A differentiated data compliance verification and correction method suitable for multiple working conditions

Abstract

The application discloses a differentiated data compliance verification and correction method suitable for multiple working conditions, comprising the following steps: adopting an LLE algorithm to compress a historical data set to construct a constitutive manifold, and combining real-time monitoring to generate a benchmark model state value; constructing a sensor node alliance, calculating a confidence interval by using a Student's t-distribution, and combining a characteristic function to perform compliance determination on a measured value, and outputting compliant data; comparing the compliant data with the benchmark model state value to calculate a correction variable value; based on the correction variable value, solving a correction curve through a GENERIC framework, and verifying an output compliant correction curve by taking a degeneracy condition as a hard constraint; and according to time integration of the benchmark model state value and the correction curve, calculating a corrected model state value, taking the corrected model state value as a new data point to update the historical data set, and re-executing the LLE algorithm. The application improves the data compliance verification accuracy and the physical credibility of correction under multiple working conditions, and has a closed-loop self-evolution ability.

Description

Technical Field

[0001] This invention relates to the field of engineering monitoring technology, specifically to a method for verifying and correcting differentiated data compliance under multiple working conditions. Background Technology

[0002] With the rapid development of industrial IoT, smart construction, and digital monitoring technologies for infrastructure, engineering systems (such as bridges, tunnels, wind turbines, smart buildings, and industrial production lines) are gradually transforming towards a data-driven operation and maintenance management model. By deploying a large number of various types of sensors (such as strain gauges, temperature sensors, vibration accelerometers, and pressure sensors) in key areas, real-time perception of the structural status and environmental parameters can be achieved, providing a data foundation for safety assessment, fault diagnosis, and predictive maintenance.

[0003] Currently, some studies have utilized data dimensionality reduction techniques (such as Locally Linear Embedding, LLE) to process high-dimensional monitoring data, employed game theory methods for multi-agent coordination, or used the GENERIC framework to describe non-equilibrium thermodynamic processes. However, existing technologies lack a systematic solution that organically integrates these methods and specifically addresses the problem of "model correction and data verification in multiple engineering scenarios." Specifically, this manifests in the following technical deficiencies: The existing LLE methods lack the ability to generate benchmark models that are adaptive to multiple scenarios. They are usually only used for data visualization and feature dimensionality reduction, and do not combine the constitutive manifold they construct with the dynamic matching mechanism of the current engineering scenario. As a result, they cannot automatically generate benchmark models that are adapted to different working conditions based on real-time monitoring data, which leads to a significant increase in model prediction bias as the scenario changes.

[0004] There is a lack of differentiation and confidence quantification in data compliance verification. Existing cooperative game theory methods mostly focus on the formation of decision-making alliances and the distribution of benefits, and have not yet applied their alliance voting mechanisms to the compliance determination of multi-source sensor data. Furthermore, there is a lack of confidence interval construction methods based on small sample statistical theories (such as Student's t-distribution), making it difficult to quantify the reliability of data when the number of sensors is limited, resulting in insufficient accuracy in distinguishing between abnormal and compliant data.

[0005] Model correction processes are prone to violating fundamental physical laws. Existing data-driven model correction methods (such as neural network fitting and least squares identification) typically prioritize data fitting accuracy as the sole optimization objective, neglecting the potential violation of thermodynamic laws such as energy conservation and entropy increase during the correction process. Even with the introduction of the GENERIC framework, current applications are mostly forward modeling rather than data correction, lacking the technical means to use its degeneracy conditions as hard constraints to ensure the physical compliance of the correction results.

[0006] The system lacks closed-loop feedback and continuous learning capabilities. Existing methods are mostly offline batch processing modes, and successful correction experiences cannot be fed back into the benchmark model library. As a result, the system cannot use historical correction data to optimize itself. As the engineering system runs for a long time, the model library gradually becomes out of touch with actual working conditions, and the correction frequency and computational cost continue to rise.

