An automated boiler operation control system and method

Abstract

This invention discloses an automated boiler operation control system and method, belonging to the field of industrial automation control. It includes: acquiring temperature sensor readings and edge video streams to construct a three-dimensional mesh spatial model of the furnace; calculating the spatiotemporal temperature gradient based on steady-state temperature data to generate a node reference entropy generation rate matrix; acquiring real-time temperature mappings to generate a node real-time entropy generation rate array; performing Gaussian mixture fitting on the reference matrix to generate a dynamic triggering baseline, and then extracting abnormal voxel node data; broadcasting the abnormal data to adjacent candidate nodes and constructing a Boolean constraint equation system; solving the Boolean constraint equation system in conjunction with the current valve opening parameters, and converting the positive solution vector into an incremental control level output to the actuator. This invention employs a decentralized strategy based on spatial neighborhood negotiation and local thermodynamic features, which can effectively overcome the hysteresis defects of traditional global control and improve the combustion stability and feedback response accuracy of the boiler under complex and variable operating conditions.

Description

Technical Field

[0001] This invention relates to the field of industrial automation control, and in particular to an automated boiler operation control system and method. Background Technology

[0002] As a core piece of equipment in industrial production and energy conversion, the level of automation in the operation and control of boilers directly affects the safety, stability, and economy of the system. In complex industrial automation scenarios, the boiler combustion process is affected by multiple factors such as load fluctuations, fuel characteristics, and physical disturbances. Therefore, it is necessary to use an automated control system to monitor and dynamically adjust the internal thermal operating status of the boiler in real time to maintain stable system operation.

[0003] In related technologies, Chinese invention patent application CN120428789A discloses a digital twin-based steam temperature optimization control system and method for coal-fired boilers. This method collects real-time operating data of the coal-fired boiler and inputs the operating data into a three-dimensional structural diagram of the coal-fired boiler to mark the high-temperature area inside the boiler. It obtains the geometric center coordinates of the high-temperature area and the position coordinates of the desuperheater nozzle in the three-dimensional structural diagram, calculates the adjustment angle and test radius of the desuperheater nozzle, and finally calculates the cooling water flow rate required for the high-temperature area. It simulates the cooling process in the three-dimensional structural diagram and adjusts the cooling water flow rate according to the steam temperature after cooling.

[0004] The aforementioned technologies primarily rely on a centralized architecture to locate the geometric center of local high-temperature areas and simulate unidirectional physical cooling. When faced with the complex and ever-changing dynamic conditions inside the boiler, this centralized command issuance method based on macroscopic temperature thresholds and geometric coordinates exhibits significant control response lag. Furthermore, this adjustment mechanism lacks adaptive scheduling capabilities for the overall dynamic balance of the system, making it prone to neglecting certain aspects when encountering drastic fluctuations in operating conditions. This results in low overall accuracy of automated control and poor operational stability. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an automated boiler operation control system and method. It employs a decentralized strategy based on spatial neighborhood negotiation and local thermodynamic characteristics, which effectively overcomes the lag defects of traditional global control and improves the combustion stability and feedback response accuracy of the boiler under complex and variable operating conditions.

[0006] The above objectives can be achieved through the following approach: An automated boiler operation control system includes a furnace voxel mesh construction module for acquiring time-series readings from temperature sensors and edge video streams in the boiler furnace, performing 3D point cloud fusion and voxelization spatial discretization operations to construct a 3D furnace mesh spatial model; a baseline entropy generation rate generation module for extracting time-series temperature data of mesh nodes from the steady-state operation phase using the furnace 3D mesh spatial model, calculating the spatiotemporal temperature gradient matrix for adjacent voxel nodes, and generating a node baseline entropy generation rate matrix; a real-time entropy generation rate mapping module for acquiring real-time temperature readings, calculating the real-time temperature gradient matrix between adjacent voxel nodes, and mapping to generate a node real-time entropy generation rate array; and an abnormal voxel node detection module. The system performs Gaussian mixture distribution fitting on the node baseline entropy generation rate matrix to generate a dynamic trigger baseline, and extracts abnormal voxel node data that are higher than the dynamic trigger baseline from the node real-time entropy generation rate array. A Boolean constraint equation construction module sends the abnormal voxel node data as a communication broadcast packet to spatially adjacent candidate voxel nodes, extracts the real-time entropy generation rate array of the candidate voxel nodes, and constructs a Boolean constraint equation set. An incremental control level solution module uses the candidate voxel nodes to extract the current fuel valve opening parameter, solves the Boolean constraint equation set to obtain a positive solution vector, and converts the positive solution vector into an incremental control level output to the fuel valve actuator.

[0007] Optionally, the furnace voxel mesh construction module includes: a multimodal thermal point cloud generation unit, used to extract a two-dimensional flame contour image using the edge video stream, reconstruct the two-dimensional flame contour image into a three-dimensional spatial geometric coordinate set, and map and anchor the time-series readings of the temperature sensor to the three-dimensional spatial geometric coordinate set to generate a multimodal thermal point cloud set; and a voxel mesh spatial modeling unit, used to extract a three-dimensional spatial bounding box using the multimodal thermal point cloud set, perform spatial segmentation calculations on the three-dimensional spatial bounding box to generate three-dimensional voxel units, and assemble the three-dimensional voxel units to construct a three-dimensional mesh spatial model of the furnace.

[0008] Optionally, the baseline entropy generation rate generation module includes: a spatiotemporal gradient extraction unit, used to extract historical grid node time-series temperature data from the three-dimensional grid spatial model of the furnace, and perform first-order derivative calculation of the sliding window to generate a temperature change rate sequence; a steady-state interval truncation unit, used to perform kernel density estimation on the temperature change rate sequence, extract the change rate envelope of the probability density peak, and truncate the continuous time segments in the temperature change rate sequence that fall into the change rate envelope as the steady-state operation stage; and a thermodynamic dissipation mapping unit, used to perform partial derivative calculation on the temperature data of the steady-state operation stage along the spatial coordinate axis to generate a spatiotemporal temperature gradient matrix, and map it to generate a node baseline entropy generation rate matrix.

[0009] Optionally, the real-time entropy generation rate mapping module includes: a reading noise reduction and smoothing unit, used to acquire the real-time temperature measurement raw signal using adjacent voxel nodes, perform filtering processing on the real-time temperature measurement raw signal, extract convergent state variables to generate a net-state real-time temperature reading; and a quadratic dissipation mapping unit, used to extract the spatial coordinate difference using adjacent voxel nodes, calculate the real-time temperature gradient matrix in combination with the net-state real-time temperature reading, and output the real-time entropy generation rate array of the generating nodes.

[0010] Optionally, the abnormal voxel node detection module includes: a parameter iteration and benchmark construction unit, used to perform expectation maximization iteration on the node benchmark entropy generation rate matrix, extract the mean vector and covariance matrix of the Gaussian components, fuse the mean vector and covariance matrix to calculate the upper boundary envelope equation, and generate a dynamic triggering benchmark line; and an out-of-bounds stripping and extraction unit, used to project the node real-time entropy generation rate array onto the upper boundary envelope equation, perform element-wise algebraic difference calculation, extract the target array coordinates with a difference greater than zero, and extract abnormal voxel node data based on the target array coordinates.

[0011] Optionally, the system further includes: extracting the spatial geometric boundary of the three-dimensional bounding box, extracting the steady-state fluctuation extreme value of the rate of change envelope, performing physical property mapping and binding between the spatial geometric boundary and the steady-state fluctuation extreme value, and generating a spatial thermal safety tolerance band.

[0012] Optionally, the Boolean constraint equation construction module includes: a geometric addressing and broadcasting unit, used to extract the three-dimensional spatial coordinates of the abnormal voxel node data, perform coplanar adjacent boundary retrieval on the furnace three-dimensional mesh space model to mark candidate voxel nodes, and convert the abnormal voxel node data into communication broadcast packets to be sent to the candidate voxel nodes; and a tolerance boundary transformation unit, used to extract the real-time entropy generation rate array of the candidate voxel nodes, perform logical inequality transformation between the real-time entropy generation rate array and the spatial thermal safety tolerance band, extract the feasible adjustment domain boundary of the node, and assemble and generate a set of Boolean constraint equations.

