A method and system for detecting the operating status of a multi-parameter detection device

By using a comprehensive anomaly index detection method combining time-varying probability benchmarks and directed nonlinear coupled networks, the problem of difficulty in identifying early weak anomalies in multi-parameter equipment is solved, and efficient and accurate equipment status monitoring is achieved.

CN123116069APending Publication Date: 2026-11-17ZHONGTONG WEIYI TECH SERVICE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611288815.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In the existing technology, early weak anomalies of multi-parameter equipment are difficult to identify effectively and have a high false alarm rate. This is mainly because the directed nonlinear coupling relationship between parameters is not fully utilized and the static judgment criteria cannot adapt to the dynamic changes in the operating status of the equipment.

Method used

By calculating the individual anomaly probability based on a time-varying probability benchmark, a directed nonlinear coupling network is constructed to generate a coupling anomaly index. This index is then nonlinearly aggregated to generate a comprehensive anomaly index, which adaptively tracks changes in equipment parameters and captures the dynamic evolution of coupling relationships.

Benefits of technology

It reduces the false alarm rate, improves the ability to identify early subtle anomalies, balances detection efficiency and accuracy, and enables accurate judgment of equipment operating status.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116069A_ABST
    Figure CN123116069A_ABST
Patent Text Reader

Abstract

This application discloses a method and system for detecting the operating status of a multi-parameter detection device, relating to the field of equipment operating status monitoring technology. It solves the technical problem of existing technologies where early weak anomalies in multi-parameter devices are difficult to effectively identify and have a high false alarm rate. The method includes: calculating the individual anomaly probability of each parameter based on the current observed values ​​of multiple operating parameters of the device under test, using a time-varying probability benchmark corresponding to each parameter; determining a time-varying coupling network based on the directed nonlinear coupling strength between the multiple operating parameters; extracting the network topology features of the time-varying coupling network; generating a coupling anomaly index based on the network topology features; nonlinearly aggregating the largest individual anomaly probability among the multiple operating parameters with the coupling anomaly index to generate a comprehensive anomaly index; and determining the operating status of the device under test based on the comprehensive anomaly index, thus achieving operating status detection that balances detection efficiency and accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application belongs to the field of equipment operation status monitoring technology, specifically a method and system for detecting the operation status of a multi-parameter detection device. Background Technology

[0002] The operational status monitoring of multi-parameter equipment typically relies on the collection and analysis of multiple operating parameters. Currently, the status monitoring methods in related technologies mainly employ single-parameter threshold comparison or simple multi-parameter listing monitoring. Single-parameter threshold comparison only focuses on whether each parameter exceeds a preset fixed upper or lower limit, failing to reflect the correlation between parameters; while simple multi-parameter listing monitoring collects multiple parameters simultaneously, each parameter is still processed independently, and the coupling information between parameters is not effectively utilized.

[0003] In actual operation, there are directed nonlinear coupling relationships between the operating parameters of multi-parameter devices. For example, changes in electrical parameters can drive the response of thermal parameters, and abnormalities in mechanical parameters may be transmitted through the coupling network and affect communication parameters. When a device exhibits early, weak anomalies, the deviation of a single parameter is often small and has not yet reached a fixed threshold, but the coupling topology between parameters may have already undergone perceptible changes. Because existing technologies fail to fully utilize the directed nonlinear coupling relationships between parameters and the judgment criteria are static, they cannot adapt to the dynamic changes in the operating status of the device, resulting in a high false alarm rate and difficulty in effectively identifying early, weak anomalies. Summary of the Invention

[0004] This application provides a method and system for detecting the operating status of a multi-parameter detection device, which solves the problem that existing technologies have difficulty in effectively identifying early weak anomalies in multi-parameter devices and have a high false alarm rate. This application uses a method and system for detecting the operating status of a multi-parameter detection device to quantify the individual anomaly probability of each parameter based on a time-varying probability benchmark. At the same time, it constructs a time-varying coupling network based on the directed nonlinear coupling strength between parameters, generates a coupling anomaly index according to the changes in network topology characteristics, and then nonlinearly aggregates the maximum individual anomaly probability and the coupling anomaly index to obtain a comprehensive anomaly index, thereby achieving operating status detection that balances detection efficiency and accuracy.

[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, a method for detecting the operating status of a multi-parameter detection device is provided, including: Based on the current observation values ​​of multiple operating parameters of the device under test, the individual anomaly probability of each parameter is calculated using the time-varying probability benchmark corresponding to each parameter. A time-varying coupling network is determined based on the directed nonlinear coupling strength between the multiple operating parameters, the network topology features of the time-varying coupling network are extracted, and a coupling anomaly index is generated based on the network topology features. The maximum individual anomaly probability among the multiple operating parameters is nonlinearly aggregated with the coupled anomaly index to generate a comprehensive anomaly index. The operating status of the tested equipment is determined based on the comprehensive anomaly index.

[0006] In the above scheme, the individual anomaly probability of each parameter is calculated based on a time-varying probability benchmark rather than relying on a fixed threshold, thus enabling adaptive tracking of the slow drift of equipment parameters and changes in operating conditions. Meanwhile, the coupling anomaly index is determined based on the topological characteristics of the time-varying coupling network generated by the directed nonlinear coupling strength between parameters, which can capture the directionality, strength, and dynamic evolution of the coupling relationship between parameters. The comprehensive anomaly index uses a nonlinear aggregation method to jointly determine individual anomalies and coupling anomalies. When the coupling topology has become abnormal but the deviation of individual parameters is not yet obvious, the comprehensive anomaly index can be significantly increased, thereby reducing the false alarm rate and improving the ability to identify early weak anomalies.

[0007] In conjunction with the first aspect above, in one possible implementation, based on the current observed values ​​of multiple operating parameters of the device under test, and using the time-varying probability benchmark corresponding to each parameter, the individual anomaly probability of each parameter is calculated, including: Based on historical normal operation data within the first sliding time window, the probability distribution parameters of each parameter are updated; based on the position of the current observation value in the corresponding probability distribution, the individual anomaly probability of each parameter is obtained by a two-sided calculation method; wherein, the two-sided calculation method assigns a consistent anomaly measure to parameters that are too high and parameters that are too low.

[0008] In the above scheme, the probability distribution parameters are continuously updated based on historical normal operation data within the first sliding time window, so that the probability benchmark can be dynamically adjusted according to the changes in the equipment's operating status; the two-sided calculation method covers both abnormal directions of parameter high and parameter low, avoiding missed detection caused by one-way detection.

[0009] In conjunction with the first aspect above, in one possible implementation, a time-varying coupled network is determined based on the directed nonlinear coupling strength among the plurality of operating parameters, network topology features of the time-varying coupled network are extracted, and a coupling anomaly index is generated based on the network topology features, including: The directional nonlinear coupling strength between each pair of parameters in the plurality of operating parameters is quantified by transfer entropy; the time-varying coupling network is obtained by using the plurality of operating parameters as nodes and the transfer entropy value as the edge weight; the network topology features of the time-varying coupling network are extracted; and a coupling anomaly index is generated based on the network topology features.

[0010] In the above scheme, the transfer entropy is asymmetric, which can reveal the direction of coupling between parameters, and based on information theory, it can capture any form of nonlinear dependence, thus more accurately reflecting the directionality and strength of the coupling relationship between parameters.

[0011] In conjunction with the first aspect above, in one possible implementation, the highest individual anomaly probability among the plurality of operating parameters is nonlinearly aggregated with the coupled anomaly index to generate a comprehensive anomaly index, including: The comprehensive anomaly index is generated by multiplicative nonlinear fusion using the largest individual anomaly probability among the plurality of operating parameters as the base term and the monotonically decreasing function corresponding to the coupled anomaly index as the penalty term; wherein the comprehensive anomaly index increases as the coupled anomaly index increases.

[0012] In the above scheme, the multiplicative nonlinear fusion avoids the subjectivity of weighted summation. The larger the coupling anomaly index, the smaller the penalty term, and the more the comprehensive anomaly index tends to be an anomaly extreme, thus realizing the collaborative judgment of individual anomalies and coupling anomalies.

