Boiler whole-stack self intelligent monitoring method and device based on domestic trust and creativity and storage medium

Abstract

According to the boiler full-stack body intelligent monitoring method and device based on the domestic trust creativity and the storage medium, the method comprises the steps that the connectivity and holes of dynamic feature flows are quantified through homology groups, topological enhanced dynamic feature flows are constructed, and the anomaly recognition accuracy is improved in boiler pipeline dense areas and smoke shielding scenes; by calculating the enhanced physical residual loss, the verification result is forced to conform to the thermal fluid and energy conservation law, and the misjudgment rate of normal working condition fluctuation is greatly reduced; a strong dominant root cause is screened through a dominant root cause positioning formula, a verification result and a root cause report are output, the contribution degree of each candidate factor is quantified, a core fault point is directly locked, and the fault handling time is greatly shortened; the sample quality and the training contribution of each power plant are quantified according to the node contribution degree calculation formula through the node local income and the node cooperation relationship, the defect that the weight is high if multiple samples exist is avoided, and the adaptation rate of the model to boilers with different load rates and different fuel types is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of boiler-embedded intelligent monitoring technology, specifically to a boiler full-stack embedded intelligent monitoring method, equipment, and storage medium based on domestically developed information technology innovation. Background Technology

[0002] Intelligent boiler monitoring is a modern monitoring system that relies on the Internet of Things, sensors, big data analytics, and artificial intelligence technologies to achieve real-time perception, precise analysis, and intelligent early warning throughout the entire boiler operation process. It breaks through the limitations of traditional manual inspections, enabling a transformation from "passive maintenance" to "proactive prediction," providing core support for the safe, efficient, and low-carbon operation of boilers.

[0003] Existing boilers have dense internal piping and uneven combustion fields, making purely dynamic data susceptible to occlusion interference. This makes it impossible to distinguish between valid anomalies and occlusion pseudo-features. Traditional monitoring relies on threshold comparisons, which do not consider the physical constraints of boiler operation and are prone to misjudging normal operating fluctuations as anomalies. Boiler anomalies are often caused by the coupling of multiple factors, and traditional methods cannot quantify the contribution of each factor, resulting in poor targeted fault handling. Furthermore, boiler operating conditions vary greatly among different power plants, and some power plants have scarce anomaly samples. If the model parameters are aggregated by average weighting, the model will be biased towards power plants with abundant samples, resulting in poor adaptability. Summary of the Invention

[0004] This invention proposes a full-stack embodied intelligent monitoring method, equipment, and storage medium for boilers based on domestically developed information technology. This addresses the challenges of existing boiler systems, such as dense internal piping, uneven combustion fields, susceptibility to interference from obstruction in purely dynamic data, inability to distinguish between valid anomalies and false features caused by obstruction, reliance on threshold comparisons in traditional monitoring which fails to consider the physical constraints of boiler operation and easily misjudges normal operating fluctuations as anomalies, and the fact that boiler anomalies are often the result of multiple coupled factors. Traditional methods cannot quantify the contribution of each factor, leading to poor targeted fault handling. Furthermore, boiler operating conditions vary significantly across different power plants, and some power plants have scarce anomaly samples. If the model parameters are aggregated using average weighting, the model may become biased towards power plants with abundant samples, resulting in poor adaptability.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The boiler full-stack embodied intelligent monitoring method based on domestic information technology innovation of the present invention includes: S1. By linking neuromorphic sensors with a digital twin engine, unstructured dynamic feature flow inside the boiler is collected. S2. Input the dynamic feature stream into the hybrid architecture of Vision Mamba and spiking neural network, dynamically adjust the system parameters using attention weights, prioritize memorizing the high-risk region features marked by the digital twin engine, and generate a sparse latent state tensor that integrates physical realism and virtual correlation. S3. Construct an industrial world model that integrates a digital twin engine and a causal inference model, map the sparse latent state tensor to a high-dimensional manifold space, verify the current physical state corresponding to the sparse latent state tensor, trace the causal relationship between anomalies and key events in the dynamic feature flow, and output the verification results and root cause report. S4. Receive the verification results and root cause report, integrate the virtual simulation data of the digital twin engine, and generate collaborative control instructions; S5. Using bee colony learning and fully homomorphic encryption technology, the industrial world model is decrypted and updated based on the verification results, the root cause report, and the collaborative control instructions. Based on the updated industrial world model, a boiler intelligent monitoring scheme is generated.

