synergistic optimization.

[0100] It should be noted that in this embodiment, closed-loop control is performed using a rolling optimization method. After step S150 is completed, at the next sampling time, steps S110 to S150 are executed again to re-acquire the bed pressure difference signal and temperature signal of each monitoring point, update the virtual partition division, and re-solve and execute the air volume control quantity.

[0101] In this embodiment, the workflow is as follows: During the current control cycle, the bed pressure difference signal and temperature signal of each monitoring point are collected, a multi-dimensional fluidization state feature vector is extracted, virtual partitions are generated through dynamic clustering, a prediction model is established, and the optimal airflow control quantity is solved. The control quantity is then output to the actuator. When the next sampling time arrives, a new round of signals from each monitoring point is collected again, and the clustering and partitioning are performed again based on the updated fluidization state. The new airflow control quantity is then solved and executed again, and this process is repeated continuously.

[0102] This rolling optimization closed-loop control method possesses a certain degree of self-diagnosis and self-recovery capabilities. When sensors detect abnormal local parameters, such as a sharp increase in pressure differential in a certain area indicating potential slagging, the system can automatically identify that area in the next sampling cycle and increase the airflow in that area for purging intervention, which helps prevent the escalation of abnormal operating conditions. Compared with existing technologies that rely on human experience and manual intervention, the closed-loop rolling optimization method reduces the intensity of manual monitoring and improves the system's automation level and responsiveness. Furthermore, each optimization calculation is based on the latest measured data, enabling control decisions to continuously adapt to the time-varying characteristics of equipment and operating conditions.

[0103] Example 2 The circulating fluidized bed air distribution optimization combustion system provided in this embodiment includes a monitoring mechanism and a controller. The monitoring mechanism has multiple monitoring points arranged in a grid above the air distribution plate. Each monitoring point includes a pressure monitoring element and a temperature monitoring element. The controller has a feature extraction and fusion calculation module, a zone decision module, an airflow setting module, an air cap execution array, and a bed fluidization state feedback calculation module.

[0104] The system comprises several modules: a feature extraction and fusion calculation module, which processes the acquired bed pressure difference signal to extract time-domain fluctuation and temperature deviation features, and combines these features at each monitoring point to form a multidimensional fluidization state feature vector. The time-domain fluctuation features include the standard deviation within the sliding time window calculated from the bed pressure difference signal and the dominant frequency extracted from the signal via fast Fourier transform. A zoning decision module combines the spatial coordinates of each monitoring point with the multidimensional fluidization state feature vector, using a density-based clustering algorithm to dynamically cluster all monitoring points and generate one or more virtual zones. An airflow setting module establishes a simplified dynamic model for each virtual zone and uses a model predictive control algorithm to solve for the optimal airflow control value for each virtual zone. An airflow cap execution array receives the optimal airflow control value and adjusts the airflow distribution for each virtual zone. A bed fluidization state feedback calculation module feeds back the updated bed pressure difference and temperature signals from the monitoring points to the feature extraction and fusion calculation module, triggering the zoning decision module and the airflow setting module to perform rolling optimization calculations, achieving closed-loop rolling optimization.

[0105] In this embodiment, the monitoring mechanism forms multiple monitoring points above the air distribution plate, arranged in a grid pattern. Each monitoring point includes a pressure monitoring element and a temperature monitoring element. Specifically, the pressure monitoring element in this embodiment is a bed differential pressure sensor, used to monitor the differential pressure above and below the air distribution plate to reflect the material bed resistance. The temperature monitoring element is specifically a thermocouple, used to monitor the bed temperature distribution in real time.

[0106] As an feasible approach, monitoring points can also be equipped with non-contact particle concentration probes or velocity probes to obtain richer fluidization state information.

[0107] The controller has a feature extraction and fusion calculation module, a zone decision module, an air volume setting module, an air cap execution array, and a bed fluidization state feedback calculation module. These modules work together to form a complete intelligent air distribution control closed loop.

[0108] The feature extraction and fusion calculation module works as follows: it receives bed pressure difference signals and temperature signals collected by the monitoring agency, processes the collected bed pressure difference signals, extracts time-domain fluctuation features and temperature deviation features, and combines the time-domain fluctuation features and temperature deviation features of each monitoring point to form a multi-dimensional fluidized state feature vector. The time-domain fluctuation features include the standard deviation within the sliding time window calculated based on the bed pressure difference signal and the dominant frequency of the spectrum extracted by fast Fourier transform.

[0109] The zoning decision module works as follows: it receives the multidimensional fluidized state feature vector output by the feature extraction and fusion calculation module, combines it with the spatial coordinates of each monitoring point, and uses a density-based clustering algorithm to dynamically cluster all monitoring points, generating one or more virtual zones. This module does not change the physical structure of the air distribution plate, but dynamically defines the areas that need to be adjusted through the algorithm, so that the zoning can adaptively adjust with changes in actual operating conditions.