[0013] Optionally, the incremental control level solving module includes: a constraint iterative optimization unit, used to extract the current fuel valve opening parameter based on the physical mapping address of the candidate voxel node, fuse the Boolean constraint equations with the space thermal safety tolerance band to construct an optimization penalty function, perform gradient descent calculation starting from the current fuel valve opening parameter, and solve to extract the positive solution vector; and a duty cycle mapping output unit, used to obtain the hardware drive gain coefficient of the fuel valve actuator, perform scalar multiplication mapping between the positive solution vector and the hardware drive gain coefficient to generate an incremental control level, and send the incremental control level to the fuel valve actuator.

[0014] Based on the same inventive concept, this invention also provides an automated boiler operation control method, the method comprising: acquiring time-series readings of temperature sensors in the boiler furnace and edge video streams; performing three-dimensional point cloud fusion and voxel spatial discretization operations to construct a three-dimensional mesh spatial model of the furnace; extracting time-series temperature data of mesh nodes from the steady-state operation stage using the three-dimensional mesh spatial model of the furnace; calculating the spatiotemporal temperature gradient matrix for adjacent voxel nodes to generate a node baseline entropy generation rate matrix; acquiring real-time temperature readings; calculating the real-time temperature gradient matrix between adjacent voxel nodes; and mapping and generating a node real-time entropy generation rate array; and for... The node baseline entropy generation rate matrix is ​​fitted with a Gaussian mixture distribution to generate a dynamic trigger baseline, and abnormal voxel node data above the dynamic trigger baseline are extracted from the node real-time entropy generation rate array. The abnormal voxel node data is sent as a communication broadcast packet to spatially adjacent candidate voxel nodes, and the real-time entropy generation rate array of the candidate voxel nodes is extracted to construct a Boolean constraint equation system. The current fuel valve opening parameter is extracted using the candidate voxel nodes, the Boolean constraint equation system is solved to obtain the positive solution vector, and the positive solution vector is converted into an incremental control level and output to the fuel valve actuator.

[0015] Compared with the prior art, the present invention has the following advantages: 1. Traditional boiler control relies heavily on a small number of macroscopic sensors, which easily leads to monitoring blind spots and control dead zones. By using multimodal point cloud fusion and spatial voxel discretization, the traditional "black box" global coarse adjustment is transformed into precise mapping at the voxel node level. This can accurately locate local thermal anomaly areas and perform targeted control, effectively avoiding energy waste and equipment thermal stress damage caused by global indiscriminate adjustment from the physical space dimension. 2. Traditional technologies rely heavily on pre-set empirical models, which are prone to failure when faced with poor-quality fuels or varying operating conditions. This new technology uses "local entropy generation rate" as the sole objective evaluation indicator and employs Gaussian distribution dynamic fitting to replace manually set fixed thresholds. This mechanism eliminates reliance on prior human experience, ensuring that the boiler can spontaneously evolve and converge to an optimal energy efficiency state with minimal global thermodynamic irreversible dissipation under any severe disturbance. 3. Traditional control systems employ centralized computation, which suffers from computational bottlenecks and the risk of single-point failures. This paper abandons a global central controller and achieves spontaneous game theory and local arbitration of local supply and demand among lower-level execution units by broadcasting abnormal data between adjacent voxel nodes and constructing a Boolean constraint equation system based on physical safety tolerance. This improves fault isolation capabilities, self-healing capabilities, and overall robustness under extreme operating conditions and local hardware failures.

[0016] 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 may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the 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 based on these drawings without creative effort.

[0018] Figure 1 This is a framework diagram of an automated boiler operation control system according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of an automated boiler operation control system according to an embodiment of the present invention.

[0020] Figure 3 This is a scatter plot of the physical dissipation characteristics of the node baseline entropy generation rate and the real-time entropy generation rate in an embodiment of the present invention.

[0021] Figure 4 This is a contour thermal distribution diagram of the thermal safety tolerance zone of the furnace cross-section in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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.

[0023] Reference Figure 1 One embodiment of the present invention proposes an automated boiler operation control system that adopts a decentralized strategy based on spatial neighborhood negotiation and local thermodynamic characteristics, which can effectively overcome the lag defects of traditional global control and improve the combustion stability and feedback response accuracy of the boiler under complex and variable operating conditions.

[0024] like Figure 2 As shown, the system in this embodiment specifically includes: The furnace voxel mesh construction module is used to acquire the time-series readings of temperature sensors and edge video streams in the boiler furnace, perform 3D point cloud fusion and voxel spatial discretization operations, and construct a 3D mesh spatial model of the furnace. Optionally, the furnace voxel mesh construction module includes: A multimodal thermal point cloud generation unit is used to extract a two-dimensional flame contour image using the edge video stream, reconstruct the two-dimensional flame contour image into a three-dimensional spatial geometric coordinate set, and map and anchor the time-series readings of the temperature sensor to the three-dimensional spatial geometric coordinate set to generate a multimodal thermal point cloud set. The system acquires edge video streams from a camera installed at the furnace observation port and time-series readings from temperature sensors collected by wall-mounted thermocouples. The edge video stream is a continuous sequence of images reflecting real-time combustion in the furnace. The time-series readings from the temperature sensors represent discrete temperature values ​​recorded on a continuous time scale. A multi-layer image feature extractor analyzes the edge video stream, identifies the flame boundary using an edge detection operator, and extracts a two-dimensional flame contour image. The two-dimensional flame contour image defines the flame's luminous area within a two-dimensional pixel plane. A multimodal thermal point cloud generation unit, based on the principle of binocular visual parallax, performs depth calculations on the two-dimensional flame contour image, reconstructing it into a three-dimensional spatial geometric coordinate set. The three-dimensional spatial geometric coordinate set represents the set of coordinate points on the flame surface in physical space. The reconstruction calculation logic of the three-dimensional spatial geometric coordinate set satisfies the following formula: , Among them, letters Represents the coordinates of a point in a three-dimensional geometric coordinate set, with the dimension of meters; letters This represents the camera's rated focal length, measured in pixels. The rated focal length is set based on pre-shipment calibration test data; the letter... The baseline physical distance representing the binocular camera, measured in meters, is set based on pre-shipment calibration test data; the letter... The letter represents the disparity value of a pixel in a two-dimensional flame outline image, with the dimension being pixels; The dimensionless transformation coefficient matrix represents the conversion from pixel coordinates to physical coordinates. This matrix is ​​set based on pre-shipment calibration test data. The multimodal thermal point cloud generation unit extracts time-series readings from the temperature sensor and maps and anchors them to a three-dimensional spatial geometric coordinate set. Mapping and anchoring refers to using the IDW (Inverse Distance Weighted Wave) algorithm to assign discrete temperature values ​​to corresponding spatial geometric points, generating a multimodal thermal point cloud. A multimodal thermal point cloud is a feature point array that simultaneously carries spatial location coordinates and temperature attribute information. The calculation logic for temperature mapping and anchoring satisfies the following formula: , Among them, letters The temperature value represents the target point of the multimodal thermal point cloud concentration, with dimensions in Kelvin; the letter represents the temperature value of the target point. Represents the time-series readings of the temperature sensor, with the dimension of Kelvin; letters Represents the dimensionless weighting coefficients calculated based on spatial Euclidean distance; letters This represents the total number of sensors involved in the calculation and is dimensionless. The dimensionless weighting coefficients are set based on the physical law of distance attenuation, meaning that the closer the spatial distance, the greater the weight score, and the sum of all weighting coefficients involved in the calculation is a constant value of 1.

[0025] For example, the multimodal thermal point cloud generation unit receives the edge video stream. A multilayer image feature extractor extracts a two-dimensional flame contour image. During the reconstruction of the three-dimensional spatial geometric coordinate set, the calibration focal length is determined based on pre-shipment calibration test data. for Pixels, baseline physical distance for Meters. The multimodal thermal point cloud generation unit detected the disparity value of a certain pixel. for Pixels, combined with a dimensionless transformation coefficient matrix The depth coordinates of this point in the three-dimensional geometric coordinate set are calculated to be... Meters. The multimodal thermal point cloud generation unit acquires time-series readings from temperature sensors and reads the temperature value of a certain sensor at the target time. for Kelvin. Dimensionless weighting coefficients were calculated based on the physical laws of distance attenuation. for By performing temperature mapping anchoring calculations, the temperature values ​​of corresponding points in the multimodal thermal point cloud are determined. for Kelvin.