[0007] Therefore, there is an urgent need for an intelligent correction method that can adapt to multiple engineering scenarios, achieve differentiated data compliance verification, ensure that the correction process conforms to physical laws, and has continuous learning capabilities, in order to solve the above-mentioned technical problems.

[0008] The above statements are for the purpose of providing background information in relation to this application only and do not necessarily constitute prior art. Summary of the Invention

[0009] The present invention proposes a differentiated data compliance verification and correction method that adapts to multiple working conditions, which can at least solve one of the technical problems in the background art.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: A differentiated data compliance verification and correction method adaptable to multiple operating conditions, comprising the following steps performed via computer equipment: S100 uses the LLE algorithm to compress historical datasets under different working conditions, constructs constitutive manifold functions, and combines real-time monitoring data vectors to generate benchmark model state values ​​adapted to a working condition. S200 constructs a sensor node alliance, calculates confidence intervals using Student's t-distribution, and uses feature functions to determine the compliance of each node's measurement values, then filters and outputs a compliant dataset. S300 determines the correction variable values ​​by comparing the output compliance dataset with the baseline model state values. S400, based on the corrected variable values, constructs a dynamic model to solve the corrected curve through the GENERIC framework, and sets the degeneracy condition of the GENERIC framework as a hard constraint to verify the output compliant corrected curve. S500: Based on the baseline model state value and the time integral of the compliance correction curve, the corrected model state value is calculated. The corrected model state value is then used as a new data point to update the historical dataset. The LLE algorithm is then re-executed to construct the updated constitutive manifold function.

[0011] Furthermore, the method for constructing the constitutive manifold function is as follows: S11a, collect historical datasets under different operating conditions. ,in, Represents a high-dimensional vector; S12a, based on historical datasets, determines each data point Find the nearest neighbors and solve for the reconstruction weights. The weights are calculated by minimizing the reconstruction error:

[0012] The constraints are:

[0013] In the formula, This represents the mathematical optimization objective of "minimizing". Represents the first in the historical dataset A high-dimensional data point, Indicates the reconstructed weights. Indicates the first The first high-dimensional data point One nearest neighbor data point; S13a, while maintaining the reconstruction weights Under the premise of invariance, solve for the low-dimensional embedded coordinates and construct the constitutive manifold function, specifically as follows:

[0014] In the formula, Indicates the first m The coordinate vectors of each data point in the low-dimensional embedding space. Indicates the first The coordinate vectors of the nearest neighbor data points in the low-dimensional embedding space.

[0015] Furthermore, the method for generating the baseline model state value under a certain working condition in step S100 is specifically as follows: S11b, obtain The real-time monitoring data vector at time [time] And based on the constitutive manifold function and its corresponding high-dimensional data points Calculate the position of the real-time monitoring data vector in the constitutive manifold function and determine the corresponding high-dimensional data points. of One nearest neighbor data point; S12b, based on weights Interpolate the baseline model parameters corresponding to the nearest neighbor data points to generate the baseline model state under the current operating condition. .

[0016] Furthermore, the specific method for filtering and outputting the compliant dataset is as follows: S201, arranges multi-sensor data into a grid according to spatial location to construct a monitoring data matrix. Simultaneously, the neighborhood is determined by Manhattan distance, and a node alliance is built. ; S202, Compute Node Alliance Average eigenvalues ​​of internal data The confidence interval for the absolute difference is calculated using Student's t-distribution, specifically as follows:

[0017] In the formula, This indicates the number of nodes in the node alliance. Indicates the significance level. Indicates the first A node alliance The confidence interval, Indicates a node alliance The average eigenvalue of the internal data, In Student's t-distribution value, Indicates a node alliance The measured data value of a certain node within the system; S203, based on confidence intervals, utilizes characteristic functions Perform compliance assessments and output a compliance dataset. ; The compliance determination operation is specifically as follows: Take the node alliance Each node data in With the confidence interval Perform a size comparison.