[0013] In conjunction with the first aspect mentioned above, one possible implementation involves updating the probability distribution parameters of each parameter, including: The probability distribution corresponding to each parameter is set to a Gaussian distribution, and the probability distribution parameters are updated in real time through maximum likelihood estimation. For the i-th parameter, its probability density function is: : ; Where i is the number of the running parameter, For the observed value of the i-th parameter, Let be the sample mean of the i-th parameter within the first sliding time window. The sample variance of the i-th parameter within the first sliding time window; Based on the position of the current observation in the corresponding probability distribution, the individual anomaly probability of each parameter is obtained using a two-sided calculation method, including: Calculate the current observation value cumulative distribution function value : ; in, For integration variables; Calculate the individual abnormality probability of the i-th parameter based on the cumulative distribution function value. : ; in, Let be the individual anomaly probability of the i-th parameter at time t. Let i be the cumulative distribution function value of the current observation of the i-th parameter. This is for calculating the minimum value.

[0014] In the above scheme, the Gaussian distribution, as a specific form of probability distribution, has a small computational cost for parameter estimation and good real-time performance; the two-sided calculation method maps the observed value to the anomaly probability in the interval [0,1] through the cumulative distribution function value. When the observed value is too high or too low, the individual anomaly probability approaches 1, realizing an equivalent measure of two-way anomalies.

[0015] In conjunction with the first aspect above, in one possible implementation, extracting the network topology features of the time-varying coupled network includes: Extract at least one topological feature from the time-varying coupled network, including network coupling density, global transmission efficiency, and overall imbalance. The extracted topological features are combined into a topological feature vector; Generate a coupling anomaly index based on the network topology characteristics, including: Obtain the baseline topology feature vector; calculate the deviation between the current topology feature vector and the baseline topology feature vector. and the rate of change of the current topological feature vector over time. ; The coupling anomaly index is generated according to the multiplicative nonlinear structure: ; in, Let be the coupling anomaly index at time t. The deviation is mentioned. The rate of change, This represents the total historical normal operating time of the tested equipment. This is the rate of change adjustment coefficient. The sensitivity coefficient, It is the hyperbolic tangent function.

[0016] In the above scheme, network coupling density, global transmission efficiency, and overall imbalance characterize the topological features of the coupled network from three dimensions: quantity, quality, and directional structure of coupling, respectively. The coupling anomaly index adopts a multiplicative nonlinear structure with deviation and rate of change enhancement terms. Only deviation without change may indicate slow degradation, while only change without deviation may indicate a normal operating condition switch. Only when both exist simultaneously does it constitute a strong anomaly signal.

[0017] In conjunction with the first aspect above, in one possible implementation, the comprehensive anomaly index is generated through multiplicative nonlinear fusion, including: denoting the largest individual anomaly probability among the plurality of operating parameters as... The coupling anomaly index is denoted as ; Generate the comprehensive anomaly index: ; in, Let be the comprehensive anomaly index at time t. The highest individual anomaly probability among the multiple operating parameters. The coupling anomaly index is... It is an exponential function with the natural constant e as its base.

[0018] In the above scheme, The term maps the coupling anomaly index to a penalty coefficient in the range (0,1). The larger the coupling anomaly index, the smaller the penalty coefficient, and the closer the comprehensive anomaly index is to 1. When the maximum individual anomaly probability is 0 and the coupling anomaly index is 0, the comprehensive anomaly index is 0, indicating no anomaly. When the maximum individual anomaly probability is 1, the comprehensive anomaly index is 1, indicating that the comprehensive anomaly has reached its maximum.

[0019] In conjunction with the first aspect above, in one possible implementation, after generating the comprehensive anomaly index, the method further includes: When the comprehensive anomaly index is lower than the first preset threshold, the operating status of the tested device is determined to be normal. When the comprehensive anomaly index is between the first preset threshold and the second preset threshold, the device under test is subjected to fine classification diagnosis. When the comprehensive anomaly index is greater than or equal to the second preset threshold, trend prediction diagnosis is performed on the tested device. The outputs of different diagnostic layers are fused using Bayesian rules to obtain a comprehensive diagnostic result.

[0020] In the above scheme, different levels of diagnostic processes are triggered based on the threshold range of the comprehensive anomaly index. Under normal conditions, rapid screening is performed to save computing resources, while under abnormal conditions, in-depth diagnosis is performed gradually to improve accuracy. The outputs of each layer are fused through Bayesian rules to balance detection efficiency and diagnostic accuracy.

[0021] Secondly, this application provides a system for detecting the operational status of a multi-parameter detection device, comprising: The probabilistic characterization module is used to calculate the individual anomaly probability of each parameter based on the current observation values ​​of multiple operating parameters of the device under test and the time-varying probability benchmark corresponding to each parameter. The coupling analysis module is used to determine the time-varying coupling network based on the directed nonlinear coupling strength between the multiple operating parameters, extract the network topology features of the time-varying coupling network, and generate a coupling anomaly index based on the network topology features. The comprehensive diagnostic module is used to perform nonlinear aggregation of the largest individual anomaly probability among the multiple operating parameters and the coupled anomaly index to generate a comprehensive anomaly index, and to determine the operating status of the device under test based on the comprehensive anomaly index.

[0022] In the above system, the probabilistic characterization module, the coupling analysis module, and the comprehensive diagnosis module respectively carry out the functions of time-varying probability benchmark calculation, directed coupled network analysis, and nonlinear aggregation judgment. The three modules work together to achieve the same detection effect as the above method.

[0023] In conjunction with the second aspect above, in one possible implementation, the system further includes an adaptive update module, used to update the probability distribution parameters, baseline topological features, and diagnostic model when the comprehensive anomaly index is lower than a preset normal threshold within multiple consecutive sampling periods.

[0024] In the above scheme, the adaptive update module automatically triggers updates during the continuous normal operation of the equipment, so that the probability distribution parameters, baseline topology features and diagnostic models can continuously adapt to changes in the individual characteristics of the equipment and maintain detection accuracy.

[0025] This application provides a method and system for detecting the operating status of a multi-parameter detection device. It can replace the fixed threshold with a time-varying probability benchmark. The probability distribution parameters of each parameter are continuously updated based on historical normal operation data within the first sliding time window. It can adaptively track the slow drift of device parameters and changes in operating conditions, overcoming the defect of high false alarm rate of traditional static threshold method when the operating status of the device changes dynamically.

[0026] By quantifying the directed nonlinear coupling strength between parameters through entropy transfer, a time-varying directed weighted coupled network is constructed and network topology features are extracted. This allows for in-depth exploration of the nonlinear directed coupling relationship between parameters, capturing perceptible changes in early weak anomalies at the coupled topology level, and compensating for the shortcomings of single-parameter threshold comparison in early anomaly identification.

[0027] By performing a multiplicative nonlinear aggregation of the maximum individual anomaly probability and the coupling anomaly index, the collaborative determination of individual anomalies and coupling anomalies is achieved. When the deviation of individual parameters is not obvious but the coupling topology has become abnormal, the comprehensive anomaly index can be significantly increased, thereby improving the ability to identify early weak anomalies while reducing the false alarm rate.

[0028] By triggering a hierarchical and progressive state diagnosis based on a comprehensive anomaly index, rapid screening is performed under normal conditions, and in-depth diagnosis is conducted under abnormal conditions. The outputs of each layer are fused through Bayesian rules, balancing detection efficiency and diagnostic accuracy.

[0029] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the method steps for detecting the operating status of the multi-parameter detection equipment in this application; Figure 2 This is a schematic diagram of the module connection of the operating status detection system of the multi-parameter detection equipment in this application. Detailed Implementation

[0032] The technical solutions of this application will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0033] Example 1: Please see Figure 1This embodiment provides a method for detecting the operating status of a multi-parameter detection device. This method is used to detect the operating status of a device under test with multiple operating parameters. The device under test can be any device that generates multiple collectable parameters during operation, such as an energy meter, a communication integrated testing instrument, an industrial controller, or a power conversion device. These multiple operating parameters typically include at least two of the following: electrical parameters, thermal parameters, mechanical parameters, and communication parameters, such as voltage, current, power factor, internal temperature of the device, vibration acceleration, and bit error rate. This embodiment only describes the core upper-level process of the method and does not limit the specific probability distribution type, specific coupling strength quantification method, specific topological feature name, or specific aggregation formula.