[0026] The voxel mesh spatial modeling unit is used to extract a three-dimensional spatial bounding box using the multimodal thermal point cloud, perform spatial segmentation calculations on the three-dimensional spatial bounding box to generate three-dimensional voxel units, and assemble the three-dimensional voxel units to construct a three-dimensional mesh spatial model of the furnace.

[0027] A 3D spatial bounding box is extracted using a multimodal thermal point cloud set. The 3D spatial bounding box is the smallest orthogonal hexahedron that can completely cover the point cloud data set. Voxel mesh spatial modeling units perform spatial partitioning calculations on the 3D spatial bounding box, that is, discretizing and cutting it along the three axes of the Cartesian coordinate system with a defined step size to generate 3D voxel elements. A 3D voxel element is the smallest geometric unit with an independent volume after spatial discretization. The logic of the spatial partitioning calculation satisfies the following formula: , Among them, letters Represents the total number of generated three-dimensional voxel units, dimensionless; letter With letters These represent the maximum and minimum coordinates of the 3D bounding box along the horizontal axis, respectively, with the unit of meter; the letters... With letters These represent the maximum and minimum coordinate values ​​along the vertical axis, respectively, with the unit of measurement being meters; the letters... With letters These represent the maximum and minimum coordinate values ​​along the vertical axis, respectively, with the unit of measurement being meters; the letters... ,letter and letters These represent the segmentation step size along the three coordinate axes, with units of meters. The segmentation step size is set based on historical fluid dynamics simulation test data to balance computational resolution and real-time processing overhead. The voxel mesh spatial modeling unit assembles three-dimensional voxel elements according to their original topological relationships to construct a three-dimensional mesh spatial model of the furnace. The three-dimensional mesh spatial model of the furnace refers to a digital model reflecting the distribution of physical fields inside the furnace.

[0028] The baseline entropy generation rate generation module is used to extract the temporal temperature data of the grid nodes from the steady-state operation stage using the three-dimensional grid space model of the furnace, calculate the spatiotemporal temperature gradient matrix for adjacent voxel nodes, and generate the node baseline entropy generation rate matrix. Optionally, the baseline entropy generation rate generation module includes: The spatiotemporal gradient extraction unit is used to extract historical grid node time-series temperature data from the three-dimensional grid spatial model of the furnace and perform first-order derivative calculation of the sliding window to generate a temperature change rate sequence. Historical grid node time-series temperature data is extracted from the three-dimensional mesh spatial model of the furnace. This data refers to the set of discrete temperature values ​​for each spatial grid node arranged according to time sequence within a historical tracking time window. The spatiotemporal gradient extraction unit performs a sliding window first-order derivative on the historical grid node time-series temperature data. The sliding window first-order derivative involves extracting a fixed-length data interval on the time axis, calculating the differential slope of the temperature over time within that interval, and iterating this calculation repeatedly while shifting forward along the time axis. This derivative operation generates a temperature change rate sequence. The temperature change rate sequence is a set of values ​​representing the rate of temperature increase or decrease at each spatial grid node at each moment. The calculation logic of the sliding window first-order derivative satisfies the following formula: , Among them, letters Represents the characteristic elements in the temperature change rate sequence, with dimensions in Kelvin per second; letters This represents the historical grid node time-series temperature data at the current end of the sliding window, measured in Kelvin; letters The historical grid node time-series temperature data represents the starting time of the sliding window, with the unit being Kelvin; the letter is... The time step represents the sliding window, measured in seconds. The value of the time step is not preset manually, but is determined directly based on the fixed sampling crystal oscillator period of the underlying hardware of the temperature sensor, to ensure that the derivative calculation extracts the real physical changes rather than interpolation noise.

[0029] For example, the spatiotemporal gradient extraction unit acquires historical time-series temperature data of a spatial grid node. Based on the fixed sampling crystal oscillator period of the temperature sensor hardware, the time step of the sliding window is determined as follows: Seconds. The spatiotemporal gradient extraction unit reads the temperature data at the current end of the sliding window. Kelvin, read the temperature data at the start of the sliding window. Kelvin. The spatiotemporal gradient extraction unit substitutes the above values ​​into the first-order derivative formula of the sliding window and performs subtraction to calculate the temperature difference value. Kelvin, then divide the temperature difference by the time step. Seconds, the characteristic elements in the temperature change rate sequence are calculated as follows: Kelvin per second.

[0030] The steady-state interval extraction unit is used to perform kernel density estimation on the temperature change rate sequence, extract the rate of change envelope of the probability density peak, and extract the continuous time segment of the temperature change rate sequence that falls into the rate of change envelope as the steady-state operation stage; The steady-state interval extraction unit performs kernel density estimation on the temperature change rate sequence. Kernel density estimation refers to using a smoothed mathematical kernel function to estimate the probability distribution curve of the temperature change rate occurring in different numerical intervals over a continuous time period. The steady-state interval extraction unit extracts the probability density peak value on this probability distribution curve and then extracts the rate of change envelope from it. The probability density peak value refers to the temperature change rate value with the highest frequency and probability density. The rate of change envelope refers to the upper and lower limit intervals formed by extending outwards from the probability density peak value. The calculation logic of the rate of change envelope satisfies the following formula: , Among them, letters With letters These represent the upper and lower boundaries of the rate of change envelope, respectively, with dimensions in Kelvin per second; the letters... The letter represents the peak probability density derived from the kernel density estimate, with dimensions in Kelvin per second; The dynamic standard deviation represents the time-series temperature data of historical grid nodes, measured in Kelvin per second. The value of the dynamic standard deviation is calculated by the root mean square deviation of all rate-of-change characteristic elements within the historical tracking time window, and is entirely adaptively driven by the operating data. The steady-state interval extraction unit determines the numerical trend of the temperature rate-of-change sequence, extracting continuous time segments that fall within the rate-of-change envelope as the steady-state operating phase. The steady-state operating phase refers to the period during which the local thermodynamic state of the furnace remains in dynamic equilibrium without drastic disturbances.

[0031] For example, the steady-state interval truncation unit performs kernel density estimation on the temperature change rate sequence, extracting the probability density peak as... Kelvin per second. The steady-state interval truncation unit extracts the dynamic standard deviation of the temperature data within that time period by calculating the root mean square deviation. Kelvin per second. The steady-state interval truncation element substitutes the peak value and dynamic standard deviation into the calculation formula, performs addition and subtraction operations, and calculates the upper boundary of the rate of change envelope as... Kelvin per second, lower bound is Kelvin per second. During the period from the tenth to the thirtieth second on the time axis, the numerical value of the temperature change rate sequence remains at... to The Kelvin fluctuates every second. The steady-state interval extraction unit determines that the continuous value falls completely within the upper and lower boundary intervals, and then accurately extracts this continuous time segment of twenty seconds as the steady-state operation stage.

[0032] The thermodynamic dissipation mapping unit is used to perform partial derivative calculations on the temperature data of the steady-state operation phase along the spatial coordinate axes to generate a spatiotemporal temperature gradient matrix, and to map and generate a node reference entropy generation rate matrix.

[0033] The thermodynamic dissipation mapping unit performs partial derivative calculations on the temperature data during steady-state operation along the spatial coordinate axes to generate a spatiotemporal temperature gradient matrix. The spatial coordinate axes encompass the three orthogonal directions of a three-dimensional Cartesian coordinate system. The spatiotemporal temperature gradient matrix is ​​a set of spatial derivative vectors reflecting the degree of non-uniformity in local temperature distribution. The thermodynamic dissipation mapping unit maps the spatiotemporal temperature gradient matrix to generate a nodal reference entropy generation rate matrix. The nodal reference entropy generation rate matrix is ​​a set of data characterizing the rate of irreversible energy dissipation caused by heat conduction in fluid micro-elements; a lower value indicates that local combustion and heat transfer are closer to an ideal reversible state. The calculation logic for generating the nodal reference entropy generation rate matrix satisfies the following formula: , Among them, letters The characteristic elements within the nodal baseline entropy generation rate matrix are represented by watts per cubic meter in Kelvin; letters The thermal conductivity of the fluid medium at the target grid location, measured in watts per meter Kelvin, is obtained by querying a physical property lookup table of the actual boiler fluid medium built into the control system; letters This represents the absolute average temperature of the grid node during the steady-state operation phase, with dimensions in Kelvin; the letter... ,letter With letters These represent the spatial partial derivatives of the spatiotemporal temperature gradient matrix in the directions corresponding to the horizontal, vertical, and longitudinal axes, respectively, with dimensions in Kelvin per meter.