[0018] Furthermore, the corrected variable value The specific calculation formula is as follows:

[0019] In the formula, Represents the actual, compliant physical state. This represents the uncorrected baseline model state value.

[0020] Furthermore, in step S400, state variables are defined. The evolution calculation equations for the modified dynamic model are defined as follows:

[0021] In the formula, This indicates a correction to the state variable; The Poisson matrix is ​​used to describe the invertible behavior of a system. Represents the energy gradient; This represents the total energy of the system; This represents the friction matrix, used to describe the dissipation behavior of the system; Represents the entropy gradient; This represents the system entropy.

[0022] Furthermore, the solution for the correction curve is calculated as follows:

[0023] In the formula, Indicates at time Correct the state variable value, Indicates the initial time. Correct the state variable value, Represents the Poisson matrix. Represents the energy gradient. Represents the friction matrix. This represents the entropy gradient.

[0024] Furthermore, the degeneracy condition specifically refers to:

[0025]

[0026] That is, the first condition ensures that the reversible process does not generate entropy, and the second condition ensures that the dissipative process does not change the total energy.

[0027] Furthermore, the verification output of the compliance correction curve is specifically as follows: Complete formula:

[0028] Constraints:

[0029] In the formula, The compliance correction curve represents the rate of change of state variables that conforms to physical laws after verification. Indicates being bound by, The first degeneracy condition ensures that a reversible process does not generate entropy. This indicates the second degeneracy condition, which ensures that the dissipation process does not change the total energy.

[0030] Furthermore, the specific formula for calculating the corrected model state value is as follows:

[0031] In the formula, This represents the corrected model state value. Represents the state values ​​of the baseline model. This indicates a compliance correction curve.

[0032] 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.

[0033] 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.

[0034] As can be seen from the above technical solution, the beneficial effects are as follows: (1) This invention constructs a constitutive manifold function through the LLE algorithm and performs adaptive interpolation by combining real-time monitoring data, so that the state value of the benchmark model can be generated quickly and accurately as the working conditions change. This effectively overcomes the problem of increased prediction deviation of the traditional fixed model when switching between multiple scenarios and improves the model's adaptability to complex engineering environments. (2) This invention innovatively constructs a sensor node alliance and uses Student's t-distribution to calculate the confidence interval. Combined with feature function, it makes compliance judgment. It can reliably quantify data uncertainty under limited sensor and small sample conditions, significantly improve the discrimination accuracy and anti-interference ability of outliers in multi-source heterogeneous data, and ensure the accurate output of compliant datasets. (3) This invention introduces the GENERIC framework to establish a modified dynamic model and uses the degeneracy condition as a hard constraint to ensure that the evolution process of the modified variable strictly satisfies the basic thermodynamic laws such as energy conservation and entropy increase, thus avoiding the risk of non-physical pseudo-solutions generated by pure data-driven modification and ensuring the credibility and practicality of the modification results at the physical level. (4) The present invention feeds back the corrected model state value to the historical dataset and re-executes the LLE algorithm to update the constitutive manifold function, giving the system continuous learning and adaptive evolution capabilities, avoiding the problem of the offline model library gradually becoming invalid as the working conditions change, reducing the computational overhead and correction frequency in the long-term operation and maintenance process, and realizing efficient and sustainable data-driven closed-loop optimization. Attached Figure Description

[0035] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0036] 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.