[0006] Preferably, in S1, the neuromorphic sensor collects dynamic feature streams and outputs (x,y,t,p) quadruple streams, which map the microsecond-level dynamic feature streams to the boiler digital twin engine in real time, realizing data synchronization between the physical device and the virtual image. The digital twin engine automatically marks high-risk areas, quantifies the connectivity of the dynamic feature streams through the 0th-order homology group and identifies holes in the dynamic feature streams through the 1st-order homology group, and constructs a topology-enhanced dynamic feature stream. The homology group formula is as follows: ; ; in, It is a homology group of order 0. It is a first-order homology group. For dynamic feature flow, Boundary operators The core space, Boundary operators Image space, For topology-enhanced dynamic feature flow, as well as These are the weights of the 0th-order homology group and the 1st-order homology group, respectively.

[0007] Preferably, in S2, the spiking neural network transforms the topology-enhanced dynamic feature flow into a sparse pulse tensor, inputs the time series of the topology-enhanced dynamic feature flow, performs dynamic stability evaluation, and then calculates the enhanced mutual information between the topology-enhanced dynamic feature flow and the sparse pulse tensor. The dynamic stability evaluation formula is as follows: ; in, for The maximum Lyapunov exponent, quantizing the dynamic stability of the feature flow. For time variables, For time window, For the Vision Mamba model time Feature mapping function, For feature mapping function pairs The first derivative; The formula for enhancing mutual information is as follows: ; ; in, For sparse latent state tensors To enhance mutual information, quantification and The degree of correlation, For original mutual information, The chaos suppression coefficient is... To obtain The maximum value of 0.

[0008] Preferably, the verification of the current physical state corresponding to the sparse latent state tensor in step S3 includes the following steps: S31. By constructing a physical state density matrix, the discrete sparse latent state tensor is transformed into an uncertain probability matrix. S32. Conduct a physical state uncertainty assessment, and the physical verification formula corrects the physical verification result by incorporating uncertainty. S33. Substitute the topological features of the input anomaly and dynamic feature flow into the order parameter change rate formula to calculate the change rate of each root cause. S34. Screen strong dominant root causes using the dominant root cause localization formula, and output the verification results and root cause report.

[0009] Preferably, the formula for the physical state density matrix is ​​as follows: ; The formula for assessing the uncertainty of the physical state is as follows: ; in, Let Y be the physical state density matrix. Let Y be the transpose of Y. For the trace operation of a matrix, For Feng Neumann entropy is used to quantify the uncertainty of a physical state; The physical verification formula is as follows: ; ; in, For the original physical residual loss, For physical constraint weights, ) is the PDE operator acting on As a result, The root cause correction coefficient. For enhanced physical residual loss; The formula for the rate of change of the order parameter is as follows: ; The formula for determining the dominant root cause is as follows: ; in, For the first Rate of change of order parameters of candidate root causes For the first The order parameter of each candidate root cause The optimized dominant root cause order parameter, The root cause correlation coefficient, For sparse latent state tensors under abnormal states. Assuming the first Anomalous feature tensor when candidate root causes are absent As an external driving force, This is the uncertainty correction coefficient.