[0034] For example, the thermodynamic dissipation mapping unit queries the built-in boiler fluid medium physical property lookup table to extract the thermal conductivity of the fluid medium at a certain grid node location. Watts per meter Kelvin. The thermodynamic dissipation mapping element extracts the absolute average temperature of this node during steady-state operation. Kelvin. The spatial partial derivatives of the spatiotemporal temperature gradient matrix in the horizontal, vertical, and axial directions are obtained through spatial partial derivative calculations. Kelvin per meter Kelvin per meter and Kelvin per meter. The thermodynamic dissipation mapping unit substitutes the above values ​​into the mapping formula, first calculating the sum of squares of the spatial partial derivatives in the three directions, i.e. Plus In addition The total gradient squared value is obtained. Then the thermal conductivity Divide by absolute average temperature The square of this value yields a ratio coefficient of 0.2 parts per million. Finally, this ratio coefficient is compared with the total gradient squared value. Performing a multiplication operation, the characteristic element within the node baseline entropy generation rate matrix is ​​calculated as follows: Watts per cubic meter of Kelvin.

[0035] The real-time entropy generation rate mapping module is used to collect real-time temperature readings, calculate the real-time temperature gradient matrix between adjacent voxel nodes, and map and generate a node real-time entropy generation rate array. Optionally, the real-time entropy generation rate mapping module includes: The reading noise reduction and smoothing unit is used to acquire the real-time temperature measurement raw signal using adjacent voxel nodes, perform filtering processing on the real-time temperature measurement raw signal, extract convergent state variables, and generate net-state real-time temperature readings. The reading noise reduction and smoothing unit acquires the raw real-time temperature measurement signal using adjacent voxel nodes. The raw real-time temperature measurement signal refers to the initial temperature data directly detected by the underlying hardware sensors at the current physical moment, unprocessed and mixed with high-frequency electromagnetic environmental noise. The reading noise reduction and smoothing unit performs filtering processing on the raw real-time temperature measurement signal. Filtering processing refers to a mathematical iterative process that uses a dynamic state estimation algorithm to weight and fuse the actual physical measurement value at the current moment with the theoretical prediction value based on the system state deduction at the previous moment, and corrects for errors. During the filtering iteration process, the reading noise reduction and smoothing unit extracts convergent state variables, thereby generating the net-state real-time temperature reading. The convergent state variable refers to the optimal temperature estimate corresponding to the error covariance matrix within the estimation algorithm approaching the minimum boundary. The net-state real-time temperature reading refers to the temperature data that accurately reflects the true thermodynamic state of the furnace space nodes after completely filtering out random high-frequency noise. The calculation logic for filtering processing and generating the net-state real-time temperature reading satisfies the following formula: , Among them, letters Represents the net-state real-time temperature reading, with dimensions in Kelvin; letter This represents the predicted temperature value derived from the state transition matrix of the previous sampling period, with the dimension Kelvin; the letter... Represents the raw, real-time temperature measurement signal collected, with the dimension of Kelvin; letters Represents the Kalman gain coefficient, which is dimensionless. The value of the Kalman gain coefficient is not subjectively set, but is dynamically determined by algebraic ratio calculation based on the measurement noise variance parameter specified in the temperature sensor hardware's manufacturer's technical specifications, combined with the controller's inherent process noise variance parameter. It is used to accurately allocate the trust weights between the measured value and the predicted value.

[0036] For example, the reading noise reduction and smoothing unit acquires the real-time raw temperature measurement signal using adjacent voxel nodes. In the current control cycle, the reading noise reduction and smoothing unit reads the real-time raw temperature measurement signal from the underlying sensor. for Kelvin. Simultaneously, the reading noise reduction and smoothing unit retrieves the predicted temperature value derived from the previous cycle. for Kelvin. The measurement noise variance is extracted based on the underlying temperature sensor hardware specifications and calculated against the inherent process noise variance. The reading noise reduction and smoothing unit then determines the Kalman gain coefficient for the current moment. for The reading noise reduction and smoothing unit then performs filtering calculations, first reducing the noise of the original real-time temperature measurement signal. Subtract the predicted temperature value The temperature deviation value was obtained. Kelvin. Next, the reading noise reduction and smoothing unit modifies this temperature deviation value with the Kalman gain coefficient. Perform scalar multiplication to calculate the correction compensation value. Kelvin. Finally, the reading noise reduction and smoothing unit will predict the temperature value. With correction compensation value Perform addition to extract convergent state variables and generate the net-state real-time temperature reading. Kelvin.

[0037] The quadratic dissipative mapping unit is used to extract the spatial coordinate difference from adjacent voxel nodes, calculate the real-time temperature gradient matrix by combining the net real-time temperature reading, and output the real-time entropy generation rate array of the generating nodes.

[0038] The quadratic dissipative mapping unit extracts spatial coordinate differences using adjacent voxel nodes. Spatial coordinate difference refers to the linear physical distance between two adjacent mesh nodes with physical topological connections in the three-dimensional mesh space model of the furnace, along three orthogonal directions in the Cartesian coordinate system. The quadratic dissipative mapping unit calculates the real-time temperature gradient matrix by combining the net real-time temperature readings. The real-time temperature gradient matrix is ​​the set of derivative vectors of the temperature field of the target node in three spatial directions at the current operating moment. The calculation logic of the horizontal component of the real-time temperature gradient matrix satisfies the following formula: , Among them, letters Represents the component of the real-time temperature gradient matrix along the horizontal axis, with dimensions in Kelvin per meter; letters Represents the net real-time temperature reading of adjacent voxel nodes along the positive horizontal axis, in Kelvin; letters This represents the net real-time temperature reading of the current target node, in Kelvin; the letter represents the temperature reading of the current target node itself. The horizontal axis represents the spatial coordinate difference between adjacent voxel nodes, measured in meters. The gradient components along the vertical and vertical axes utilize the same spatial partial derivative difference operation logic. Subsequently, the quadratic dissipation mapping unit outputs an array of real-time entropy generation rates for generating nodes. This array represents the set of energy dissipation rates of each grid node due to irreversible heat conduction at the current running moment; it is the core reference data driving subsequent adaptive control equations. The calculation logic for outputting the array of real-time entropy generation rates for generating nodes satisfies the following quadratic operator formula: , Among them, letters The feature element in the array representing the real-time entropy generation rate of nodes, with the unit being watts per cubic meter in Kelvin; letters The real-time thermal conductivity of the fluid medium at the node's location is expressed in watts per meter (Kelvin). This real-time thermal conductivity is obtained by interpolating and matching the pressure and temperature data of the current node using a built-in fluid medium property database. This represents the net real-time temperature reading of the current target node, in Kelvin; the letter represents the temperature reading of the current target node itself. ,letter With letters These represent the components of the real-time temperature gradient matrix along the horizontal, vertical, and axial directions, respectively, with dimensions in Kelvin per meter.

[0039] For example, the quadratic dissipative mapping unit extracts the spatial coordinate difference using adjacent voxel nodes to obtain the spatial coordinate difference in the horizontal axis direction. for Meters. The quadratic dissipative mapping unit acquires the net real-time temperature readings of adjacent voxel nodes, and reads the net real-time temperature readings of adjacent voxel nodes along the positive horizontal axis. for Kelvin, the current target node's own net real-time temperature reading. for Kelvin. The quadratic dissipative mapping unit calculates the algebraic difference between the two as... Kelvin, divided by the spatial coordinate difference Meters, the components of the real-time temperature gradient matrix in the horizontal direction are calculated. for Kelvin per meter. Using the same spatial partial derivative difference operation logic, the quadratic dissipative mapping unit calculates the component in the vertical direction. for Kelvin per meter, vertical component for Kelvin per meter. The quadratic dissipative mapping unit queries the fluid medium property characteristic library based on the current node state to determine the real-time thermal conductivity of the fluid. for Watts per meter Kelvin. The quadratic dissipative mapping unit performs calculations using the quadratic operator formula, first calculating the sum of squares of the gradient components in the three directions, obtaining the spatial gradient sum of squares value as follows: Subsequently, the real-time thermal conductivity will be... Divide by the current target node's net real-time temperature reading The square of this factor yields a conversion factor of 0.15 parts per million. Finally, this conversion factor is compared with the sum of squared spatial gradients. Performing a numerical multiplication operation, the quadratic dissipative mapping unit outputs the feature elements in the real-time entropy generation rate array of the generating nodes. for Watts per cubic meter of Kelvin.