[0034] Step S100: Based on the current observed values ​​of multiple operating parameters of the device under test, calculate the individual anomaly probability of each parameter using the time-varying probability benchmark corresponding to each parameter.

[0035] Specifically, during the operation of the device under test, multiple operating parameters are collected synchronously, and the values ​​of each parameter collected at the current moment constitute the current observation vector. Each operating parameter corresponds to a time-varying probability benchmark, which is used to describe the value distribution law of the corresponding parameter under normal operating conditions. Unlike fixed thresholds, the time-varying probability benchmark is not an unchanging judgment boundary, but a probability reference system that is dynamically adjusted according to the operating state of the device. For example, when the device experiences a slow parameter drift due to a change in operating conditions or ambient temperature, the time-varying probability benchmark can be updated to follow this drift, so that the reference range of the normal state is always consistent with the current actual operating state of the device.

[0036] Based on the position of the current observation in the time-varying probability benchmark, the individual anomaly probability of each parameter can be calculated. The individual anomaly probability is used to measure the degree to which the current value of a single parameter deviates from its normal distribution range. The closer the value is to 0, the closer it is to the normal state, and the closer the value is to 1, the greater the degree of deviation. There are many ways to implement the time-varying probability benchmark. For example, the probability distribution parameters can be estimated based on historical normal operation data within a sliding time window, or the probability benchmark can be constructed by using kernel density estimation, mixed distribution models, etc. This embodiment does not limit this.

[0037] Step S200: Determine the time-varying coupled network based on the directed nonlinear coupling strength between multiple operating parameters, extract the network topology features of the time-varying coupled network, and generate a coupling anomaly index based on the network topology features; Specifically, the multiple operating parameters of the device under test are not independent of each other, but have a directed nonlinear coupling relationship. Direction means that the coupling relationship has directionality, and the driving effect of parameter A on parameter B and the driving effect of parameter B on parameter A may be different in strength. Nonlinear means that the dependence between parameters cannot be fully described by a linear correlation coefficient, and there may be nonlinear characteristics such as threshold effect and saturation effect.

[0038] Based on the directed nonlinear coupling strength between parameter pairs, a time-varying coupled network can be determined. This network uses each operating parameter as a node and the coupling strength between parameters as edges, with the direction of the edges corresponding to the direction of the coupling effect. Since the coupling strength changes over time, the time-varying coupled network also evolves over time. Network topology features are extracted from the time-varying coupled network. These topology features are used to characterize the overall state of the coupled network at the structural level, such as the density of coupling relationships, the efficiency of information transmission in the network, and the balance of the driving and driven relationships between nodes in the network. Changes in topology features can reflect anomalies in the coupling structure. When an early fault occurs in a device, even if the deviation of a single parameter value is not obvious, the coupling structure between parameters may have already undergone perceptible changes, which will be reflected in the changes in topology features. A coupling anomaly index is generated based on the network topology features. The coupling anomaly index is used to measure the degree to which the topology of the coupled network deviates from the normal state.

[0039] Step S300: Nonlinearly aggregate the largest individual anomaly probability among multiple operating parameters with the coupled anomaly index to generate a comprehensive anomaly index.

[0040] Specifically, the maximum value is taken from the individual abnormal probabilities of each parameter calculated in step S100. This maximum value represents the degree of abnormality corresponding to the parameter that deviates most significantly from the normal state among all parameters. The maximum individual abnormal probability is nonlinearly aggregated with the coupling abnormality index obtained in step S200 to generate a comprehensive abnormality index. Nonlinear aggregation is different from linear weighted summation. Linear weighted summation requires manually setting the weight coefficients of each component. The selection of weights is subjective, and the components are independent of each other and cannot produce a synergistic effect. Nonlinear aggregation, on the other hand, uses a nonlinear combination method to produce a synergistic judgment effect between individual abnormalities and coupling abnormalities.

[0041] For example, when the individual anomaly probability of a certain parameter is only 0.2, which has not yet reached the alarm level corresponding to the traditional fixed threshold, if the coupling topology has changed significantly at this time, resulting in a high coupling anomaly index, then through nonlinear aggregation, the comprehensive anomaly index can be significantly increased, thereby triggering anomaly judgment. This collaborative judgment mechanism enables this embodiment to identify potential faults in the early stage when the deviation of individual parameters is not obvious, but the coupling structure has already shown signs of anomaly.

[0042] Step S400: Determine the operating status of the tested equipment based on the comprehensive anomaly index.

[0043] Specifically, the comprehensive anomaly index comprehensively reflects the synergistic effect of the deviation of individual parameters and the degree of anomaly in the coupling structure between parameters. The operating status of the tested equipment can be judged based on the comprehensive anomaly index by comparing it with preset judgment conditions, or by inputting the comprehensive anomaly index into the subsequent diagnostic process for further analysis. The judgment result can include normal status and abnormal status, and can also be further subdivided into different anomaly levels.

[0044] Through steps S100 to S400, this embodiment establishes a higher-level conceptual framework, encompassing everything from time-varying probability benchmarks to individual anomaly probabilities, from directed nonlinear coupling strength to time-varying coupled networks and coupled anomaly indices, and from nonlinear aggregation to comprehensive anomaly indices. Since this embodiment does not limit specific probability distribution types, specific coupling strength quantification methods, or specific aggregation formulas, the scope of protection of this higher-level framework is not limited by the replacement of specific implementation methods.

[0045] Example 2: This embodiment, based on embodiment 1, further explains the construction method of the time-varying probability benchmark and the calculation method of the individual abnormal probability in step S100.

[0046] In Example 1, the time-varying probability benchmark is described as a probability reference system that is dynamically adjusted according to the operating state of the device. This example provides a specific implementation method: based on the historical normal operation data within the first sliding time window, the probability distribution parameters of each parameter are updated.

[0047] Specifically, the first sliding time window is a data extraction interval that continuously slides along the time axis. The window length W is a configurable parameter, and the window contains historical normal operation data of the most recent W sampling points. Taking a sampling period of 1 second and W of 300 as an example, the first sliding time window covers 300 sampling points within 300 seconds before the current moment. Whenever new sampling data arrives, the window slides forward by one sampling period, the earliest sampling point is removed from the window, and the latest sampling point is included in the window. The probability distribution parameter is continuously updated based on the data currently contained in the window, enabling the probability benchmark to adaptively track the slow drift of equipment parameters and changes in operating conditions. The value of the window length W can be adjusted according to the rate of change of the parameters of the equipment under test. For equipment with slower parameter changes, a larger W value can be used to obtain a more stable statistical estimate, while for equipment with faster parameter changes, a smaller W value can be used to respond to environmental changes more quickly.

[0048] Based on the position of the current observation in the corresponding probability distribution, the individual anomaly probability of each parameter is obtained by bilateral calculation.

[0049] The meaning of bilateral calculation is that it assigns a consistent anomaly measure to both parameters that are too high and too low. Traditional one-way detection usually only focuses on whether the parameter exceeds the upper limit or only focuses on whether the parameter is below the lower limit. When the parameter deviates in the opposite direction, it may be missed. Bilateral calculation covers both deviation directions at the same time. No matter whether the current observation falls at the high end or the low end of the probability distribution, the individual anomaly probability can reflect the degree of deviation, avoiding the missed detection caused by one-way detection.

[0050] Based on the above intermediate layer implementation, this embodiment further provides a bottom layer implementation method: the probability distribution corresponding to each parameter is set as a Gaussian distribution, and the probability distribution parameters are updated in real time through maximum likelihood estimation.

