Weak current system fault prediction method based on digital twinning

Abstract

The application discloses a weak current system fault prediction method based on digital twinning, which comprises the following steps: creating a twin definition code based on a preset domain-specific language, automatically generating a digital twinning graph model, realizing automatic and transactional update of the digital twinning graph model when the code is changed through a version control system and a CI / CD process, comparing actual data of a physical device with an expected state defined by the code, quantifying system drift from two dimensions of configuration and state in parallel, using a state space model of an adaptive Kalman filter to predict a future evolution track of a health degree vector composed of two-dimensional drift, and generating an early warning with a time window when the predicted track reaches a fault threshold; the application significantly improves the accuracy, reliability and timeliness of the weak current system in fault prediction through physical constraint embedding, two-dimensional drift quantification and track prediction.

Description

Technical Field

[0001] This invention relates to the field of automation control technology, and in particular to a fault prediction method for low-voltage systems based on digital twins. Background Technology

[0002] With the rapid development of IoT, big data, and AI technologies, the scale and complexity of low-voltage systems in modern buildings, data centers, and industrial parks, such as security monitoring, building automation, and network communication systems, are increasing daily. To improve operational efficiency and system reliability, digital twin technology has been introduced into the full lifecycle management of low-voltage systems. This technology constructs high-fidelity virtual images of physical entities to achieve real-time monitoring, state simulation, and behavior prediction of low-voltage systems. Building upon this, and combining data-driven predictive maintenance methods, machine learning or deep learning models are used to analyze the massive amounts of time-series data generated by the twin, aiming to detect potential faults in advance. This has become an important development direction in the field of intelligent operation and maintenance.

[0003] However, existing technologies still face several technical bottlenecks in the integration of digital twins and fault prediction, significantly limiting the accuracy and reliability of predictions. First, physical low-voltage systems frequently undergo equipment replacement, topology adjustments, and firmware upgrades. Traditional digital twin models are mostly statically constructed, and their updates rely on manual intervention, resulting in delayed responses and a high risk of errors. This leads to a rapid disconnect between the digital model and physical reality, making predictions based on outdated models inherently inaccurate. Second, purely data-driven prediction models are like "black boxes," with opaque decision-making logic that may produce predictions that contradict physical principles and are difficult to gain the trust of maintenance personnel. More importantly, existing methods primarily focus on "state drift" characterized by abnormal sensor data, while generally lacking effective quantification and monitoring methods for "configuration drift" caused by configuration errors, unauthorized changes, or version incompatibility, thus missing numerous early warning signs of faults. Furthermore, existing prediction methods remain limited to anomaly detection based on the current state or simple threshold alarms, failing to accurately depict the deterioration trend of system health and providing insufficient warning time and clear fault evolution paths. Summary of the Invention

[0004] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0005] In view of the aforementioned existing problems, this invention is proposed. Therefore, this invention provides a fault prediction method for low-voltage systems based on digital twins to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a fault prediction method for low-voltage systems based on digital twins, comprising: Based on a pre-defined domain-specific language, create twin definition code that describes the topology and equipment attributes of a low-voltage system, and generate and deploy a digital twin graph model based on this code; When the twin definition code changes, the deployed digital twin graph model is automatically updated according to the change to maintain consistency between the model and the code definition; The configuration drift and state drift of the system are quantified by comparing the actual data collected from the physical device with the expected state represented by the twin definition code or the digital twin graph model. Based on the quantification results of the configuration drift and the state drift, the future evolution trajectory of the system's health state is predicted using a state-space model, and a fault warning is generated when the predicted evolution trajectory reaches a preset fault threshold.

[0007] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the twin definition code further includes physical rule constraints based on first principles of physics. This step further includes: Before generating the digital twin graph model, the twin definition code is statically validated according to the physical rule constraints.

[0008] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the physical rule constraints include: Total power consumption constraints based on the device power budget, or temperature-related constraints based on the device heat dissipation model.

[0009] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, when the twin definition code changes, the deployed digital twin graph model is automatically updated according to the change to maintain consistency between the model and the code definition, including: The twin definition code will be included in the version control system for management. When the code is changed and committed, the compilation and verification process is automatically triggered; After verification, the differences between the digital twin graph models corresponding to the old and new versions of the code are calculated, and the deployed digital twin graph models are updated in a transactional manner based on these differences.