[0040] An abnormal voxel node detection module is used to fit the node baseline entropy generation rate matrix to a Gaussian mixture distribution, generate a dynamic trigger baseline, and extract abnormal voxel node data that are higher than the dynamic trigger baseline from the node real-time entropy generation rate array. Optionally, the abnormal voxel node detection module includes: The parameter iteration and benchmark construction unit is used to perform expectation maximization iteration on the node benchmark entropy generation rate matrix, extract the mean vector and covariance matrix of the Gaussian components, fuse the mean vector and covariance matrix to calculate the upper boundary envelope equation, and generate a dynamic triggering benchmark line. The parameter iteration and benchmark building unit performs expectation-maximization iteration on the node baseline entropy generation rate matrix. The node baseline entropy generation rate matrix records the underlying irreversible thermodynamic dissipation baseline at each spatial location during steady-state operation. Expectation-maximization iteration refers to a purely mathematical iterative process that alternates between expectation and maximization steps to approximate the mixed model parameters underlying the data probability density. Through this iteration, the parameter iteration and benchmark building unit extracts the mean vector and covariance matrix of the Gaussian components. The mean vector represents the statistical clustering center of the local node entropy generation rate under steady-state conditions; the covariance matrix represents the divergence of node data in various spatial dimensions and the multivariate coupling relationship. The parameter iteration and benchmark building unit fuses the mean vector and covariance matrix to calculate the upper boundary envelope equation, thereby generating a dynamic triggering baseline. The dynamic triggering baseline refers to an anomaly detection surface that adapts to the physical location characteristics of each node on a three-dimensional spatial scale, rather than a globally uniform rigid threshold. The calculation logic of the dynamic triggering baseline satisfies the following formula: , Among them, letters Represents the dynamic trigger baseline, measured in watts per cubic meter (Kelvin); letters This represents the mean vector extracted through the expected value maximization iteration, with units of watts per cubic meter in Kelvin; the letter... Represents the covariance matrix; letters This represents the variance vector formed by performing the main diagonal element extraction operation on the covariance matrix, with dimensions of watt squared and meter Kelvin squared; the letter... The confidence multiplier coefficient is dimensionless. It is not arbitrarily assigned, but rather derived by reverse-engineering the statistical quantile corresponding to the highest tolerance index for false alarm rate in the industrial boiler safety operation specifications, thus ensuring the rigor and legality of the anomaly detection boundary.

[0041] For example, parameter iteration and benchmark building units perform expectation-maximization iteration on the node benchmark entropy generation rate matrix to extract the mean vector of a specific spatial Gaussian component. for Watts per cubic meter (Kelvin). Parameter iteration and benchmark building blocks extract the covariance matrix of this component and obtain the main diagonal elements to obtain the variance vector. for The standard deviation of the square root of the watt squared to the sixth power of the meter Kelvin squared is: Watts per cubic meter (Kelvin). Based on the false alarm rate tolerance specified in the industrial boiler safety operation specifications, the confidence multiplier coefficient obtained from the lookup table is extracted. for Substituting the above values ​​into the formula and performing numerical multiplication and addition operations, parameter iteration and baseline building element calculation generate the dynamic trigger baseline for this local mesh location. for Watts per cubic meter of Kelvin.

[0042] The out-of-bounds stripping and extraction unit is used to project the real-time entropy generation rate array of the nodes onto the upper boundary envelope equation, perform element-wise algebraic difference calculation, extract the target array coordinates with a difference greater than zero, and extract abnormal voxel node data based on the target array coordinates.

[0043] The boundary stripping and extraction unit projects the node real-time entropy generation rate array onto the upper boundary envelope equation. The node real-time entropy generation rate array reflects the transient physical dissipation of each grid node in the furnace at the current moment. The projection operation aligns the spatial index of the real-time monitoring data with the dynamic trigger baseline in the same three-dimensional grid coordinate system. After alignment, the boundary stripping and extraction unit performs element-wise algebraic difference calculation. Element-wise algebraic difference calculation refers to a scalar subtraction operation that subtracts the upper limit of the envelope of the corresponding coordinate point from the instantaneous dissipation value of each grid node. The calculation logic of the boundary stripping and extraction unit satisfies the following formula: , Among them, letters The decision difference represents the output of the element-wise algebraic difference calculation, measured in watts per cubic meter (Kelvin); the letter... The feature element in the array representing the real-time entropy generation rate of nodes, with the unit being watts per cubic meter in Kelvin; letters The dynamic trigger baseline represents the corresponding coordinate point, with the dimension watts per cubic meter (Kelvin). The boundary stripping and extraction unit determines the polarity of the difference, extracting the target array coordinates where the difference is greater than zero. The target array coordinates refer to the three-dimensional spatial index where the degree of thermodynamic irreversible dissipation exceeds the allowable upper limit. Based on the target array coordinates, the boundary stripping and extraction unit extracts anomalous voxel node data. Anomalous voxel node data refers to the complete three-dimensional coordinates and state slice information of local mesh nodes experiencing significant combustion deterioration or pipe wall heat transfer failure. For example... Figure 3As shown in the figure, the discrete points represent independent three-dimensional voxel nodes. When the real-time entropy generation rate of a voxel node breaks through the dynamic trigger baseline generated by the expectation maximization iteration and falls into the gray abnormal voxel boundary stripping area, the corresponding abnormal voxel node data can be accurately extracted based on the coordinate difference.

[0044] For example, the boundary stripping and extraction unit projects the real-time entropy generation rate array of nodes onto the upper boundary envelope equation. It reads the feature elements of the real-time entropy generation rate array of a given grid node. for For every watt per cubic meter of Kelvin, retrieve the dynamic trigger baseline corresponding to that coordinate point. for Watts per cubic meter Kelvin. The boundary stripping and extraction unit performs element-wise algebraic difference calculation, performs scalar subtraction, and obtains the judgment difference. for Watts per cubic meter Kelvin. Due to the difference in judgment. If the value is strictly greater than zero, the out-of-bounds stripping and extraction unit locks the grid point and extracts its target array coordinates. Based on the target array coordinates, it completely packages and extracts the abnormal voxel node data of the local grid for subsequent neighborhood adaptive broadcasting.

[0045] Optionally, the system further includes: Extract the spatial geometric boundary of the three-dimensional bounding box and the steady-state fluctuation extreme value of the rate of change envelope. Perform physical property mapping and binding on the spatial geometric boundary and the steady-state fluctuation extreme value to generate a spatial thermal safety tolerance band.

[0046] The spatial thermal tolerance generation module extracts the spatial geometric boundary of the 3D spatial bounding box. The spatial geometric boundary refers to the outermost absolute physical coordinates of the point cloud model enclosing the entire furnace fluid. Simultaneously, the module extracts the steady-state fluctuation extreme values ​​of the rate of change envelope. These extreme values ​​refer to the highest permissible upper limit of the absolute value of the temperature change rate intercepted throughout the entire furnace during historical fault-free steady-state operation. The module then performs physical attribute mapping and binding between the spatial geometric boundary and the steady-state fluctuation extreme values. This physical attribute mapping and binding involves scaling the spatial gradient of the uniform steady-state fluctuation extreme values ​​across the entire domain based on the physical attenuation deviation of each grid node from the furnace's geometric center. Through this physical attribute mapping and binding operation, the module generates a spatial thermal safety tolerance band. This band refers to the set of dynamic temperature fluctuation limits with differentiated stringency constraints implemented at different physical locations within the furnace. The calculation logic for the physical attribute mapping and binding satisfies the following formula: , Among them, letters Represents the characteristic extreme value within the space thermal safety tolerance zone, with dimensions in Kelvin per second; letters Represents the extracted steady-state fluctuation extreme value, with dimensions in Kelvin per second; letters This represents a dimensionless geometric adjustment factor calculated based on spatial geometric boundaries. The value of this factor is determined by the linear reciprocal relationship of the Euclidean distance between the grid node coordinates and the geometric center coordinates of the 3D bounding box. This aligns with the objective physical evolution of the thermal flow field, where intense combustion in the furnace center allows for high tolerances, while combustion in the perimeter wall region requires low tolerances. For example... Figure 4 As shown, the physical topological distribution of the extreme values ​​of safety tolerance decreases in a Euclidean distance gradient from the geometric center of the furnace to the edge tube wall, which confirms the digital spatial mapping logic of implementing differentiated constraint boundaries for different combustion zones.