[0051] For the i-th parameter, its probability density function is: : ; Where i is the number of the running parameter, For the observed value of the i-th parameter, Let be the sample mean of the i-th parameter within the first sliding time window. The sample variance of the i-th parameter within the first sliding time window; Gaussian distribution is composed of mean and variance Both parameters are completely determined. Mean Characterizing the central position of the normal range of parameter values, variance It characterizes the dispersion of the parameter's normal values. Within the first sliding time window, it is updated in real-time using maximum likelihood estimation. and The maximum likelihood estimation selects the parameter value that maximizes the probability of the sample data appearing within the window as the current estimate. For a Gaussian distribution, the analytical result of the maximum likelihood estimation is the arithmetic mean and sample variance of the samples within the window. It involves little computation and is suitable for real-time operation.

[0052] Based on the position of the current observation value in the corresponding probability distribution, the individual anomaly probability of each parameter is obtained using a two-sided calculation method, including: calculating the current observation value. cumulative distribution function value : ; Where u is the integration variable.

[0053] Cumulative distribution function value This means that under the current Gaussian distribution, the parameter value is less than or equal to the current observed value. The probability of; The value ranges from 0 to 1. The closer the value is to 0.5, the closer the current observation is to the center of the probability distribution. The closer the value is to 0 or 1, the further the current observation deviates from the center of the probability distribution.

[0054] Calculate the individual anomaly probability of the i-th parameter based on the cumulative distribution function value. : ; in, Let be the individual anomaly probability of the i-th parameter at time t. Let be the cumulative distribution function value of the current observation for the i-th parameter, and min{·} is the minimum value operation. When the observed value is high, Approaching 1, at this point Pick When the observed value is low, the probability of an individual being abnormal approaches 1, and the probability of an abnormal individual approaches 0. Approaching 0, at this point Pick Similarly, the probability of an individual being an anomaly also approaches 1. When the observed value happens to be located at the center of the probability distribution, , The probability of individual anomalies is 0. Therefore, the two-sided calculation method assigns a consistent anomaly measure to both high and low parameters, achieving an equivalent measure of bidirectional anomalies.

[0055] This embodiment also provides a comparative example to illustrate the advantages of a time-varying probability benchmark over a fixed threshold. If a fixed threshold [μ] is used... Using [3σ, μ+3σ] as the judgment standard, when the device parameters drift slowly, for example, the mean gradually drifts from 220.0 to 224.0, the fixed threshold cannot adaptively adjust. It still uses the initial μ=220.0 and σ=2.0 as the benchmark. In this case, the parameter with a normal value of 224.0 will be judged as deviating from the center by 2σ, and the probability of individual anomalies is overestimated, leading to an increased false alarm rate. The time-varying probability benchmark continuously updates μ and σ through a first sliding time window. The sample mean within the window will gradually adjust to around 224.0 as the parameters drift, ensuring that the probability benchmark always remains consistent with the current actual operating state of the device, thereby maintaining detection accuracy.

[0056] It should be noted that the Gaussian distribution is a specific choice of probability distribution. This embodiment uses the Gaussian distribution because of its advantages of low computational cost and good real-time performance in parameter estimation. In other embodiments, when the historical data of the parameters exhibits a multimodal distribution, the probability distribution can also adopt a mixture Gaussian distribution or other distribution forms, as long as the distribution parameters can be continuously updated using data within a sliding time window.

[0057] Example 3: Based on Example 1, this embodiment further explains the quantization method of the directed nonlinear coupling strength, the construction method of the time-varying coupling network, and the generation method of the coupling anomaly index in step S200.

[0058] In Example 1, the directed nonlinear coupling strength is described as a measure of the directional nonlinear dependency between parameters. This example provides a specific implementation method. The directed nonlinear coupling strength between parameter pairs among multiple operating parameters is quantified using transfer entropy.

[0059] Specifically, transfer entropy is an information-theoretic measure of coupling strength used to quantify the reduction in uncertainty of the future state of one parameter by the historical state of another. Transfer entropy is asymmetric; that is, the transfer entropy from parameter j to parameter i is usually not equal to the transfer entropy from parameter i to parameter j. This asymmetry reveals the direction of coupling, distinguishing between the driver and the responder. Furthermore, transfer entropy is calculated based on the conditional dependencies between probability distributions, not on linear correlation coefficients. Therefore, it can capture arbitrary forms of nonlinear dependencies, including coupling characteristics that linear methods cannot effectively describe, such as threshold effects and saturation effects. There are various ways to calculate transfer entropy, such as estimating the probability density function from data within a sliding time window based on kernel density estimation, or using nonparametric estimation methods such as histograms and nearest neighbor methods. This embodiment does not limit the specific methods used.

[0060] A time-varying coupled network is obtained by using multiple operating parameters as nodes and the transfer entropy value as the edge weight. Specifically, the N operating parameters of the device under test correspond to N nodes in the network. For any two parameters i and j, if the transfer entropy value from parameter j to parameter i is greater than a preset significance threshold, a directed edge is established in the network from node j to node i, and the weight of the edge is the corresponding transfer entropy value. Due to the asymmetry of transfer entropy, the edge from node j to node i and the edge from node i to node j are two independent edges, and their weights can be different. The significance threshold is set to filter statistically insignificant weak associations, making the resulting time-varying coupled network a directed weighted sparse network, avoiding the inclusion of a large number of spurious associations in subsequent analysis. Since the transfer entropy value changes with time, the network's edge set and edge weights also evolve with time; therefore, this network is a time-varying coupled network.

[0061] Extract the network topology features of time-varying coupled networks and generate a coupling anomaly index based on the network topology features; Specifically, time-varying coupled networks have a specific topology at each moment, and topological features are used to characterize the statistical properties of the structure as a whole. Unlike the weight of a single edge, topological features reflect the structural state at the network level. When the device is in normal operation, the topological features of the coupled network fluctuate around the baseline. When the device malfunctions, the coupling relationship between parameters changes, and the topological features deviate from the baseline. Based on the topological features, a coupling anomaly index is generated, which quantifies the structural changes at the topological level into a numerical indicator that can be fused with other anomaly measures.

[0062] Based on the above intermediate layer implementation, this embodiment further provides a bottom layer implementation method: extracting at least one topological feature from the time-varying coupled network, including network coupling density, global transmission efficiency, and overall imbalance, and combining the extracted topological features into a topological feature vector; Network coupling density reflects the density of coupling relationships. When the overall coupling relationship between the parameters of the device is strengthened or weakened, the number of effective directed edges in the network will change, and the network coupling density will increase or decrease accordingly. Network coupling density can be defined as the ratio of the number of effective directed edges in the current network to the maximum possible total number of directed edges, and the value is between 0 and 1.

[0063] Global transmission efficiency reflects the efficiency of information transmission in a coupled network. When there are many short path connections in the coupled network, information can be transmitted from one node to another in a few hops, resulting in high global transmission efficiency. When the coupled network becomes sparse or has local breaks, information transmission requires more hops or may not be possible, leading to a decrease in global transmission efficiency. Global transmission efficiency can be calculated based on the reciprocal of the shortest directed path length between any two nodes in the network.

[0064] The overall imbalance reflects the directional balance of information flow in a coupled network. For each node in the network, its weighted in-degree and weighted out-degree can be calculated separately. The weighted in-degree represents the sum of information transmitted from other parameters to this parameter, and the weighted out-degree represents the sum of information transmitted from this parameter to other parameters. When the weighted in-degree and weighted out-degree of each node in the network are generally close, the direction of information flow is relatively balanced. When the in-degree of some nodes is significantly greater than the out-degree or the out-degree is significantly greater than the in-degree, the network exhibits a directional imbalance. The overall imbalance can be calculated by summing the absolute values ​​of the difference between the weighted in-degree and weighted out-degree of each node.