[0037] like Figure 1As shown in the figure, the differentiated data compliance verification and correction method adapted to multiple working conditions described in this embodiment performs the following steps through a computer device: S100 uses the LLE algorithm to compress historical datasets under different working conditions, constructs constitutive manifold functions, and combines real-time monitoring data vectors to generate benchmark model state values ​​adapted to a working condition. S200 constructs a sensor node alliance, calculates confidence intervals using Student's t-distribution, and uses feature functions to determine the compliance of each node's measurement values, then filters and outputs a compliant dataset. S300 determines the correction variable values ​​by comparing the output compliance dataset with the baseline model state values. S400, based on the corrected variable values, constructs a dynamic model to solve the corrected curve through the GENERIC framework, and sets the degeneracy condition of the GENERIC framework as a hard constraint to verify the output compliant corrected curve. S500: Based on the baseline model state value and the time integral of the compliance correction curve, the corrected model state value is calculated. The corrected model state value is then used as a new data point to update the historical dataset. The LLE algorithm is then re-executed to construct the updated constitutive manifold function.

[0038] The following provides a detailed explanation of each step: Step S1: Using local linear embedding technology, high-dimensional historical monitoring data from different engineering scenarios are compressed to construct a low-dimensional constitutive manifold, which serves as a feature library for subsequent scene matching and benchmark model generation.

[0039] To adapt to the characteristics of multiple scenarios in engineering, we first construct constitutive manifolds covering different working conditions based on Local Linear Embedding (LLE) technology, and establish a scenario feature library. Locally Linear Embedding (LLE) is a nonlinear dimensionality reduction method. Its core idea is to assume that high-dimensional data is distributed on a low-dimensional manifold, and that each data point and its nearest neighbors approximately satisfy a linear relationship within a local region of the manifold. LLE first finds the k nearest neighbors for each data point in the high-dimensional space and solves for a set of reconstruction weights such that the point can be linearly reconstructed by its neighbors with minimal error. Then, while keeping these weights unchanged, it finds a new set of coordinates in the low-dimensional space, preserving the same linear reconstruction relationship as much as possible. In this way, LLE can effectively reveal the inherent geometric structure of high-dimensional data while preserving the local topological properties of the data, and is commonly used for tasks such as data visualization, feature extraction, and manifold learning.

[0040] Sub-step S101: Collect historical monitoring datasets for different engineering scenarios Each of them Represents a high-dimensional vector (containing state parameters such as temperature, stress, and strain); Sub-step S102: Based on the historical monitoring dataset, determine each data point Find the nearest neighbors and solve for the reconstruction weights. The weights are calculated by minimizing the reconstruction error:

[0041] The constraints are .

[0042] In the formula, This represents the mathematical optimization objective of "minimizing". Represents the first in the dataset A high-dimensional data point, Indicates the reconstructed weights. Indicates the first Data points ( ) Nearest neighbor data points.

[0043] Sub-step S103: While maintaining weights Under the premise of invariance, solve for the low-dimensional embedding coordinates and construct the constitutive manifold:

[0044] In the formula, Indicates the first m The coordinate vectors of each data point in the low-dimensional embedding space. Indicates the first The coordinate vectors of the nearest neighbor data points in the low-dimensional embedding space.

[0045] Step S2: By mapping the current real-time monitoring data onto the constitutive manifold constructed in step S1, and using LLE weights to interpolate the baseline model parameters of the nearest neighbor points, a baseline model state adapted to the current engineering scenario is generated.

[0046] Sub-step S201: Based on the constitutive manifold and its corresponding high-dimensional data points, receive the current time step. Engineering monitoring data vector Calculate the position of the engineering monitoring data vector in the low-dimensional manifold, and find its position on the manifold. One nearest neighbor data point; Sub-step S202: Utilize the LLE weights obtained in step 1 Interpolate the baseline model parameters corresponding to the nearest neighbor data points to generate the baseline model state for the current scene. This ensures the compatibility of the baseline model with the current engineering scenario.

[0047] Specifically, the baseline model parameters refer to the set of vectors that describe the basic physical state of an engineering system under specific working conditions. Its specific composition consists of characteristic physical quantities or state variables (such as stress and strain, ambient temperature, vibration frequency, structural displacement, etc.) monitored by various types of sensors deployed in key locations.