[0110] The working principle of the air volume setting module is as follows: it receives the virtual partition information output by the partition decision module, establishes a simplified dynamic model for each virtual partition, and uses the model predictive control algorithm to solve for the optimal air volume control quantity of each virtual partition. The optimal air volume control quantity is the target air volume of the corresponding virtual partition.

[0111] The working principle of the wind cap execution array is as follows: it receives the optimal air volume control output from the air volume setting module and adjusts the air volume distribution of each virtual zone.

[0112] The working principle of the bed fluidization state feedback calculation module is as follows: the updated bed pressure difference signal and temperature signal from the monitoring points are fed back to the feature extraction and fusion calculation module to trigger the zoning decision module and the air volume setting module to perform a new round of rolling optimization calculation, thereby realizing closed-loop rolling optimization.

[0113] In this embodiment, a complete closed loop is achieved, from state perception to intelligent analysis and then to execution control. The perception layer monitors the bed state in real time through a multi-parameter sensor network; the analysis layer realizes virtual partitioning decisions through feature extraction and dynamic clustering; the control layer achieves precise airflow allocation through model predictive control and wind cap execution array; and the feedback layer achieves continuous self-adjustment through rolling optimization. This enables the system to perceive and dynamically compensate for fluidization unevenness caused by fuel and load changes in real time, reduce fluidization dead zones, prevent local coking, and improve bed temperature uniformity.

[0114] In this embodiment, the wind cap execution array includes primary air branch pipes and electric regulating valves; there are multiple primary air branch pipes and multiple electric regulating valves, and the multiple electric regulating valves are respectively installed on multiple primary air branch pipes and located on the air inlet side of the wind cap. The control terminal of the electric regulating valve is electrically connected to the controller.

[0115] The primary air branch pipes connect the main primary air pipe to each air cap, with each branch pipe serving a group of air caps. Electric regulating valves are installed on the primary air branch pipes, located at the air inlet side of the air caps, between the air caps and the primary air source. By adjusting the opening of the electric regulating valves, the primary air volume entering the corresponding group of air caps can be precisely controlled.

[0116] The control terminal of the electric regulating valve is electrically connected to the controller. The controller converts the optimal air volume control quantity of each virtual zone calculated by the air volume setting module into the corresponding electric regulating valve opening command, and sends it to the actuator motor of each electric regulating valve to drive the valve plate to rotate to the target opening position, thereby changing the air volume of the branch pipe.

[0117] As an feasible approach, the wind cap actuator array can also adopt a design with an adjustable stopcock structure on the wind cap head. That is, each wind cap has its own adjustable stopcock mechanism. By changing the opening of the small hole of the wind cap, the outlet resistance can be adjusted, thereby changing the local fluidization air volume of the wind cap. This can further shorten the adjustment response time and achieve higher precision local air volume control.

[0118] In this structure, the wind cap execution array can independently and precisely adjust the airflow of each virtual zone, thereby achieving the effect of adjusting the airflow distribution of each virtual zone according to the target airflow of each virtual zone, thus completing a complete control closed loop from perception and decision-making to execution.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0120] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0121] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing combustion in a circulating fluidized bed, characterized in that, Includes the following steps: Real-time acquisition of bed pressure differential and temperature signals from multiple monitoring points above the air distribution plate; For each monitoring point, feature extraction is performed on the bed pressure difference signal and the temperature signal at the monitoring point to obtain a multidimensional fluidization state feature vector of the monitoring point; Based on the multidimensional fluidized state feature vector and spatial location of each monitoring point, all monitoring points are clustered into one or more virtual partitions. For each virtual partition, a prediction model is established by combining the multidimensional fluidization state feature vector of the monitoring points within the virtual partition, and the target air volume of the virtual partition is calculated by combining the preset rules. Adjust the airflow allocation corresponding to each virtual partition according to the target airflow of each virtual partition.

2. The circulating fluidized bed air distribution optimization combustion method according to claim 1, characterized in that, The step of extracting features from the bed pressure difference signal and temperature signal at the monitoring point to obtain the multidimensional fluidization state feature vector of the monitoring point includes: The collected bed pressure difference signal is subjected to time-domain feature extraction to extract time-domain fluctuation features characterizing the local fluidization state; Based on the difference between the temperature signal at the monitoring point and the average temperature of the bed, the temperature deviation characteristics are calculated. The time-domain fluctuation characteristics and the temperature deviation characteristics are combined to form a multidimensional fluidization state feature vector.