[0065] The three topological features described above characterize the topological state of the coupled network from the three dimensions of quantity, quality, and directional structure of coupling, respectively, forming a multi-dimensional three-dimensional characterization of the coupling relationship. The extracted topological features are combined into a topological feature vector, for example, the network coupling density, global transmission efficiency, and overall imbalance are arranged in order into a three-dimensional vector. It should be noted that the selection of topological features is not limited to the above three dimensions. In other embodiments, other topological features that can reflect the structural state of the coupled network can also be extracted, such as the average clustering coefficient, network diameter, modularity, etc., as long as they can reflect the changes in the topological structure of the coupled network.

[0066] The process involves obtaining a baseline topology feature vector, calculating the deviation between the current topology feature vector and the baseline topology feature vector, and the rate of change of the current topology feature vector over time. Specifically, the baseline topology feature vector characterizes the coupled network topology of the device under a healthy baseline state. The baseline topology feature vector can be the average of the topology feature vectors at various points within the device's healthy baseline period, which can be the initial period after the device's first commissioning or its most recent successful calibration. The deviation measures the statistical distance between the current topology feature vector and the baseline topology feature vector; a larger deviation indicates a greater difference between the current coupled network topology and the healthy baseline. The deviation can be calculated using Mahalanobis distance, which considers the correlation between topology features and the differences in the dimensions of different features, providing a more reasonable measure of the deviation between multidimensional feature vectors compared to Euclidean distance. The rate of change measures the drastic change of the topology feature vector over time and can be calculated based on the difference between the topology feature vector at the current moment and the previous sampling moment.

[0067] Generate coupling anomaly indices based on multiplicative nonlinear structures: ; in, Let be the coupling anomaly index at time t. For deviation, For the rate of change, This represents the total historical normal operating time of the tested equipment. This is the rate of change adjustment coefficient. Let be the sensitivity coefficient, and tanh(·) be the hyperbolic tangent function. The above formula employs a multiplicative nonlinear structure. Basic deviation term, This is the rate-of-change enhancement term; the range of the rate-of-change enhancement term is [1, 1+]. When the rate of change When the value is 0, the hyperbolic tangent function is 0, the enhancement term is 1, and the coupling anomaly index equals the deviation; when the rate of change... As the value increases, the hyperbolic tangent function value gradually approaches 1, and the enhancing term gradually approaches 1+. The coupling anomaly index is amplified on top of the deviation. Used to normalize the rate of change to comparable dimensions. Controlling the rate of change to maximize the contribution of the coupling anomaly index. Nonlinear saturation rate that controls the effect of rate of change.

[0068] The meaning of multiplicative nonlinear structure lies in achieving coordinated judgment of deviation and rate of change; when there is only deviation without change, it may be a slow degradation process of equipment parameters. In this case, the coupling anomaly index is mainly determined by the deviation, and the contribution of the rate of change enhancement term is limited; when there is only change without deviation, it may be a temporary adjustment of the coupling structure caused by the switching of normal operating conditions of the equipment. In this case, the deviation is small, and the coupling anomaly index will not increase significantly; only when the deviation and rate of change exist at the same time, that is, when the coupled network topology deviates from the healthy baseline and is changing rapidly, will the coupling anomaly index be significantly amplified, constituting a strong anomaly signal.

[0069] This embodiment also provides a comparative example to illustrate the advantages of the coupling anomaly index compared to relying solely on individual anomaly probabilities. If only individual anomaly probabilities are relied upon without constructing a coupled network, when a device experiences early, weak anomalies, the individual anomaly probabilities of each parameter are typically between 0.2 and 0.4, not yet reaching the alarm level corresponding to a traditional fixed threshold. At this point, anomalies cannot be effectively identified solely from the perspective of individual parameter deviations. However, in the early, weak anomaly stage, the coupling topology between parameters may have undergone perceptible changes. For example, the propagation entropy values ​​between certain parameter pairs may have changed significantly, causing the network coupling density, global propagation efficiency, or overall imbalance to deviate from the baseline. Therefore, the coupling anomaly index... The index then rises. By nonlinearly aggregating the coupling anomaly index with the individual anomaly probability, the comprehensive anomaly index can be significantly increased, thereby identifying potential anomalies when individual parameter deviations are not yet obvious.

[0070] It should be noted that transfer entropy is a specific choice for quantifying the strength of directed nonlinear coupling. This embodiment uses transfer entropy because of its asymmetric and nonlinear capture capabilities. In other embodiments, other quantification methods can be used for the strength of directed nonlinear coupling, as long as the directionality and nonlinear dependence of the coupling relationship between parameters can be quantified.

[0071] Example 4: This embodiment further explains the specific implementation of nonlinear aggregation in step S300, based on embodiment 1.

[0072] In Example 1, nonlinear aggregation is described as a combination method that is different from linear weighted summation. This example provides a specific implementation method: taking the largest individual anomaly probability among multiple operating parameters as the base term, and the monotonically decreasing function corresponding to the coupling anomaly index as the penalty term, a comprehensive anomaly index is generated through multiplicative nonlinear fusion, wherein the comprehensive anomaly index increases as the coupling anomaly index increases.

[0073] Specifically, the maximum individual anomaly probability The maximum value among the individual anomaly probabilities of each parameter calculated in step S100 is taken, representing the degree of anomaly corresponding to the parameter that deviates most significantly from the normal state among all operating parameters. Coupling Anomaly Index From step S200, it represents the degree to which the inter-parameter coupled network topology deviates from the normal state. Multiplicative nonlinear fusion will... As a basic item, The corresponding monotonically decreasing function serves as the penalty term, and the two are combined through a multiplicative relationship. When As the value increases, the penalty term decreases monotonically, and the overall abnormality index increases accordingly and tends to be abnormally extreme.

[0074] The key difference between a multiplicative structure and a weighted summation structure lies in the fact that weighted summation requires manually setting the weight coefficients for each component. Weight selection relies on experience and is subjective; different maintenance personnel may provide different weight configurations, leading to different judgments for the same set of input data. A multiplicative structure, on the other hand, does not require manual weight setting. The multiplicative relationship between the base term and the penalty term is automatically adjusted by the coupling anomaly index. The stronger the coupling anomaly, the smaller the penalty term, and the closer the comprehensive anomaly index tends to 1, achieving collaborative judgment of individual anomalies and coupling anomalies.

[0075] Building upon the aforementioned intermediate layer implementation, this embodiment further presents a lower-level implementation. The highest individual anomaly probability among multiple operating parameters is denoted as... The coupling anomaly index is denoted as Generate a comprehensive anomaly index:

[0076] in, Let be the comprehensive anomaly index at time t. The highest individual anomaly probability among multiple operating parameters. Let exp(·) be the coupling anomaly exponent, and let exp(·) be the exponential function with the natural constant e as the base.

[0077] In this formula, The item will couple the anomaly index The mapping is a penalty coefficient within the range (0,1). When hour, The penalty coefficient is set to its maximum value of 1. In this case, the overall anomaly index is entirely determined by the probability of the largest individual anomaly. = ;when When it increases, As the value monotonically decreases and approaches 0, the penalty coefficient approaches 0. Approaching 0, The index approaches 1; therefore, the larger the coupling anomaly index, the smaller the penalty coefficient, the closer the comprehensive anomaly index is to 1, and the higher the degree of anomaly.

[0078] From the perspective of boundary conditions, when and hour, This indicates no abnormality. When At any time, regardless What value should be taken? =1 indicates that the overall anomaly has reached its maximum. When When it is between 0 and 1, Never less than ,and The larger, Exceeding The larger the magnitude, the more it reflects the amplifying effect of coupling anomalies on the comprehensive anomaly index.

[0079] It should be noted that, It is a specific choice of a monotonically decreasing function. This embodiment uses an exponential function because it is simple to calculate, its range naturally falls in the (0,1] interval, and it increases with... The characteristic of rapid decay is enhanced. In other embodiments, the penalty term can also be other forms of monotonically decreasing functions, as long as they satisfy the condition that they monotonically decrease with the increase of the coupling anomaly exponent and their value range falls within the interval (0,1].

[0080] Example 5: This embodiment, based on embodiment 1, further explains the diagnostic process after determining the operating status of the tested equipment according to the comprehensive anomaly index in step S400.