[0048] In the constitutive manifold space, the baseline model parameters are used as high-dimensional data points to participate in LLE dimensionality reduction calculation. That is, each low-dimensional embedded coordinate on the manifold is uniquely mapped to and associated with a set of baseline model parameter vectors containing the real physical state.

[0049] The initial method for obtaining the baseline model parameters is to collect actual data through a multi-sensor network and determine them through physical model calibration such as finite element method during the early stage of engineering system construction or under specific known stable operating conditions. Subsequently, the parameters are used as an important part of the historical dataset to participate in the construction and iterative update of the constitutive manifold.

[0050] Step S3: By constructing a sensor node alliance, calculating the confidence interval based on Student's t-distribution, and using the characteristic function to determine the compliance of the measurement value of each node, a set of compliant data is selected from the multi-source sensor data and output as the actual measurement value representing the real physical state.

[0051] For multi-source sensor data in the current scenario, a cooperative game theory algorithm is adopted to identify data compliance and distinguish between normal and abnormal data through a coalition voting mechanism.

[0052] Cooperative game theory (or cooperative game algorithm) is an important branch of game theory that studies game scenarios where participants can reach binding cooperative agreements, form alliances, and achieve greater gains through collaboration. In contrast, non-cooperative game theory studies the decision-making behavior of participants when they cannot cooperate and each pursues their own self-interest.

[0053] Sub-step S301: Construct the monitoring data matrix (Arrange multi-sensor data into a grid according to spatial location), determine the neighborhood based on Manhattan distance, and build a node alliance. ; Sub-step S302: Calculate the average eigenvalue of the data within the node consortium. And the absolute difference, and construct confidence intervals based on Student's t-distribution. :

[0054] In the formula, Indicates the number of alliance nodes. Indicates the significance level (e.g., 1%). Indicates the first Alliance The confidence interval, Indicates alliance The average eigenvalue of the internal data, In Student's t-distribution t value, Indicates alliance The measured data value of a certain node within the system; The Student's t-distribution, also known as the student t-distribution, is an important probability distribution in statistics. It was named after William Sealy Gosset, who published it in 1908 under the pseudonym "Student." It is primarily used to handle statistical inference problems involving small samples, especially when the population standard deviation is unknown and the sample size is small (usually less than 30), to estimate the mean of a normally distributed population. The t-distribution resembles the standard normal distribution, exhibiting a unimodal, symmetrical bell-shaped curve, but its tails are thicker. This means that at the same confidence level, the t-distribution produces a wider confidence interval, thus more conservatively reflecting the uncertainty of small sample estimates. As the sample size (i.e., degrees of freedom) increases, the t-distribution gradually approaches the standard normal distribution. In practical applications, the t-distribution is widely used to construct confidence intervals, perform t-tests (such as comparing whether there is a significant difference between the means of two groups), and test the significance of regression coefficients. Average eigenvalues The calculation method is for the alliance All V The measurement values ​​from each sensor node are taken as an arithmetic average, that is, the measurement values ​​from each node are first averaged. Add them together and then divide by the total number of nodes. V The absolute difference in the formula actually refers to the sum of squared deviations between the measured value at each node and the average value. It is calculated by first calculating the sum of squared deviations of each node's measured value from the average value. and The difference is calculated, then each difference is squared, and finally the squared values ​​of all nodes are summed. This sum of squares is used to calculate the standard deviation of the data within the consortium, and then combined with Student's t-distribution to construct confidence intervals, thereby quantifying the dispersion of the data within the consortium and identifying outlier nodes; Sub-step S303: Based on the confidence interval, use the characteristic function Perform compliance assessments and output a compliance dataset. : For the alliance Data from each sensor node in Compare it with the confidence interval calculated in sub-step S302. Compare them. Characteristic functions. The definition is usually: if the measured value of this node Falling within the confidence interval If the value is within the range (i.e., greater than or equal to the lower limit of the interval and less than or equal to the upper limit of the interval), then the data at that node is considered compliant data, and the feature function output value is 1; if... Data falling outside the confidence interval is considered non-compliant (abnormal), and the feature function output value is 0. Then, all sensor nodes in the current scene are traversed, and all data marked as compliant (…) are collected. u The node data (=1) is aggregated and output as a compliance data set. .Should This will be used as the actual measurement value representing the true physical state in subsequent steps, and compared with the baseline model to calculate the correction target.