[0010] Preferably, step S4 includes the following steps: S41. Receive the verification result and root cause report, input the enhanced physical residual loss, and determine the control error; S42. Obtain the optimal PID parameters through the norm-based anti-interference optimization formula; S43. Finally, substitute the parameters into the fractional-order PID control law formula to calculate the control law and generate cooperative control instructions, which include DCS adjustment instructions and robot inspection instructions. The formula for calculating the control error is as follows: ; Among them, 0.04 is The normal threshold; The norm-based anti-interference optimization formula is as follows: ; ; ; in, To find the minimum value of the PID parameters, This is the proportionality coefficient. The integral coefficient is... The differential coefficients are... For interference To control error The transfer function, To cause interference at the industrial site, For sliding surface functions, For fractional order, For integral weights; The formula for the fractional-order PID control law is as follows: ; in, For control laws, For fractional differential operators, For fractional integral operators, For sliding mode switching gain, It is a symbolic function.

[0011] Preferably, updating the industrial world model in S5 includes the following steps: S51, each power plant edge node is based on local data , as well as Train the Vision Mamba model and the causal inference model to obtain local model parameters; S52. Based on the node's local revenue and node collaboration relationship, and according to the node contribution calculation formula, quantify the node's contribution to the model. S53. Each node generates a quantum key through the BB84 protocol and verifies qualified nodes according to the key rate formula and enhancement formula. S53. Based on bee colony learning, a leader node is randomly elected in each round of training. The leader node inputs the contribution of all nodes and obtains the updated industrial world model through the aggregation formula.

[0012] Preferably, the formula for calculating the node contribution is as follows:

[0013] in, For the first The contribution of each edge node. The damping coefficient is... For the first Local revenue of each edge node For the first The set of incoming edges of each edge node For the first The contribution of each edge node. For the first Local revenue of each edge node For the first The number of outgoing edges from each edge node; The key rate formula is as follows: ; The enhancement formula is as follows: ; in, This is the raw key rate for the BB84 protocol. For enhanced security key rate, For photon transmission probability, For detection efficiency, It is a binary entropy function. The original bit error rate. The bit error rate after error correction. This is the interference correction factor; The aggregation formula is as follows: ; in, For aggregation model parameters, The total number of edge nodes participating in the collaboration. For the first The result of encrypting the local model parameters of each edge node.

[0014] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0015] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0016] As can be seen from the above technical solution, this invention provides a full-stack intelligent monitoring method for boilers based on domestically developed information technology. Compared with the prior art, this invention has the following advantages: 1. By quantizing the connectivity of the dynamic feature flow using the 0th-order homology group and identifying holes in the dynamic feature flow using the 1st-order homology group, a topology-enhanced dynamic feature flow is constructed. The dynamic feature flow that integrates topological features does not rely on a single pixel signal, but rather selects effective features based on spatial structural correlation. This improves the accuracy of anomaly identification in dense boiler pipe areas and flue gas obstruction scenarios.

[0017] 2. By calculating the enhanced physical residual loss, the forced verification results conform to the laws of thermal fluid and energy conservation, and the misjudgment rate of normal operating condition fluctuations is greatly reduced.

[0018] 3. By using the dominant root cause localization formula to screen strong dominant root causes, output verification results and root cause reports, quantify the contribution of each candidate factor, directly pinpoint the core fault point, and greatly reduce fault handling time.

[0019] 4. By considering the local benefits of nodes and the collaborative relationships between nodes, and based on the node contribution calculation formula, the contribution of nodes to the model is quantified. According to bee colony learning, a leader node is randomly elected in each round of training. The leader node inputs the contribution of all nodes, and the updated industrial world model is obtained through the aggregation formula. This quantifies the sample quality and training contribution of each power plant, avoiding the drawback of high weight due to a large number of samples, and greatly improving the model's adaptability to boilers with different load rates and different fuel types. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the full-stack embodied intelligent monitoring method for boilers based on domestically developed information technology innovation, as described in this invention. Detailed Implementation

[0021] 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 some embodiments of the present invention, but not all embodiments.