[0047] For example, the spatial thermal tolerance generation module extracts the spatial geometric boundary of the three-dimensional bounding box and calculates and confirms the absolute geometric center of the furnace. For a grid node on the edge of the tube wall that is far from the center, the spatial thermal tolerance generation module calculates and determines the dimensionless geometric adjustment factor at that location based on its long-distance Euclidean spatial attenuation relationship. for The space thermal tolerance generation module extracts the extreme values ​​of steady-state fluctuations captured in the previous stage. for Kelvin per second. The space thermal tolerance generation module performs physical property mapping and binding between the dimensionless coefficients calculated from the space geometric boundary and the steady-state fluctuation extreme values, substitutes them into the formula for numerical multiplication, and generates the space thermal safety tolerance band at the edge-fitting mesh node location. for Kelvin per second is used as the extremely stringent physical operating limit for this sensitive area.

[0048] The Boolean constraint equation construction module is used to send the abnormal voxel node data as a communication broadcast packet to the spatially adjacent candidate voxel nodes, and extract the real-time entropy generation rate array of the candidate voxel nodes to construct a set of Boolean constraint equations. Optionally, the Boolean constraint equation construction module includes: The geometric addressing and broadcasting unit is used to extract the three-dimensional spatial coordinates of the abnormal voxel node data, perform coplanar adjacent boundary retrieval in the three-dimensional mesh space model of the furnace to mark candidate voxel nodes, and convert the abnormal voxel node data into a communication broadcast packet and send it to the candidate voxel node. The geometric addressing and broadcasting unit extracts the three-dimensional spatial coordinates of anomalous voxel node data. Anomalous voxel node data refers to the physical coordinates and state information of a specific mesh node identified as experiencing anomalous thermodynamic dissipation in the preceding detection steps. The three-dimensional spatial coordinates represent the absolute position values ​​of the anomalous node in the horizontal, vertical, and lateral directions within a three-dimensional Cartesian coordinate system. Based on this coordinate information, the geometric addressing and broadcasting unit performs a coplanar adjacent boundary search within the furnace's three-dimensional mesh space model. The furnace's three-dimensional mesh space model is a digital entity composed of all three-dimensional voxel elements. The coplanar adjacent boundary search is a spatial addressing process that uses the anomalous node as its physical center to find adjacent mesh nodes that directly share a two-dimensional square geometric surface. The geometric addressing and broadcasting unit marks candidate voxel nodes by calculating the spatial Manhattan distance. Candidate voxel nodes are surrounding mesh nodes that are physically adjacent to the anomalous node and possess the potential to participate in local thermodynamic collaborative regulation through heat conduction. The calculation logic of the spatial Manhattan distance satisfies the following formula: , Among them, letters Represents the Manhattan distance in space, measured in meters; letters ,letter With letters These represent the three-dimensional spatial coordinates of the target mesh node along the horizontal, vertical, and y-axis directions during the traversal search process, respectively, with the unit being meters; the letters... ,letter With letters These represent the three-dimensional spatial coordinates of the anomalous voxel node data along the horizontal, vertical, and triangular axes, respectively, with units of meters. When the spatial Manhattan distance is strictly equal to the spatial partitioning step size, the geometric addressing and broadcasting unit determines that the node is coplanar and adjacent to the anomalous node and marks it as a candidate voxel node. Subsequently, the geometric addressing and broadcasting unit converts the anomalous voxel node data into communication broadcast packets. A communication broadcast packet is a digital signal data frame encapsulated according to the industrial bus control protocol, carrying anomaly location codes and thermal dissipation deviation amplitudes. The geometric addressing and broadcasting unit sends the communication broadcast packets point-to-point to the marked candidate voxel nodes, thereby triggering a distributed collaborative adjustment mechanism without a central controller among the underlying physical nodes.

[0049] For example, the geometric addressing and broadcasting unit extracts the three-dimensional spatial coordinates of a certain anomalous voxel node data and reads the horizontal axis coordinates. for meters, vertical axis coordinates for meters, vertical axis coordinates for Meters. The geometric addressing and broadcasting unit performs coplanar adjacent boundary retrieval in the three-dimensional mesh space model of the furnace. During traversal calculation, the geometric addressing and broadcasting unit reads the three-dimensional spatial coordinates of a certain adjacent mesh node, its horizontal axis coordinate... for meters, vertical axis coordinates for meters, vertical axis coordinates for Meters. The geometric addressing and broadcasting unit substitutes the above coordinates into the distance formula to calculate the absolute difference of the horizontal axis coordinates, which is [value missing]. Meters, the absolute difference between the vertical and ordinate coordinates is 1. Meters. The spatial Manhattan distance is calculated by summing the measurements. for Meters. The spatial partitioning step size of the three-dimensional mesh model of the furnace is known to be set to... Since the calculated Manhattan distance is exactly equal to the segmentation step size, the geometric addressing and broadcasting unit officially marks the adjacent grid node as a candidate voxel node. Subsequently, the geometric addressing and broadcasting unit encapsulates and converts the coordinates and deviation values ​​of the abnormal node according to the protocol to generate a communication broadcast packet, and sends it to the candidate voxel node via the communication bus.

[0050] The tolerance boundary transformation unit is used to extract the real-time entropy generation rate array of the candidate voxel nodes, perform logical inequality transformation between the real-time entropy generation rate array and the space thermal safety tolerance zone, extract the feasible adjustment domain boundary of the node, and assemble and generate a Boolean constraint equation system.

[0051] The tolerance boundary transformation unit extracts the real-time entropy generation rate array of candidate voxel nodes. The real-time entropy generation rate array refers to the energy dissipation rate of the candidate voxel node due to irreversible heat conduction at the current physical moment. To assess the physical safety of the node's participation in coordinated regulation, the tolerance boundary transformation unit performs a logical inequality transformation between the real-time entropy generation rate array and the space thermal safety tolerance band. The space thermal safety tolerance band refers to the safety limit generated by the preceding module based on the spatial geometric boundary, defining the maximum allowable temperature fluctuation extreme value within a specific spatial region. The logical inequality transformation refers to the algebraic process of unifying parameters of different physical units to the same dissipation dimension by introducing dimensional mapping coefficients and establishing inequality extreme value constraints. The tolerance boundary transformation unit extracts the node's feasible regulation domain boundary through this transformation process. The node's feasible regulation domain boundary refers to the theoretical upper limit of the maximum additional entropy generation rate that a candidate voxel node can withstand without causing secondary thermodynamic anomalies such as its own wall overheating or local flow field collapse. The calculation logic of the node's feasible regulation domain boundary satisfies the following formula: , Among them, letters The representative node represents the feasible adjustment domain boundary, with the dimension of watts per cubic meter (Kelvin); the letter represents... The thermal dissipation mapping coefficient, measured in watt-seconds per cubic meter Kelvin squares, is determined based on the measured product of the isobaric specific heat capacity and fluid density parameters of the local fluid medium within the industrial boiler. This coefficient serves to physically bridge the temperature change rate to the voxel dissipation rate. This represents the numerical value within the space thermal safety tolerance zone, with dimensions in Kelvin per second. Furthermore, the tolerance boundary transformation unit assembles and generates a Boolean constraint equation set using the logical comparison results. The Boolean constraint equation set refers to the set of algebraic equations composed of the binary logical determination results of whether multiple candidate voxel nodes possess adjustment margins. The constraint state generation logic of the tolerance boundary transformation unit satisfies the following formula: , Among them, letters Represents the dimensionless, binary state variable assigned to the candidate voxel node in the Boolean constraint equations; letters This represents a mathematical symbol function; it outputs a positive 1 when the value inside the parentheses is greater than zero and a negative 1 when it is less than or equal to zero; letters... Represents the boundary of the extracted node feasible adjustment domain, with the dimension of watts per cubic meter Kelvin; letter The feature element in the array representing the real-time entropy generation rate of candidate voxel nodes is measured in watts per cubic meter (Kelvin). When the real-time entropy generation rate of a candidate voxel node is less than the boundary of the node's feasible adjustment domain, it indicates that the node still has a safety margin for adjustment, and the binarized state variable is calculated to a value of one; otherwise, it is assigned a value of zero, and the binarized state variables of all nodes are assembled together to generate a Boolean constraint equation system.