[0055] Step S4: Compare the scene baseline model generated in step S2. Compliance data verified in step S3 Calculate the deviation and determine the target that needs to be corrected.

[0056] Sub-step S401: Calculate the correction variable :

[0057] In the formula, Represents the actual, compliant physical state. This represents the uncorrected baseline model prediction. The purpose is to quantify the gap between the model and actual compliance data.

[0058] Step S5: Use the GENERIC framework to perform dynamic modeling of the correction variables. Solve for the correction curve by defining state variables and adjusting the friction matrix, so as to ensure that the physical laws of energy conservation and entropy increase of the system are not violated while correcting the data.

[0059] In order to correct the data without destroying the physical laws of the system, the GENERIC framework is used to perform dynamic modeling of the correction amount calculated in step S4.

[0060] The GENERIC framework (General Equation for Non-Equilibrium Reversible-Irreversible Coupling) is a mathematical framework for constructing the evolution equations of non-equilibrium thermodynamic systems. Its core idea is to decompose the dynamic evolution of complex systems into the superposition of two independent contributions: one derived from the Poisson matrix. The reversible (conservative) part described corresponds to the energy-driven dissipative motion of the system, such as Hamiltonian dynamics in mechanics; the second is determined by the friction matrix. The irreversible (dissipative) portion described corresponds to the entropy-increasing process of the system, such as viscous diffusion and heat conduction. To ensure that the evolutionary process conforms to the fundamental laws of thermodynamics, the GENERIC framework mandates two degeneracy conditions: the reversible portion must not generate entropy (…). The dissipated portion does not change the total energy. This framework is widely used in fields such as complex fluid dynamics, polymer dynamics, and non-equilibrium thermodynamics modeling, and can ensure that the constructed macroscopic or mesoscopic models strictly satisfy the laws of energy conservation and entropy increase in mathematics; Sub-step S501: Define state variables Define the evolution equations for the modified dynamics:

[0061] In the formula, This indicates a correction to the state variable. The Poisson matrix represents the reversible (conservative) behavior of a system. Represents the energy gradient. Represents the total energy of the system. The friction matrix represents the dissipative (irreversible) behavior of the system. Represents the entropy gradient. Represents system entropy; Sub-step S502: Adjust the friction matrix based on the quality characteristics of the compliance data in step S3. This allows us to reflect the dissipation characteristics of the data (such as the gradual error of the data) and thus solve for the specific correction curve.

[0062] The solution for the correction curve is calculated as follows:

[0063] In the formula, Indicates at time Correct the state variable value, Indicates the initial time. Correct the state variable value, Represents the Poisson matrix. Represents the energy gradient. Represents the friction matrix. This represents the entropy gradient.

[0064] Step S6: By applying the degeneracy condition of the GENERIC framework as a hard constraint, verify and force the adjustment of the Poisson matrix and the friction matrix to ensure that the correction process satisfies the first law of thermodynamics (energy conservation) and the second law (entropy increase), and finally output a physically compliant correction curve.

[0065] The specific verification output of the compliance correction curve is as follows:

[0066] Constraints:

[0067]

[0068] In the formula, The compliance correction curve represents the rate of change of state variables that conforms to physical laws after verification. Indicates being bound by, The first degeneracy condition ensures that a reversible process does not generate entropy. This indicates the second degeneracy condition, which ensures that the dissipation process does not change the total energy.