[0022] like Figure 1 As shown in this embodiment, the boiler full-stack embodied intelligent monitoring method based on domestic information technology innovation includes: S1. By linking neuromorphic sensors with a digital twin engine, unstructured dynamic feature flow inside the boiler is collected. S2. Input the dynamic feature stream into the hybrid architecture of Vision Mamba and spiking neural network, use attention weights to dynamically adjust system parameters, prioritize memorizing the high-risk region features marked by the digital twin engine, and generate a sparse latent state tensor that integrates physical reality and virtual correlation. S3. Construct an industrial world model that integrates a digital twin engine and a causal inference model. Map the sparse latent state tensor to a high-dimensional manifold space, verify the current physical state corresponding to the sparse latent state tensor, trace the causal relationship between anomalies and key events in the dynamic feature flow, and output the verification results and root cause report. S4. Receive the verification results and root cause report, integrate the virtual simulation data from the digital twin engine, and generate collaborative control commands; S5. Utilize bee colony learning and fully homomorphic encryption technology to decrypt and update the industrial world model based on verification results, root cause reports, and collaborative control instructions. Generate a boiler intelligent monitoring solution based on the updated industrial world model.

[0023] In S1, the neuromorphic sensor collects dynamic feature streams and outputs (x,y,t,p) quadruple streams. The microsecond-level dynamic feature streams are mapped to the boiler digital twin engine in real time, realizing data synchronization between the physical device and the virtual image. The digital twin engine automatically marks high-risk areas, quantifies the connectivity of the dynamic feature streams through the 0th-order homology group, and identifies holes in the dynamic feature streams through the 1st-order homology group, thus constructing a topology-enhanced dynamic feature stream. The homology group formula is as follows: ; ; in, It is a homology group of order 0. It is a first-order homology group. For dynamic feature flow, Boundary operators The core space, Boundary operators Image space, For topology-enhanced dynamic feature flow, as well as These are the weights of the 0th homology group and the 1st homology group, respectively. In practical applications, six DVS sensors are deployed at the four corners of the boiler and on both sides of the screen-type superheater to collect the original dynamic characteristic flow within a 100ms time window. Substituting these samples into the homology group formula, the topology calculation module of the edge nodes automatically solves the problem. The continuous pulses of leakage turbulence in the dynamic characteristic flow form three spatial propagation paths, leading to the conclusion that... The presence of two pulseless vortex core regions at the center of the flame indicates... ; Set weights , Substitute into the fusion formula to calculate: , and thus ,Will , , Synchronized to the boiler digital twin engine, the engine is based on Mark 3 leakage propagation paths, based on Two high-risk vortex regions were marked to provide a dual-dimensional input of "dynamic + spatial" for subsequent coding.

[0024] In summary, through homology groups (Number of connected components) (Number of voids) quantifies spatial structure, accurately identifying features that traditional monitoring methods cannot capture, such as turbulence propagation paths and flame gaps, reducing the spatial location error of anomalies such as leaks and flame shifts from meters to centimeters; it also integrates topological features... Instead of relying on a single pixel signal, it filters effective features based on spatial structural correlation, improving the anomaly recognition accuracy by more than 30% in dense boiler pipe areas and flue gas obstruction scenarios.

[0025] In S2, the spiking neural network transforms the topology-enhanced dynamic feature flow into a sparse pulse tensor. The time series of the topology-enhanced dynamic feature flow is input for dynamic stability evaluation, and then the enhanced mutual information between the topology-enhanced dynamic feature flow and the sparse pulse tensor is calculated. The dynamic stability assessment formula is as follows: ; in, for The maximum Lyapunov exponent, quantizing the dynamic stability of the feature flow. For time variables, For time window, For the Vision Mamba model time Feature mapping function, For feature mapping function pairs The first derivative; The formula for enhancing mutual information is as follows: ; ; in, For sparse latent state tensors To enhance mutual information, quantification and The degree of correlation, For original mutual information, The chaos suppression coefficient is... To obtain The maximum value of 0; In practical applications, receiving The data is sliced ​​into 100ms time windows, with each slice containing approximately 800,000 feature data points. The spiking neural network compresses these 800,000 feature data points into 400,000. Substitute into the dynamic stability evaluation formula It was determined to be in a "weakly chaotic state"; First, calculate the original mutual information, then obtain it through the information entropy calculation module of the edge nodes. Bit, Bits, get ; Recalculate enhanced mutual information and set , , It was further reduced to 200,000 entries.