[0081] In Example 1, step S400 determines the operating status of the device under test based on the comprehensive anomaly index. This example provides a hierarchical and progressive diagnostic implementation method. After generating the comprehensive anomaly index, when the comprehensive anomaly index is lower than the first preset threshold, the operating status of the device under test is determined to be normal. When the comprehensive anomaly index is between the first preset threshold and the second preset threshold, the device under test is subjected to fine classification diagnosis. When the comprehensive anomaly index is greater than or equal to the second preset threshold, the device under test is subjected to trend prediction diagnosis. The outputs of different diagnostic layers are fused using Bayesian rules to obtain a comprehensive diagnostic result.

[0082] Specifically, the first and second preset thresholds are pre-defined judgment boundaries that divide the range of the comprehensive anomaly index into three segments. The first preset threshold is lower than the second preset threshold, and both can be set between 0 and 1. The specific values ​​can be configured based on the operating characteristics of the device under test, historical fault statistics, and maintenance strategies. For example, for devices that are more sensitive to false alarms, the first preset threshold can be set relatively high to allow more samples to enter the fine-grained classification and diagnosis stage; for devices that are more sensitive to false negatives, the first preset threshold can be set relatively low to make the determination of normal status more stringent.

[0083] When the comprehensive anomaly index is below the first preset threshold, it indicates that the deviation of individual parameters and the anomaly of the coupling structure are both at a low level, and the tested device is likely in normal operating condition. At this point, the detection result is directly output and the process returns to the next round of monitoring, without triggering subsequent fine-grained classification diagnosis and trend prediction diagnosis, thus saving computational resources. This layer can be called the initial diagnosis layer, and its function is to quickly screen a large number of normal samples, sending only samples with a comprehensive anomaly index reaching a certain level into a deeper diagnostic process.

[0084] When the comprehensive anomaly index falls between the first and second preset thresholds, it indicates that the device under test has shown signs of anomaly to a certain extent, but the degree of anomaly has not yet reached the severity level that requires immediate trend prediction. At this point, a fine-grained classification diagnosis is performed on the device under test. The current observation vector, coupled network information, and coupled anomaly index are input into a pre-trained classification model, which outputs the posterior probability of each fault type.

[0085] The current observation vector consists of the current observation values ​​of each operating parameter. The coupling network information describes the structural state of the coupling relationships between parameters at the current moment, and the coupling anomaly index describes the degree to which the coupling network topology deviates from its normal state. The classification model is a pre-trained machine learning model whose input is the three types of information mentioned above, and whose output is the posterior probability corresponding to each fault type. The posterior probability represents the likelihood that the tested device belongs to each fault type under the current observation data conditions, and the sum of the posterior probabilities of all fault types is 1. The specific type of classification model can be chosen from various options, such as a classification model based on a multilayer perceptron, a classification model based on a support vector machine, or a classification model based on a decision tree; this embodiment does not limit this choice. The training data for the classification model can come from various fault samples accumulated by the tested device during its historical operation, or from experimental data of the same model of equipment under different fault conditions.

[0086] When the comprehensive anomaly index is greater than or equal to the second preset threshold, it indicates that the device under test has experienced a relatively serious anomaly. It is necessary not only to determine the current fault type but also to predict the subsequent evolution trend of the fault. In this case, trend prediction diagnosis is performed on the device under test. The observation vector sequence from multiple consecutive time points is input into a pre-trained time-series prediction model to predict the observation vectors at future time points and calculate the estimated remaining effective operating time.

[0087] A sequence of observation vectors from multiple consecutive moments constitutes a time series sample, which records the dynamic changes of various operating parameters of the device under test (DUT) over a recent period. The time series prediction model predicts observation vectors for several future moments based on this historical sequence, and the prediction results reflect the future trends of each operating parameter. Based on the predicted future observation vectors, the comprehensive anomaly index prediction value for each future moment can be further calculated, and this can be used to determine when the DUT will reach an unacceptable anomaly level. The time interval from the current moment to that moment is the estimated remaining effective operating time. Various types of time series prediction models can be selected, such as those based on recurrent neural networks, convolutional neural networks, or state-space models; this embodiment does not limit this selection.

[0088] The outputs of different diagnostic layers are fused using Bayesian rules to obtain a comprehensive diagnostic result.

[0089] The principle of Bayesian fusion lies in the fact that the posterior probability is proportional to the product of the likelihood probability and the prior probability. In this embodiment, the likelihood probability originates from the posterior probability of each fault type output by the fine-grained classification and diagnosis layer, representing the probability of each fault type occurring under the current observed data conditions; the prior probability is determined by historical diagnostic results and trend predictions, representing the baseline probability of each fault type occurring before the current data is observed. The two probabilities are multiplied by Bayesian rules and normalized to obtain the fused posterior probability of each fault type, and the state with the highest posterior probability is taken as the comprehensive diagnostic result.

[0090] There are several ways to determine prior probabilities. For example, if the trend prediction diagnostic layer predicts that the health trajectory of the tested equipment will continue to decline, the prior probability corresponding to the fault type will increase accordingly; if the historical diagnostic results are consistently normal, the prior probability of normal operation will remain at a high level. In this way, Bayesian fusion integrates the rapid screening results of the initial diagnosis layer, the fault type judgment of the fine classification diagnostic layer, and the future evolution information of the trend prediction diagnostic layer into a unified probabilistic output, avoiding the fragmentation between the output results of each diagnostic layer.

[0091] It should be noted that the hierarchical progressive diagnostic framework in this embodiment only describes the triggering conditions and functional positioning of each diagnostic layer in a higher-level manner. The specific model type, training method, and input data structure used within each diagnostic layer can be selected and replaced according to the actual application scenario. For example, the fine classification diagnostic layer can use different classification model architectures, and the trend prediction diagnostic layer can use different time series prediction model architectures, as long as the input and output requirements of the corresponding diagnostic layer are met. This embodiment, through the hierarchical progressive diagnostic framework, quickly screens under normal conditions to save computational resources, gradually deepens diagnosis under abnormal conditions to improve accuracy, and fuses the outputs of each layer through Bayesian rules, thus balancing detection efficiency and diagnostic accuracy.

[0092] Example 6: Please see Figure 2 This embodiment provides a system for detecting the operational status of a multi-parameter detection device. This system is used to detect the operational status of a device under test (DUT) with multiple operational parameters. The DUT can be any device that generates multiple collectable parameters during operation, such as an energy meter, a communication integrated testing instrument, an industrial controller, or a power conversion device. The system includes three core modules: a probabilistic characterization module, a coupling analysis module, and a comprehensive diagnostic module.

[0093] The probabilistic characterization module is used to calculate the individual anomaly probability of each parameter based on the current observation values ​​of multiple operating parameters of the device under test and the time-varying probability benchmark corresponding to each parameter.

[0094] Specifically, the probabilistic characterization module receives current observations of multiple operating parameters of the device under test (DUT). These observations can be provided by an external data acquisition device or acquired by the probabilistic characterization module itself from the DUT's sensors or sampling channels via a data interface. Internally, the probabilistic characterization module maintains a time-varying probability benchmark for each parameter. This benchmark describes the distribution of values ​​for the corresponding parameter under normal operating conditions and is dynamically adjusted according to the device's operating status. Based on the position of the current observation in the time-varying probability benchmark, the probabilistic characterization module calculates the individual anomaly probability of each parameter. This individual anomaly probability measures the degree to which the current value of a single parameter deviates from its normal distribution range. The probabilistic characterization module can be implemented in various ways, such as estimating probability distribution parameters based on historical normal operating data within a sliding time window, or constructing a probability benchmark using kernel density estimation, mixed distribution models, etc. This embodiment does not limit the specific implementation. The probabilistic characterization module can be implemented as a software module, hardware circuit, or a combination of both, such as a probability calculation program running in an embedded processor or a probability calculation unit implemented in a programmable logic device.

[0095] The coupling analysis module is used to determine the time-varying coupled network based on the directed nonlinear coupling strength between multiple operating parameters, extract the network topology features of the time-varying coupled network, and generate a coupling anomaly index based on the network topology features.