[0052] For example, the tolerance boundary transformation unit extracts the real-time entropy generation rate array of a candidate voxel node and reads the feature elements therein. for Watts per cubic meter (Kelvin). The tolerance boundary transformation unit extracts a predetermined thermal dissipation mapping coefficient based on measured data of the fluid medium's isobaric specific heat capacity and density. for Watts per second per cubic meter (Kelvin squared). The tolerance boundary transformation unit retrieves the spatial thermal safety tolerance zone corresponding to the node coordinates. The value is Kelvin per second. The tolerance boundary transformation unit performs logical inequality transformation calculations, converting the digital... With numbers Perform multiplication to extract the feasible adjustment domain boundary of the node. for Watts per cubic meter Kelvin. The tolerance boundary transformation unit then assembles and generates the constraint equation state, adjusting the nodal feasible adjustment domain boundary. Subtract its real-time feature elements The difference was found to be positive. Since the difference is greater than zero, the sign function... The output value is positive one. The tolerance boundary transformation element adds one to the positive one value and then divides it by two to calculate the binary state variable of that node in the Boolean constraint equations. for The tolerance boundary transformation unit repeats the above digital logic operations in parallel for all marked candidate voxel nodes, and finally combines all the binary state variables to generate a complete set of Boolean constraint equations, which are then sent to the subsequent control loop.

[0053] The incremental control level solving module is used to extract the current fuel valve opening parameters using the candidate voxel nodes, solve the Boolean constraint equations to obtain the positive solution vector, and convert the positive solution vector into an incremental control level output to the fuel valve actuator.

[0054] Optionally, the incremental control level solving module includes: The constraint iterative optimization unit is used to extract the current fuel valve opening parameter based on the physical mapping address of the candidate voxel node, fuse the Boolean constraint equation set with the space thermal safety tolerance band to construct the optimization penalty function, and perform gradient descent calculation starting from the current fuel valve opening parameter to solve and extract the positive solution vector. The constraint-based iterative optimization unit extracts the current fuel valve opening parameters based on the physical mapping addresses of candidate voxel nodes. The physical mapping address is the unique binding communication identifier between the virtual 3D mesh spatial coordinates and the underlying real burner mechanical valves in an industrial control system. The current fuel valve opening parameter refers to the percentage of the mechanical valves controlling the fuel medium flow rate at the current physical moment. The constraint-based iterative optimization unit integrates Boolean constraint equations with the spatial thermal safety tolerance band to construct an optimization penalty function. The optimization penalty function applies an extremum penalty to mathematical solutions that attempt to violate the physical safety boundary during the numerical optimization iteration process, thereby forcing the calculation results to converge to the target optimization equation within the absolute safety regulation domain. The construction logic of the optimization penalty function satisfies the following formula: , Among them, letters The function representing the optimization penalty is dimensionless; the letter represents the value. Represents the iterative aperture variable in the optimization process; it is dimensionless; the letter represents the aperture value. Represents the current fuel valve opening parameter, dimensionless; letter Represents the penalty weight multiplier, which is dimensionless. The value of the penalty weight multiplier is based on the overshoot divergence extreme value setting in the historical control system joint debugging log to ensure that the penalty term has absolute numerical dominance when an out-of-bounds event occurs; the letter... Represents the binary state variable of the corresponding node in the Boolean constraint equations; it is dimensionless and its value is either one or zero; the letter... The conversion factor representing the conversion from a thermal scalar to a mechanical opening, with dimensions in seconds per Kelvin, is obtained by referring to the factory thermal calibration test data of the burner's underlying hardware; the letter... This represents the space thermal safety tolerance zone, measured in Kelvin per second. Subsequently, the constrained iterative optimization unit performs gradient descent calculations starting from the current fuel valve opening parameters. Gradient descent calculation refers to the mathematical optimization search process that successively approaches the local minimum point of the function along the negative gradient direction of the optimization penalty function with a fixed iteration step size. The logic of gradient descent calculation satisfies the following formula: , Among them, letters Represents the opening variable after the next iteration update; dimensionless; letter. Represents the opening degree variable within the current iteration period; dimensionless; letter. Represents the learning rate step size, dimensionless. The value of the learning rate step size is derived from the maximum operating frequency of the controller hardware and the reciprocal of the control cycle time to strictly prevent iterative divergence; the letter... The first-order partial derivative of the optimization penalty function with respect to the iterative opening variable is dimensionless. The constrained iterative optimization unit calculates through continuous gradient descent until the function converges, extracting the positive solution vector. The positive solution vector refers to the opening change that minimizes local thermodynamic dissipation and ensures absolute safety after safety constraint filtering and iterative optimization. The calculation logic of the positive solution vector satisfies the following formula: , Among them, letters Represents the positive solution vector, dimensionless; letters The variable representing the final steady-state opening after gradient descent calculations have fully converged is dimensionless; the letter represents the final steady-state opening variable. The current fuel valve opening parameter represents the starting point for optimization and is dimensionless.

[0055] The duty cycle mapping output unit is used to obtain the hardware drive gain coefficient of the fuel valve actuator, perform scalar multiplication mapping between the forward vector and the hardware drive gain coefficient to generate an incremental control level, and send the incremental control level to the fuel valve actuator.

[0056] The duty cycle mapping output unit obtains the hardware drive gain coefficient of the fuel valve actuator. The fuel valve actuator is a physical hardware device that receives analog electrical signals and converts them into proportional mechanical stroke to throttle the fuel pipeline. The hardware drive gain coefficient is the physical amplification ratio parameter between the actuator's input electrical control signal and the actual mechanical opening displacement. The hardware drive gain coefficient is directly extracted from the fuel valve actuator's electrical parameter hardware specification. The duty cycle mapping output unit performs a scalar multiplication mapping between the forward vector and the hardware drive gain coefficient. Scalar multiplication mapping involves algebraically multiplying the dimensionless change amplitude obtained from pure logical operations with a proportionality coefficient containing physical units, thereby converting the digital quantity across modes into a physical electrical quantity. Through this operation, the duty cycle mapping output unit generates an incremental control level. The incremental control level is the high / low potential difference signal used to drive the internal servo motor of the fuel valve actuator to generate a specific mechanical displacement. The logic for generating the incremental control level satisfies the following formula: , Among them, letters Represents the generated incremental control level, in volts; the letter... The vector representing the positive solution output by the constraint-based iterative optimization unit is dimensionless; the letter represents the positive solution. This represents the hardware drive gain coefficient, measured in volts. Finally, the duty cycle mapping output unit sends the incremental control level to the fuel valve actuator via a digital-to-analog converter interface to complete the automated physical adjustment for abnormal heat dissipation in the boiler in a closed loop.

[0057] For example, the duty cycle mapping output unit obtains the positive solution vector of the upstream output. for The duty cycle mapping output unit extracts the hardware drive gain coefficient of this model of fuel valve actuator by calling the electrical parameter hardware manual. for Volt, which corresponds to 100% opening at full scale. The driving voltage is volts. The duty cycle mapping output unit performs a scalar multiplication mapping between the forward vector and the hardware drive gain coefficient, and generates the incremental control level through digital multiplication calculation. for Volt. The duty cycle mapped output unit will then contain The digital message of the volt signal command is sent to the communication pin of the fuel valve actuator, driving the mechanical valve core to perform the corresponding opening increase action.

[0058] Based on the same inventive concept, the present invention also provides an automated boiler operation control method, the method comprising: Acquire the time-series readings of temperature sensors in the boiler furnace and the edge video stream, perform 3D point cloud fusion and voxelization spatial discretization operations, and construct a 3D mesh spatial model of the furnace. Using the three-dimensional mesh space model of the furnace, the temporal temperature data of the mesh nodes are extracted from the steady-state operation stage, and the spatiotemporal temperature gradient matrix is ​​calculated for adjacent voxel nodes to generate the node reference entropy generation rate matrix. Collect real-time temperature readings, calculate the real-time temperature gradient matrix between adjacent voxel nodes, and map it to generate a node real-time entropy generation rate array. Gaussian mixture distribution fitting is performed on the node baseline entropy generation rate matrix to generate a dynamic trigger baseline, and abnormal voxel node data above the dynamic trigger baseline are extracted from the node real-time entropy generation rate array. The abnormal voxel node data is sent as a communication broadcast packet to the spatially adjacent candidate voxel nodes, and the real-time entropy generation rate array of the candidate voxel nodes is extracted to construct a Boolean constraint equation system. The current fuel valve opening parameters are extracted using the candidate voxel nodes, the Boolean constraint equations are solved to obtain the positive solution vector, and the positive solution vector is converted into an incremental control level and output to the fuel valve actuator.