[0026] In summary, through Quantify feature flow stability, by By modifying mutual information, the feature weights of chaotic disturbances such as combustion fluctuations are reduced by 40%, the amount of encoded data is reduced by 50%, and the inference latency at edge nodes is reduced to 3. ms Within.

[0027] S3 verifies the current physical state corresponding to the sparse latent state tensor, including the following steps: S31. By constructing a physical state density matrix, the discrete sparse latent state tensor is transformed into an uncertain probability matrix. S32. Conduct a physical state uncertainty assessment, and the physical verification formula corrects the physical verification result by incorporating uncertainty. S33. Substitute the topological features of the input anomaly and dynamic feature flow into the order parameter change rate formula to calculate the change rate of each root cause. S34. Screen strong dominant root causes using the dominant root cause localization formula, and output the verification results and root cause report.

[0028] The formula for the physical state density matrix is ​​as follows: ; The formula for assessing the uncertainty of physical state is as follows: ; in, Let Y be the physical state density matrix. Let Y be the transpose of Y. For the trace operation of a matrix, For Feng Neumann entropy is used to quantify the uncertainty of a physical state; The physical verification formula is as follows: ; ; in, For the original physical residual loss, For physical constraint weights, ) is the PDE operator acting on As a result, The root cause correction coefficient. For enhanced physical residual loss; The formula for the rate of change of the order parameter is as follows: ; The formula for identifying the dominant root cause is as follows: ; in, For the first Rate of change of order parameters of candidate root causes For the first The order parameter of each candidate root cause The optimized dominant root cause order parameter, The root cause correlation coefficient, For sparse latent state tensors under abnormal states. Assuming the first Anomalous feature tensor when candidate root causes are absent As an external driving force, This is the uncertainty correction coefficient.

[0029] In practical applications, real-time boiler operating data (steam temperature 540℃, pressure 25MPa, load rate 90%) is acquired synchronously as a reference benchmark for physical constraint verification. Y is a tensor of 200,000 × 102. A 10×10 covariance matrix is ​​obtained, and the trace operation is performed. After normalization, we get Calculated The bit, because it is greater than 1.0, is judged as an abnormal precursor; set up , The result is the effect of the thermal fluid PDE operator. , , Temporary ,calculate Enhanced physical residual This indicates a physical anomaly. Then, several candidate root causes were identified, including burner offset. Insufficient cooling water flow The smoke baffle is stuck. Then set , , hour, , hour, , hour, Then set Calculations yielded Output root cause report: The core root cause of the screen-type superheater abnormality: burner misalignment; In summary, through The forced verification results conform to the laws of thermal fluid and energy conservation, reducing the misjudgment rate of normal operating condition fluctuations to below 5%, significantly reducing ineffective operation and maintenance. Quantify the contribution of each candidate factor. By directly pinpointing the core fault point, the time for handling superheater overheating in 660MW units has been reduced from 30 minutes to 12 minutes, shortening the fault handling time by 60%.

[0030] S4 includes the following steps: S41. Receive the verification results and root cause report, input the enhanced physical residual loss, and determine the control error; S42. Obtain the optimal PID parameters through the norm-based anti-interference optimization formula; S43. Finally, substitute the parameters into the fractional PID control law formula to calculate the control law and generate cooperative control instructions, which include DCS adjustment instructions and robot inspection instructions. The formula for calculating control error is as follows: ; Among them, 0.04 is The normal threshold; The optimization formula for norm-based anti-interference is as follows: ; ; ; in, To find the minimum value of the PID parameters, This is the proportionality coefficient. The integral coefficient is... The differential coefficients are... For interference To control error The transfer function, To cause interference at the industrial site, For sliding surface functions, For fractional order, For integral weights; The formula for the fractional-order PID control law is as follows: ; in, For control laws, For fractional differential operators, For fractional integral operators, For sliding mode switching gain, It is a symbolic function.