[0096] Specifically, the coupling analysis module receives time-series data of multiple operating parameters of the device under test (DUT), which can come from the same data source as the probabilistic characterization module. The coupling analysis module quantifies the directed nonlinear coupling strength between parameter pairs. "Directed" means the coupling relationship has a directionality; the driving effect of parameter A on parameter B can differ in strength from the driving effect of parameter B on parameter A. "Nonlinear" means the dependency between parameters cannot be fully described by a simple linear correlation coefficient. Based on the quantified coupling strength, the coupling analysis module determines a time-varying coupling network. This network uses each operating parameter as a node and the coupling strength between parameters as edges, with the direction of the edges corresponding to the direction of the coupling effect. Since the coupling strength changes over time, the time-varying coupling network also evolves over time. The coupling analysis module extracts network topology features from the time-varying coupling network. These topology features characterize the overall state of the coupling network at the structural level, such as the density of coupling relationships, the efficiency of information transmission in the network, and the balance of driving and driven relationships among nodes. Based on the network topology features, a coupling anomaly index is generated, which measures the degree to which the coupling network topology deviates from the normal state. There are various ways to implement the coupling analysis module, such as using transfer entropy, directed mutual information, or nonlinear extensions of the Granger causality test to quantify the strength of directed nonlinear coupling. This embodiment does not limit this approach. The coupling analysis module can also be implemented as a software module, a hardware circuit, or a combination of both.

[0097] The comprehensive diagnostic module is used to nonlinearly aggregate the highest individual anomaly probability among multiple operating parameters with the coupled anomaly index to generate a comprehensive anomaly index, and to determine the operating status of the tested equipment based on the comprehensive anomaly index.

[0098] Specifically, the comprehensive diagnostic module receives the individual anomaly probabilities of each parameter from the probabilistic characterization module and the coupling anomaly index from the coupling analysis module. The comprehensive diagnostic module takes the maximum value from the individual anomaly probabilities of each parameter; this maximum value represents the degree of anomaly corresponding to the parameter that deviates most significantly from the normal state. The comprehensive diagnostic module then performs nonlinear aggregation of the maximum individual anomaly probability and the coupling anomaly index to generate a comprehensive anomaly index. Nonlinear aggregation differs from linear weighted summation. Linear weighted summation requires manually setting the weight coefficients for each component, which is subjective, and the components are independent and cannot produce a synergistic effect. Nonlinear aggregation, on the other hand, uses a nonlinear combination method to create a synergistic judgment effect between individual anomalies and coupled anomalies. The comprehensive diagnostic module determines the operating status of the tested equipment based on the comprehensive anomaly index. This can be done by comparing the comprehensive anomaly index with preset judgment conditions or by inputting the comprehensive anomaly index into subsequent diagnostic processes for further analysis. The judgment results can include normal and abnormal states, and can also be further subdivided into different anomaly levels.

[0099] The data flow between the three modules is as follows: The output of the probabilistic characterization module is the individual anomaly probability of each parameter, and the output of the coupling analysis module is the coupling anomaly index. Both are input into the comprehensive diagnosis module. The comprehensive diagnosis module performs nonlinear aggregation on the input data to generate a comprehensive anomaly index and outputs the operational status judgment result. The probabilistic characterization module and the coupling analysis module can run independently of each other, performing parallel analysis of the device under test from the dimensions of individual parameter deviation and coupling structure anomaly, respectively, without dependence on each other. This parallel architecture allows the system to use different time windows and update strategies in the probabilistic characterization module and the coupling analysis module to adapt to the characteristics of their respective analysis tasks.

[0100] It should be noted that the module names in this embodiment maintain a general overview. The probabilistic characterization module, coupling analysis module, and comprehensive diagnosis module correspond to the three functional levels of time-varying probability benchmark calculation, directed coupled network analysis, and nonlinear aggregation judgment, respectively. The module names do not limit the specific algorithms and data structures used within the modules. For example, the probabilistic characterization module can use Gaussian distribution, Gaussian mixture distribution, or other probability models; the coupling analysis module can use transfer entropy or other coupling strength quantification methods; and the comprehensive diagnosis module can use exponential functions or other monotonically decreasing functions as penalty terms. The interface methods between modules can also be flexibly configured, such as transferring data through shared memory, message queues, function calls, etc. This embodiment does not limit this.

[0101] Through the coordinated operation of the three core modules mentioned above, the multi-parameter detection equipment operation status detection system provided in this embodiment achieves the same detection effect as the method embodiment. The probabilistic characterization module calculates the probability of individual anomalies based on a time-varying probability benchmark, enabling the system to adaptively track the slow drift of equipment parameters and changes in operating conditions; the coupling analysis module constructs a time-varying coupling network based on directed nonlinear coupling strength and extracts topological features, enabling the system to capture the directionality, strength, and dynamic evolution of coupling relationships between parameters; the comprehensive diagnosis module uses nonlinear aggregation to collaboratively determine individual anomalies and coupling anomalies. When the coupling topology has become abnormal but the deviation of individual parameters is not yet obvious, the comprehensive anomaly index can be significantly increased, thereby reducing the false alarm rate and improving the ability to identify early weak anomalies.

[0102] Example 7: This embodiment, based on embodiment 6, further explains the adaptive update function of the operating status detection system of the multi-parameter detection device.

[0103] In Example 6, the system includes three core modules: a probabilistic characterization module, a coupling analysis module, and a comprehensive diagnosis module. The system provided in this example also includes an adaptive update module, used to update the probability distribution parameters, baseline topological features, and diagnostic model when the comprehensive anomaly index falls below a preset normal threshold over multiple consecutive sampling periods.

[0104] Specifically, the adaptive update module receives the comprehensive anomaly index output by the comprehensive diagnostic module and monitors the continuous performance of the comprehensive anomaly index over time. The preset normal threshold is a pre-defined judgment limit, which is lower than the first preset threshold used to trigger fine classification diagnosis in Example 5. When the comprehensive anomaly index is lower than the preset normal threshold for multiple consecutive sampling periods, the adaptive update module determines that the device under test is in a continuous normal operating state and triggers the update of the probability distribution parameters, baseline topological features, and diagnostic model.

[0105] The trigger condition means that if the comprehensive anomaly index is lower than the normal threshold for multiple consecutive sampling periods, it indicates that the device under test has not shown significant signs of individual parameter deviation or coupling structure abnormalities over a relatively long period, and the device is likely in a healthy operating state. In this case, it is safe to include the operational data collected in the current period into the training dataset, as data under abnormal conditions will not be mistakenly used as normal data for updates, thus avoiding contamination of the probability benchmark and diagnostic model. The number of consecutive sampling periods can be configured according to the operating characteristics and maintenance strategy of the device under test, for example, hundreds or thousands of sampling periods; this embodiment does not limit this.

[0106] The update targets include three levels; first, probability distribution parameters; the adaptive update module incorporates the observation data of each parameter in the current time period into the first sliding time window and recalculates the probability distribution parameters of each parameter. Taking the Gaussian distribution in Example 2 as an example, the updated sample mean and sample variance will reflect the actual operating status of the equipment in the recent period, enabling the time-varying probability benchmark to continuously track the slow drift of equipment parameters and changes in operating conditions.

[0107] Second, the baseline topology features: The adaptive update module incorporates the coupling pattern data of the current time period into the baseline data pool, updating the baseline topology feature vector and the corresponding covariance matrix. The baseline topology feature vector characterizes the coupled network topology of the device under healthy baseline conditions, and the covariance matrix is ​​used to calculate the deviation between the current topology feature vector and the baseline topology feature vector. By incorporating the coupling pattern data within the current normal operating period into the baseline data pool, the baseline topology features can adjust to changes in individual device characteristics, avoiding deviations from the actual healthy state due to normal device aging or long-term changes in operating conditions.