[0069] To ensure that the corrected result obtained in step S5 does not violate the laws of engineering physics, the degeneracy condition of the GENERIC framework is applied as a hard constraint.

[0070] Sub-step S601: Based on the correction curve, Poisson matrix, and friction matrix in step S5, verify and enforce the degeneracy condition to obtain the compliance verification result; The degeneracy condition is as follows:

[0071]

[0072] The first condition ensures that a reversible process does not produce entropy (the second law of thermodynamics), and the second condition ensures that a dissipative process does not change the total energy (the first law of thermodynamics).

[0073] Sub-step S602: Based on the compliance verification results, if the compliance verification results violate the above constraints, then adjust... or The parameters ensure that the correction process converges while satisfying the fundamental laws of physics, ultimately yielding a compliance correction curve that meets physical compliance constraints.

[0074] Step S7: The cumulative correction amount after physical compliance constraints is superimposed on the baseline model to output the final correction result. This result is then used as a new data point to update the historical dataset. The LLE algorithm is then re-executed to construct the updated constitutive manifold, thereby achieving a closed-loop feedback update between the correction output and the manifold library.

[0075] Sub-step S701: Based on the compliance correction curve, obtain the baseline model state of the current scenario from step S2. Obtain the correction curve that satisfies physical compliance constraints from step S6. Plot the correction curve against time. Integrate to obtain the cumulative correction. Add the baseline model state to the cumulative correction to calculate the final corrected model state, and use this as the final output of the system.

[0076] In the formula, This represents the final corrected model state. Indicates the state of the baseline model. Indicates the compliance correction curve; Sub-step S702: The final corrected result output by sub-step S701. As a new high-dimensional data point, it is added to the historical dataset of step S1. In this process, an expanded dataset is formed. .

[0077] Based on this expanded dataset, the LLE algorithm is re-executed: the determination of each data point is then performed. The nearest neighbors are used to recalculate and reconstruct the weights. Then resolve the low-dimensional embedding coordinates. This allows for the construction of an updated constitutive manifold. Thus, when the system encounters a similar scenario again, the updated manifold provides a more accurate baseline model, reducing the amount of subsequent corrections required.

[0078] 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.

[0079] 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.

[0080] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the differentiated data compliance verification and correction methods adapted to multiple working conditions in the above embodiments.

[0081] It is understood that the systems, devices, and storage media provided in the embodiments of the present invention correspond to the methods provided in the embodiments of the present invention, and the explanations, examples, and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0082] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially 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., solid state disk (SSD)).

[0083] 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.

[0084] 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.

[0085] 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 method for differentiated data compliance verification and correction adaptable to multiple working conditions, characterized in that, Perform the following steps using a computer device: S100 uses the LLE algorithm to compress historical datasets under different working conditions, constructs constitutive manifold functions, and combines real-time monitoring data vectors to generate benchmark model state values ​​adapted to a working condition. S200 constructs a sensor node alliance, calculates confidence intervals using Student's t-distribution, and uses feature functions to determine the compliance of each node's measurement values, then filters and outputs a compliant dataset. S300 determines the correction variable values ​​by comparing the output compliance dataset with the baseline model state values. S400, based on the corrected variable values, constructs a dynamic model to solve the corrected curve through the GENERIC framework, and sets the degeneracy condition of the GENERIC framework as a hard constraint to verify the output compliant corrected curve. S500: Based on the baseline model state value and the time integral of the compliance correction curve, the corrected model state value is calculated. The corrected model state value is then used as a new data point to update the historical dataset. The LLE algorithm is then re-executed to construct the updated constitutive manifold function.

2. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 1, characterized in that, The method for constructing the constitutive manifold function is as follows: S11a, collect historical datasets under different operating conditions. ,in, Represents a high-dimensional vector; S12a, based on historical datasets, determines each data point Find the nearest neighbors and solve for the reconstruction weights. The weights are calculated by minimizing the reconstruction error: The constraints are: In the formula, The mathematical optimization objective is "minimization". Represents the first in the historical dataset A high-dimensional data point, Indicates the reconstructed weights. Indicates the first The first high-dimensional data point One nearest neighbor data point; S13a, while maintaining the reconstruction weights Under the premise of invariance, solve for the low-dimensional embedded coordinates and construct the constitutive manifold function, specifically as follows: In the formula, Indicates the first m The coordinate vectors of each data point in the low-dimensional embedding space. Indicates the first The coordinate vectors of the nearest neighbor data points in the low-dimensional embedding space.

3. The method for differentiated data compliance verification and correction adapted to multiple working conditions as described in claim 2, characterized in that, The specific method for generating the baseline model state value under a certain working condition in step S100 is as follows: S11b, obtain The real-time monitoring data vector at time [time] And based on the constitutive manifold function and its corresponding high-dimensional data points Calculate the position of the real-time monitoring data vector in the constitutive manifold function and determine the corresponding high-dimensional data points. of One nearest neighbor data point; S12b, based on weights Interpolate the baseline model parameters corresponding to the nearest neighbor data points to generate the baseline model state under the current operating condition. .

4. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 3, characterized in that, The specific method for filtering and outputting the compliant dataset is as follows: S201, arranges multi-sensor data into a grid according to spatial location to construct a monitoring data matrix. Simultaneously, the neighborhood is determined by Manhattan distance, and a node alliance is built. ; S202, Compute Node Alliance Average eigenvalues ​​of internal data The confidence interval for the absolute difference is calculated using Student's t-distribution, specifically as follows: In the formula, This indicates the number of nodes in the node alliance. Indicates the significance level. Indicates the first A node alliance The confidence interval, Indicates node alliance The average eigenvalue of the internal data, In Student's t-distribution value, Indicates a node alliance The measured data value of a certain node within the node; S203, based on confidence intervals, utilizes characteristic functions Perform compliance assessments and output a compliance dataset. ; The compliance determination operation is specifically as follows: Take the node alliance Each node data in With the confidence interval Perform a size comparison.

5. The method for adapting to differentiated data compliance verification and correction under multiple working conditions as described in claim 4, characterized in that: The corrected variable value The specific calculation formula is as follows: In the formula, Represents the true and compliant physical state. This represents the uncorrected baseline model state value.

6. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 5, characterized in that: The state variables are defined in step S400. The evolution calculation equations for the modified dynamic model are defined as follows: In the formula, This indicates a correction to the state variable; The Poisson matrix is ​​used to describe the invertible behavior of a system. Represents the energy gradient; This represents the total energy of the system; This represents the friction matrix, used to describe the dissipation behavior of the system; Represents the entropy gradient; This represents the system entropy.

7. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 6, characterized in that: The correction curve is calculated as follows: In the formula, Indicates at time Correct the state variable value, Indicates the initial time. Correct the state variable value, Represents the Poisson matrix. Represents the energy gradient. Represents the friction matrix. This represents the entropy gradient.

8. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 7, characterized in that: The degeneracy condition is specifically as follows: That is, the first condition ensures that the reversible process does not generate entropy, and the second condition ensures that the dissipative process does not change the total energy.

9. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 8, characterized in that: The specific verification output of the compliance correction curve is as follows: Complete formula: Constraints: In the formula, The compliance correction curve represents the rate of change of state variables that conforms to physical laws after verification. Indicates being bound by, The first degeneracy condition ensures that a reversible process does not generate entropy. This indicates the second degeneracy condition, which ensures that the dissipation process does not change the total energy.

10. The method for differentiated data compliance verification and correction adaptable to multiple working conditions as described in claim 9, characterized in that: The specific formula for calculating the corrected model state value is as follows: In the formula, This represents the corrected model state value. Represents the state values ​​of the baseline model. This indicates a compliance correction curve.

Images