[0031] In practical applications, receiving Controlling error ; On-site interference For power grid fluctuations Fractional order The integral weight is 0.8. If it is 0.3, then Solving by optimizing the formula , , ,make Substituting into the fractional-order PID control law formula, we obtain... ; Will Converted to DCS command: Adjust the burner angle to the left by 2.5°, increase the opening of the right desuperheating water valve by 5.2%, and collect new data 5ms after execution. , When the value approaches zero, control takes effect; In summary, the fractional order can adapt to abnormal nonlinear dynamics, with a control response speed of <8ms, which is 50% faster than traditional PID, achieving a millisecond-level closed loop of "abnormality-control-stability".

[0032] Updating the Industrial World model in S5 includes the following steps: S51, each power plant edge node is based on local data , as well as Train the Vision Mamba model and the causal inference model to obtain local model parameters; S52. Based on the node's local revenue and node collaboration relationship, and according to the node contribution calculation formula, quantify the node's contribution to the model. S53. Each node generates a quantum key through the BB84 protocol and verifies qualified nodes according to the key rate formula and enhancement formula. S53. Based on bee colony learning, a leader node is randomly elected in each round of training. The leader node inputs the contribution of all nodes and obtains the updated industrial world model through the aggregation formula.

[0033] The formula for calculating node contribution is as follows:

[0034] in, For the first The contribution of each edge node. The damping coefficient is... For the first Local revenue of each edge node For the first The set of incoming edges of each edge node For the first The contribution of each edge node. For the first Local revenue of each edge node For the first The number of outgoing edges from each edge node; The key rate formula is as follows: ; The enhancement formula is as follows: ; in, This is the raw key rate for the BB84 protocol. For enhanced security key rate, For photon transmission probability, For detection efficiency, It is a binary entropy function. The original bit error rate. The bit error rate after error correction. This is the interference correction factor; The aggregation formula is as follows: ; in, For aggregation model parameters, The total number of edge nodes participating in the collaboration. For the first The result of encrypting the local model parameters of each edge node; In practical applications, five power plants (all with 660MW ultra-supercritical boilers) serve as edge nodes. Each node deploys the same Vision Mamba model and locally stores nearly three months' worth of data. Y, collaborative control command data; Set damping coefficient The local revenue of each edge node is 0.85. The revenues for power plants 1 (0.92), 2 (0.88), 3 (0.75), 4 (0.83), and 5 (0.87) are calculated by substituting these values ​​into the node contribution formula. , , , , ; set up It is 0.9. It is 0.85. It is 0.03. The value is 0.008, calculated as follows: , , Calculations yielded It was determined to be a safe node.

[0035] calculate ; After aggregation, we get .

[0036] In summary, through Quantifying the sample quality and training contribution of each power plant avoids the drawback of "more samples equal higher weights," thus improving the model's adaptability to boilers with different load rates (50%-100%) and different fuel types (bituminous coal, lignite) to over 90%. Downloading at each node... After decryption, the local industrial world model was updated, and the anomaly detection accuracy improved by an average of 4.3%.

[0037] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0038] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0039] As can be seen from the above technical solution, this invention provides a full-stack embodied intelligent monitoring method for boilers based on domestically developed information technology. Compared with the prior art, this invention has the following advantages: It constructs a topology-enhanced dynamic feature flow by quantizing the connectivity of the dynamic feature flow using a 0th-order homology group and identifying holes in the dynamic feature flow using a 1st-order homology group. The dynamic feature flow, which integrates topological features, does not rely on a single pixel signal but selects effective features based on spatial structure correlation. This improves the accuracy of anomaly identification in dense boiler pipe areas and flue gas obstruction scenarios. By calculating the enhanced physical residual loss, the verification results are forced to conform to the laws of thermofluidity and energy conservation, greatly reducing the misjudgment rate of fluctuations in normal operating conditions. Furthermore, it selects strong dominant factors through the dominant root cause localization formula. The root cause analysis outputs verification results and a root cause report, quantifying the contribution of each candidate factor to directly pinpoint the core fault point and significantly reduce fault handling time. By considering node local benefits and node collaboration relationships, and based on the node contribution calculation formula, the model quantifies the node's contribution to the model. Using bee colony learning, a leader node is randomly elected in each training round. The leader node inputs the contributions of all nodes, and an updated industrial world model is obtained through an aggregation formula. This quantifies the sample quality and training contribution of each power plant, avoiding the drawback of high weight due to a large number of samples, and significantly improving the model's adaptability to boilers with different load rates and fuel types. Therefore, this invention effectively overcomes the various shortcomings of existing technologies and has high industrial application value.