[0108] Third, the diagnostic model; the adaptive update module uses the normal operation data of the current period as new training samples to incrementally learn and fine-tune the diagnostic model. The diagnostic model includes the classification model used for fine-grained classification diagnosis in Example 5 and the time-series prediction model used for trend prediction diagnosis. The meaning of incremental learning fine-tuning is that, while retaining the historical knowledge already learned by the diagnostic model, the model parameters are slightly adjusted using new training samples, so that the model can adapt to changes in the individual characteristics of the equipment without having to retrain from scratch. There are various specific implementation methods for incremental learning, such as incremental learning methods based on regularization constraints, incremental learning methods based on replay mechanisms, etc., which are not limited in this embodiment.

[0109] Through these three levels of updates, the adaptive update module enables the probability distribution parameters, baseline topology features, and diagnostic model to continuously adapt to changes in individual equipment characteristics, maintaining detection accuracy. When equipment experiences slow changes in parameter distribution and coupling patterns due to normal aging, environmental changes, or adjustments in operating conditions, the adaptive update module can promptly incorporate these changes into the normal baseline, avoiding misjudging normal changes as abnormalities and thus reducing the false alarm rate. Simultaneously, the diagnostic model continuously absorbs new normal operating samples through incremental learning, and its ability to judge fault types and evolution trends gradually improves with accumulated operating time.

[0110] It should be noted that the triggering condition for the adaptive update module is not limited to the form that the comprehensive anomaly index is lower than a preset normal threshold for multiple consecutive sampling periods. In other embodiments, other auxiliary conditions can also be combined to determine whether to trigger an update. For example, it can require that the individual anomaly probability of each parameter is lower than the corresponding threshold in the current time period, or that the coupling anomaly index is lower than a preset coupling normal threshold in the current time period, so as to further improve the security of the updated data. The scope of the update object can also be adjusted according to the actual application scenario. For example, only the probability distribution parameters and the baseline topological features are updated without updating the diagnostic model, or an adaptive adjustment of the length of the first sliding time window is added.

[0111] Some of the data in the above formula are calculated by removing dimensions and taking their numerical values. The formula is the closest to the real situation obtained by software simulation of a large amount of collected data. The preset parameters and preset thresholds in the formula are set by those skilled in the art according to the actual situation or obtained through simulation of a large amount of data.

[0112] The above embodiments are only used to illustrate the technical methods of this application and are not intended to limit it. Although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of this application without departing from the spirit and scope of the technical methods of this application.

Claims

1. A method of detecting an operating state of a multi-parameter detection apparatus, characterized by, include: Based on the current observation values ​​of multiple operating parameters of the device under test, the individual anomaly probability of each parameter is calculated using the time-varying probability benchmark corresponding to each parameter. A time-varying coupling network is determined based on the directed nonlinear coupling strength between the multiple operating parameters, the network topology features of the time-varying coupling network are extracted, and a coupling anomaly index is generated based on the network topology features. The maximum individual anomaly probability among the multiple operating parameters is nonlinearly aggregated with the coupled anomaly index to generate a comprehensive anomaly index. The operating status of the tested equipment is determined based on the comprehensive anomaly index.

2. The method of claim 1, wherein The method involves calculating the individual anomaly probability of each parameter based on the current observed values ​​of multiple operating parameters of the device under test, using the time-varying probability benchmark corresponding to each parameter. This includes: Based on the historical normal operation data within the first sliding time window, update the probability distribution parameters of each parameter; Based on the position of the current observation in the corresponding probability distribution, the individual anomaly probability of each parameter is obtained by bilateral calculation.

3. The method of claim 1, wherein A time-varying coupled network is determined based on the directed nonlinear coupling strength among the multiple operating parameters. The network topology features of the time-varying coupled network are extracted, and a coupling anomaly index is generated based on these network topology features, including: The strength of the directed nonlinear coupling between each pair of parameters in the multiple operating parameters is quantified using transfer entropy. Using the multiple operating parameters as nodes and the transmission entropy value as the edge weight, the time-varying coupled network is obtained; Extract the network topology features of the time-varying coupled network; A coupling anomaly index is generated based on the network topology characteristics.

4. The method of claim 1, wherein The highest individual anomaly probability among the multiple operating parameters is nonlinearly aggregated with the coupled anomaly index to generate a comprehensive anomaly index, including: The comprehensive anomaly index is generated by multiplicative nonlinear fusion, using the largest individual anomaly probability among the multiple operating parameters as the base term and the monotonically decreasing function corresponding to the coupled anomaly index as the penalty term. The comprehensive anomaly index increases as the coupling anomaly index increases.

5. The method of claim 2, wherein The probability distribution parameters for updating each parameter include: The probability distribution corresponding to each parameter is set to a Gaussian distribution, and the probability distribution parameters are updated in real time through maximum likelihood estimation. For the i-th parameter, the probability density function is : ; where i is the number of the running parameter, is the observed value of the i-th parameter, is the sample mean of the i-th parameter within the first sliding time window, is the sample variance of the i-th parameter within the first sliding time window; Based on the position of the current observation in the corresponding probability distribution, the individual anomaly probability of each parameter is obtained using a two-sided calculation method, including: the cumulative distribution function value of the current observation :​ ; wherein is an integral variable; calculating an individual outlier probability for the i-th parameter from the cumulative distribution function value : ; wherein, is the individual anomaly probability of the i-th parameter at time t, is the cumulative distribution function value of the current observation of the i-th parameter, is the min operation.

6. The method of claim 3, wherein Extracting the network topology features of the time-varying coupled network includes: Extract at least one topological feature from the time-varying coupled network, including network coupling density, global transmission efficiency, and overall imbalance. The extracted topological features are combined into a topological feature vector; Generate a coupling anomaly index based on the network topology characteristics, including: obtaining a reference topology feature vector; calculating a deviation degree between a current topology feature vector and the reference topology feature vector , and a rate of change of the current topology feature vector over time ; The coupling anomaly index is generated according to the multiplicative nonlinear structure: ; wherein, is the coupling anomaly index at time t, is the deviation degree, is the change rate, is the total length of historical normal operation of the measured equipment, is the change rate adjustment coefficient, is the sensitivity coefficient, is the hyperbolic tangent function.

7. The method of claim 4, wherein The comprehensive anomaly index is generated through multiplicative nonlinear fusion, including: the largest individual abnormal probability in the plurality of operating parameters is denoted as the coupling abnormality index is denoted as ; Generate the comprehensive anomaly index: ; wherein, is the integrated anomaly index for time t, is the largest individual anomaly probability among the plurality of operating parameters, is the coupling anomaly index, is the exponential function with base the natural constant e.

8. The method of claim 4, wherein After generating the comprehensive anomaly index, the method further includes: When the comprehensive anomaly index is lower than the first preset threshold, the operating status of the tested device is determined to be normal. When the comprehensive anomaly index is between the first preset threshold and the second preset threshold, the device under test is subjected to fine classification diagnosis. When the comprehensive anomaly index is greater than or equal to the second preset threshold, trend prediction diagnosis is performed on the tested device. The outputs of different diagnostic layers are fused using Bayesian rules to obtain a comprehensive diagnostic result.

9. A system for detecting the operating status of a multi-parameter detection device, characterized in that, include: The probabilistic characterization module is used to calculate the individual anomaly probability of each parameter based on the current observation values ​​of multiple operating parameters of the device under test and the time-varying probability benchmark corresponding to each parameter. The coupling analysis module is used to determine the time-varying coupling network based on the directed nonlinear coupling strength between the multiple operating parameters, extract the network topology features of the time-varying coupling network, and generate a coupling anomaly index based on the network topology features. The comprehensive diagnostic module is used to perform nonlinear aggregation of the largest individual anomaly probability among the multiple operating parameters and the coupled anomaly index to generate a comprehensive anomaly index, and to determine the operating status of the device under test based on the comprehensive anomaly index.

10. The operating state detection system of a multi-parameter detection apparatus according to claim 9, wherein It also includes an adaptive update module, which updates the probability distribution parameters, baseline topological features, and diagnostic model when the comprehensive anomaly index is lower than a preset normal threshold for multiple consecutive sampling periods.