3. The circulating fluidized bed air distribution optimization combustion method according to claim 2, characterized in that, The time-domain fluctuation characteristics include the standard deviation within the sliding time window calculated based on the bed pressure difference signal, and the dominant frequency of the spectrum extracted from the bed pressure difference signal by fast Fourier transform; The step of extracting time-domain features from the acquired bed pressure difference signal to extract time-domain fluctuation features characterizing the local fluidization state includes: The standard deviation is calculated by performing a standard deviation calculation on the collected bed pressure difference signal within a sliding time window to obtain the standard deviation; The collected bed pressure difference signal is subjected to a fast Fourier transform, and the frequency with the largest amplitude in the spectrum is extracted as the dominant frequency of the spectrum.

4. The circulating fluidized bed air distribution optimization combustion method according to claim 2, characterized in that, The step of clustering all the monitoring points into one or more virtual partitions based on the multidimensional fluidized state feature vectors and spatial locations of each monitoring point includes: For each monitoring point, calculate the spatial distance between the monitoring point and each adjacent monitoring point, and the feature distance between the multidimensional fluidization state feature vector of the monitoring point and the multidimensional fluidization state feature vector of each adjacent monitoring point; By combining the spatial distance and feature distance of each monitoring point, a density-based clustering algorithm is used to cluster all the monitoring points into one or more virtual partitions.

5. The circulating fluidized bed air distribution optimization combustion method according to claim 1, characterized in that, For each virtual partition, the steps of establishing a prediction model based on the multidimensional fluidization state feature vector of the monitoring points within the virtual partition and calculating the target airflow of the virtual partition according to preset rules include: For each virtual partition, the mean of the multidimensional fluidized state feature vectors of all monitoring points within the virtual partition is used as the partition state vector; Based on the partition state vector, a simplified dynamic model of the partition is established; The model predictive control algorithm is used to solve for the optimal airflow control value of the virtual partition based on the simplified dynamic model of the partition, and the optimal airflow control value is used as the target airflow of the virtual partition.

6. The circulating fluidized bed air distribution optimization combustion method according to claim 5, characterized in that, The step of employing a model predictive control algorithm to solve for the optimal airflow control quantity of the virtual zone based on the simplified dynamic model of the zone includes: Construct an optimization objective function, which includes a deviation penalty term between the partition state vector of each virtual partition and the target state vector, and an incremental penalty term for the air volume control quantity; Each virtual partition and each feature dimension uses an independently adjustable weight matrix.

7. The circulating fluidized bed air distribution optimization combustion method according to claim 6, characterized in that, The target state vector is dynamically set by the upper-level optimizer based on boiler load and / or pollutant emission indicators.

8. The circulating fluidized bed air distribution optimization combustion method according to claim 1, characterized in that, The method performs closed-loop control using a rolling optimization approach. At the next sampling time, the bed pressure difference signal and temperature signal of each monitoring point are collected again, the division of the virtual partition is updated, and the air volume control quantity is re-solved and executed.

9. A circulating fluidized bed air distribution optimized combustion system, characterized in that, include: The monitoring mechanism has multiple monitoring points above the air distribution plate, and the multiple monitoring points are arranged in a grid pattern. Each monitoring point includes a pressure monitoring element and a temperature monitoring element. The controller has a feature extraction and fusion calculation module, a zoning decision module, an air volume setting module, an air cap execution array, and a bed fluidization state feedback calculation module; The feature extraction and fusion calculation module is used to process the collected bed pressure difference signal to extract time-domain fluctuation features and temperature deviation features, and to combine the time-domain fluctuation features and temperature deviation features of each monitoring point to form a multi-dimensional fluidized state feature vector. The time-domain fluctuation characteristics include the standard deviation within the sliding time window calculated based on the bed pressure difference signal and the dominant frequency of the spectrum extracted from the bed pressure difference signal by fast Fourier transform; The partitioning decision module is used to combine the spatial coordinates of each monitoring point and the multidimensional fluidized state feature vector, and use a density-based clustering algorithm to dynamically cluster all the monitoring points to generate one or more virtual partitions. The airflow setting module is used to establish a simplified dynamic model for each virtual partition and to use a model predictive control algorithm to solve for the optimal airflow control quantity for each virtual partition. The wind cap execution array is used to receive the optimal airflow control amount and adjust the airflow distribution of each of the virtual partitions; The bed fluidization state feedback calculation module is used to feed back the updated bed pressure difference signal and temperature signal from the monitoring point to the feature extraction and fusion calculation module, so as to trigger the zoning decision module and the air volume setting module to perform rolling optimization calculations and realize closed-loop rolling optimization.

10. The circulating fluidized bed air distribution optimized combustion system according to claim 9, characterized in that, The wind cap actuation array includes a primary air branch pipe and an electric regulating valve; There are multiple primary air branch pipes and multiple electric regulating valves. Each of the multiple electric regulating valves is installed on a corresponding primary air branch pipe and located on the air inlet side of the air cap. The control terminal of the electric regulating valve is electrically connected to the controller.

Images