[0040] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., a solid-state disk (SSD)).

[0041] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0042] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0043] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A full-stack intelligent monitoring method for boilers based on domestically developed information technology innovation, characterized in that: include: S1. By linking neuromorphic sensors with a digital twin engine, unstructured dynamic feature flow inside the boiler is collected. S2. Input the dynamic feature stream into the hybrid architecture of Vision Mamba and spiking neural network, dynamically adjust the system parameters using attention weights, prioritize memorizing the high-risk region features marked by the digital twin engine, and generate a sparse latent state tensor that integrates physical realism and virtual correlation. S3. Construct an industrial world model that integrates a digital twin engine and a causal inference model, map the sparse latent state tensor to a high-dimensional manifold space, verify the current physical state corresponding to the sparse latent state tensor, trace the causal relationship between anomalies and key events in the dynamic feature flow, and output the verification results and root cause report. S4. Receive the verification results and root cause report, integrate the virtual simulation data of the digital twin engine, and generate collaborative control instructions; S5. Using bee colony learning and fully homomorphic encryption technology, the industrial world model is decrypted and updated based on the verification results, the root cause report, and the collaborative control instructions. Based on the updated industrial world model, a boiler intelligent monitoring scheme is generated.

2. The boiler full-stack embodied intelligent monitoring method based on domestic information technology innovation as described in claim 1, characterized in that: The neuromorphic sensor in S1 collects dynamic feature streams and outputs (x,y,t,p) quadruple streams. It maps the microsecond-level dynamic feature streams to the boiler digital twin engine in real time, realizing data synchronization between physical devices and virtual mirrors. The digital twin engine automatically marks high-risk areas, quantifies the connectivity of the dynamic feature streams through the 0th-order homology group, and identifies holes in the dynamic feature streams through the 1st-order homology group, thus constructing a topology-enhanced dynamic feature stream. The homology group formula is as follows: ； ； in, It is a homology group of order 0. It is a first-order homology group. For dynamic feature flow, Boundary operators The core space, Boundary operators Image space, For topology-enhanced dynamic feature flow, as well as These are the weights of the 0th-order homology group and the 1st-order homology group, respectively.

3. The boiler full-stack embodied intelligent monitoring method based on domestic information technology innovation as described in claim 2, characterized in that: In S2, the spiking neural network transforms the topology-enhanced dynamic feature flow into a sparse pulse tensor, inputs the time series of the topology-enhanced dynamic feature flow, performs dynamic stability evaluation, and then calculates the enhanced mutual information between the topology-enhanced dynamic feature flow and the sparse pulse tensor. The dynamic stability evaluation formula is as follows: ； in, for The maximum Lyapunov exponent, quantizing the dynamic stability of the feature flow. For time variables, For time window, For the Vision Mamba model time Feature mapping function, For feature mapping function pairs The first derivative; The formula for enhancing mutual information is as follows: ； ； in, For sparse latent state tensors To enhance mutual information, quantification and The degree of correlation, For original mutual information, The chaos suppression coefficient is... To obtain The maximum value of 0.

4. The boiler full-stack embodied intelligent monitoring method based on domestic information technology innovation as described in claim 3, characterized in that: The verification of the current physical state corresponding to the sparse latent state tensor in S3 includes the following steps: S31. By constructing a physical state density matrix, the discrete sparse latent state tensor is transformed into an uncertain probability matrix. S32. Conduct a physical state uncertainty assessment, and the physical verification formula corrects the physical verification result by incorporating uncertainty. S33. Substitute the topological features of the input anomaly and dynamic feature flow into the order parameter change rate formula to calculate the change rate of each root cause. S34. Screen strong dominant root causes using the dominant root cause localization formula, and output the verification results and root cause report.