[0059] It should be noted that the formulas described above, through the principle of dimensional consistency and mathematical standardization methods (such as normalization, dimensionless parameter conversion, or unit system unification), can translate physical quantities with different properties into unitless standard values ​​or parameters that can be superimposed in the same dimension. This eliminates the interference of different dimensions on the computational logic, allowing the formulas to retain the original data distribution characteristics while possessing mathematical rationality and adaptability to objective laws. These are conventional technical methods and will not be elaborated further. The electrical connections between the various units described above do not necessarily represent direct or indirect connections; any indirect connection method is applicable to the embodiments of this invention as long as it achieves the purpose of this invention. The above descriptions are merely exemplary embodiments of this invention and should not be construed as limiting the scope of this invention.

[0060] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. An automated boiler operation control system, characterized in that, The system includes: The furnace voxel mesh construction module is used to acquire the time-series readings of temperature sensors and edge video streams in the boiler furnace, perform 3D point cloud fusion and voxel spatial discretization operations, and construct a 3D mesh spatial model of the furnace. The baseline entropy generation rate generation module is used to extract the temporal temperature data of the grid nodes from the steady-state operation stage using the three-dimensional grid space model of the furnace, calculate the spatiotemporal temperature gradient matrix for adjacent voxel nodes, and generate the node baseline entropy generation rate matrix. The real-time entropy generation rate mapping module is used to collect real-time temperature readings, calculate the real-time temperature gradient matrix between adjacent voxel nodes, and map and generate a node real-time entropy generation rate array. An abnormal voxel node detection module is used to fit the node baseline entropy generation rate matrix to a Gaussian mixture distribution, generate a dynamic trigger baseline, and extract abnormal voxel node data that are higher than the dynamic trigger baseline from the node real-time entropy generation rate array. The Boolean constraint equation construction module is used to send the abnormal voxel node data as a communication broadcast packet to the spatially adjacent candidate voxel nodes, and extract the real-time entropy generation rate array of the candidate voxel nodes to construct a set of Boolean constraint equations. The incremental control level solving module is used to extract the current fuel valve opening parameters using the candidate voxel nodes, solve the Boolean constraint equations to obtain the positive solution vector, and convert the positive solution vector into an incremental control level output to the fuel valve actuator.

2. The automated boiler operation control system according to claim 1, characterized in that, The furnace voxel mesh construction module includes: A multimodal thermal point cloud generation unit is used to extract a two-dimensional flame contour image using the edge video stream, reconstruct the two-dimensional flame contour image into a three-dimensional spatial geometric coordinate set, and map and anchor the time-series readings of the temperature sensor to the three-dimensional spatial geometric coordinate set to generate a multimodal thermal point cloud set. The voxel mesh spatial modeling unit is used to extract a three-dimensional spatial bounding box using the multimodal thermal point cloud, perform spatial segmentation calculations on the three-dimensional spatial bounding box to generate three-dimensional voxel units, and assemble the three-dimensional voxel units to construct a three-dimensional mesh spatial model of the furnace.

3. An automated boiler operation control system according to claim 2, characterized in that, The baseline entropy generation rate generation module includes: The spatiotemporal gradient extraction unit is used to extract historical grid node time-series temperature data from the three-dimensional grid spatial model of the furnace and perform first-order derivative calculation of the sliding window to generate a temperature change rate sequence. The steady-state interval extraction unit is used to perform kernel density estimation on the temperature change rate sequence, extract the rate of change envelope of the probability density peak, and extract the continuous time segment of the temperature change rate sequence that falls into the rate of change envelope as the steady-state operation stage; The thermodynamic dissipation mapping unit is used to perform partial derivative calculations on the temperature data of the steady-state operation phase along the spatial coordinate axes to generate a spatiotemporal temperature gradient matrix, and to map and generate a node reference entropy generation rate matrix.

4. An automated boiler operation control system according to claim 1, characterized in that, The real-time entropy generation rate mapping module includes: The reading noise reduction and smoothing unit is used to acquire the real-time temperature measurement raw signal using adjacent voxel nodes, perform filtering processing on the real-time temperature measurement raw signal, extract convergent state variables, and generate net-state real-time temperature readings. The quadratic dissipative mapping unit is used to extract the spatial coordinate difference from adjacent voxel nodes, calculate the real-time temperature gradient matrix by combining the net real-time temperature reading, and output the real-time entropy generation rate array of the generating nodes.

5. An automated boiler operation control system according to claim 1, characterized in that, The abnormal voxel node detection module includes: The parameter iteration and benchmark construction unit is used to perform expectation maximization iteration on the node benchmark entropy generation rate matrix, extract the mean vector and covariance matrix of the Gaussian components, fuse the mean vector and covariance matrix to calculate the upper boundary envelope equation, and generate a dynamic triggering benchmark line. The out-of-bounds stripping and extraction unit is used to project the real-time entropy generation rate array of the nodes onto the upper boundary envelope equation, perform element-wise algebraic difference calculation, extract the target array coordinates with a difference greater than zero, and extract abnormal voxel node data based on the target array coordinates.

6. An automated boiler operation control system according to claim 3, characterized in that, The system also includes: Extract the spatial geometric boundary of the three-dimensional bounding box and the steady-state fluctuation extreme value of the rate of change envelope. Perform physical property mapping and binding on the spatial geometric boundary and the steady-state fluctuation extreme value to generate a spatial thermal safety tolerance band.

7. An automated boiler operation control system according to claim 6, characterized in that, The Boolean constraint equation construction module includes: The geometric addressing and broadcasting unit is used to extract the three-dimensional spatial coordinates of the abnormal voxel node data, perform coplanar adjacent boundary retrieval in the three-dimensional mesh space model of the furnace to mark candidate voxel nodes, and convert the abnormal voxel node data into a communication broadcast packet and send it to the candidate voxel node. The tolerance boundary transformation unit is used to extract the real-time entropy generation rate array of the candidate voxel nodes, perform logical inequality transformation between the real-time entropy generation rate array and the space thermal safety tolerance zone, extract the feasible adjustment domain boundary of the node, and assemble and generate a Boolean constraint equation system.

8. An automated boiler operation control system according to claim 7, characterized in that, The incremental control level calculation module includes: The constraint iterative optimization unit is used to extract the current fuel valve opening parameter based on the physical mapping address of the candidate voxel node, fuse the Boolean constraint equation set with the space thermal safety tolerance band to construct the optimization penalty function, and perform gradient descent calculation starting from the current fuel valve opening parameter to solve and extract the positive solution vector. The duty cycle mapping output unit is used to obtain the hardware drive gain coefficient of the fuel valve actuator, perform scalar multiplication mapping between the forward vector and the hardware drive gain coefficient to generate an incremental control level, and send the incremental control level to the fuel valve actuator.

9. An automated boiler operation control method, applied to an automated boiler operation control system as described in any one of claims 1-8, characterized in that, The method includes: Acquire the time-series readings of temperature sensors in the boiler furnace and the edge video stream, perform 3D point cloud fusion and voxelization spatial discretization operations, and construct a 3D mesh spatial model of the furnace. Using the three-dimensional mesh space model of the furnace, the temporal temperature data of the mesh nodes are extracted from the steady-state operation stage, and the spatiotemporal temperature gradient matrix is ​​calculated for adjacent voxel nodes to generate the node reference entropy generation rate matrix. Collect real-time temperature readings, calculate the real-time temperature gradient matrix between adjacent voxel nodes, and map it to generate a node real-time entropy generation rate array. Gaussian mixture distribution fitting is performed on the node baseline entropy generation rate matrix to generate a dynamic trigger baseline, and abnormal voxel node data above the dynamic trigger baseline are extracted from the node real-time entropy generation rate array. The abnormal voxel node data is sent as a communication broadcast packet to the spatially adjacent candidate voxel nodes, and the real-time entropy generation rate array of the candidate voxel nodes is extracted to construct a Boolean constraint equation system. The current fuel valve opening parameters are extracted using the candidate voxel nodes, the Boolean constraint equations are solved to obtain the positive solution vector, and the positive solution vector is converted into an incremental control level and output to the fuel valve actuator.

Images