5. The full-stack embodied intelligent monitoring method for boilers based on domestically developed information technology, as described in claim 4, is characterized in that: The formula for the physical state density matrix is ​​as follows: ； The formula for assessing the uncertainty of the physical state is as follows: ； in, Let Y be the physical state density matrix. Let Y be the transpose of Y. For the trace operation of a matrix, For Feng Neumann entropy is used to quantify the uncertainty of a physical state; The physical verification formula is as follows: ； ； in, For the original physical residual loss, For physical constraint weights, ) is the PDE operator acting on As a result, The root cause correction coefficient. For enhanced physical residual loss; The formula for the rate of change of the order parameter is as follows: ； The formula for determining the dominant root cause is as follows: ； in, For the first Rate of change of order parameters of candidate root causes For the first The order parameter of each candidate root cause The optimized dominant root cause order parameter, The root cause correlation coefficient, For sparse latent state tensors under abnormal states. Assuming the first Anomalous feature tensor when candidate root causes are absent As an external driving force, This is the uncertainty correction coefficient.

6. The full-stack embodied intelligent monitoring method for boilers based on domestically developed information technology, as described in claim 5, is characterized in that: S4 includes the following steps: S41. Receive the verification result and root cause report, input the enhanced physical residual loss, and determine the control error; S42. Obtain the optimal PID parameters through the norm-based anti-interference optimization formula; S43. Finally, substitute the parameters into the fractional-order PID control law formula to calculate the control law and generate cooperative control instructions, which include DCS adjustment instructions and robot inspection instructions. The formula for calculating the control error is as follows: ； Among them, 0.04 is The normal threshold; The norm-based anti-interference optimization formula is as follows: ； ； ； in, To find the minimum value of the PID parameters, This is the proportionality coefficient. The integral coefficient is... These are the differential coefficients. For interference To control error The transfer function, To cause interference at the industrial site, For sliding surface functions, For fractional order, For integral weights; The formula for the fractional-order PID control law is as follows: ； in, For control laws, For fractional differential operators, For fractional integral operators, For sliding mode switching gain, It is a symbolic function.

7. The full-stack embodied intelligent monitoring method for boilers based on domestically developed information technology, as described in claim 6, is characterized in that: Updating the industrial world model in S5 includes the following steps: S51, each power plant edge node is based on local data , as well as Train the Vision Mamba model and the causal inference model to obtain local model parameters; S52. Based on the node's local revenue and node collaboration relationship, and according to the node contribution calculation formula, quantify the node's contribution to the model. S53. Each node generates a quantum key through the BB84 protocol and verifies qualified nodes according to the key rate formula and enhancement formula. S53. Based on bee colony learning, a leader node is randomly elected in each round of training. The leader node inputs the contribution of all nodes and obtains the updated industrial world model through the aggregation formula.

8. The full-stack embodied intelligent monitoring method for boilers based on domestic information technology innovation as described in claim 7, characterized in that: The formula for calculating the node contribution is as follows: in, For the first The contribution of each edge node. The damping coefficient is... For the first Local revenue of each edge node For the first The set of incoming edges of each edge node For the first The contribution of each edge node. For the first Local revenue of each edge node For the first The number of outgoing edges from each edge node; The key rate formula is as follows: ； The enhancement formula is as follows: ； in, This is the raw key rate for the BB84 protocol. For enhanced security key rate, For photon transmission probability, For detection efficiency, It is a binary entropy function. The original bit error rate. The bit error rate after error correction. This is the interference correction factor; The aggregation formula is as follows: ； in, For aggregation model parameters, The total number of edge nodes participating in the collaboration. For the first The result of encrypting the local model parameters of each edge node.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 8.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it causes the processor to perform the steps of the method as described in any one of claims 1 to 8.

Images