[0010] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the quantization of configuration drift includes: The actual configuration parameters collected from the physical device are compared with the expected configuration attributes defined for the device in the twin definition code, and a weighted sum is performed according to the preset weights of the differences to obtain the configuration drift index.

[0011] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the desired configuration attribute includes compliance requirements for the device firmware version; and the quantification of configuration drift further includes comparing the actual firmware version of the physical device with the compliance requirements to identify unauthorized firmware changes.

[0012] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the quantification of state drift includes: Based on the real-time telemetry data collected during the normal operation of the system, a low-dimensional data manifold representing the health status of the weak current system is constructed. Calculate the distance from the currently acquired real-time telemetry data point to the low-dimensional data manifold, and use this distance as the quantization result of the state drift.

[0013] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the method includes: based on the quantification results of the configuration drift and the state drift, using a state-space model to predict the future evolution trajectory of the system's health state, and generating a fault warning when the predicted evolution trajectory reaches a preset fault threshold, including: The quantization results of the configuration drift and the state drift are combined to form a multidimensional health state vector, and the future evolution trajectory of the health state vector is predicted using a Kalman filter model.

[0014] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the method further includes: The noise covariance parameter in the Kalman filter model is updated and adaptively adjusted in real time using the residual between the actual observed value and the predicted value of the health state vector.

[0015] As a preferred embodiment of the fault prediction method for low-voltage systems based on digital twins described in this invention, the method further includes: After generating a fault warning, the root causes leading to the configuration drift or increased state drift are analyzed, including: When the configuration drift increases, locate the configuration parameters of the physical device that do not match the expected configuration; As state drift increases, identify one or more telemetry data variables that contribute the most to the deviation from the normal state in real-time telemetry data.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By incorporating the code of the twin definition language describing the low-voltage system into the version control system and deeply integrating it with the CI / CD process, the automated and transactional updates of the digital twin graph model are realized. This changes the problems of delayed updates and model distortion caused by the reliance on manual maintenance in traditional digital twin models. When planned changes occur in the physical system (such as equipment additions or removals, configuration adjustments), maintenance personnel only need to modify the code, which serves as the "sole true source," and the twin model can automatically, safely, and reliably complete its evolution. This ensures that all subsequent analyses and predictions are based on a digital foundation that is highly consistent with physical reality, fundamentally improving the accuracy of fault prediction. 2. By directly embedding rule constraints based on the first principles of physics into the code of the twin definition language, it is ensured that the construction and operation of the model always follow physical and engineering logic. This not only avoids unreasonable design in the planning stage of low-voltage systems, but also excludes results that obviously violate physical common sense during prediction, making the behavior of the model more interpretable and greatly enhancing the confidence of operation and maintenance personnel in the prediction results. 3. By comparing the actual configuration of physical devices with the expected configuration defined in the code, the configuration drift index is quantified. At the same time, the degree of deviation of real-time running data from the healthy mode is quantified through manifold learning to obtain state drift. This can capture both static risks caused by illegal changes and configuration errors, as well as dynamic risks caused by performance degradation and abnormal load. This greatly broadens the scope of fault precursor detection and enables earlier and more comprehensive risk identification. 4. Furthermore, this invention abandons the traditional instantaneous alarm mode based on fixed thresholds and instead utilizes the state-space model of an adaptive Kalman filter to model the health state vector composed of configuration drift and state drift. This model not only smooths noise and accurately estimates the current health status of the low-voltage system, but more importantly, it can also predict its future evolution trajectory. By calculating the time it takes for the predicted trajectory to reach a preset fault threshold, this invention can generate early warning information with a clear time window. This quantifiable predictive early warning provides the operation and maintenance team with a valuable decision-making buffer period, enabling operation and maintenance personnel to calmly plan maintenance windows and allocate resources, truly realizing an operation and maintenance mode from passive response to proactive prevention. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein: Figure 1 This is a flowchart illustrating the overall process of a fault prediction method for low-voltage systems based on digital twins, as described in one embodiment of the present invention. Detailed Implementation

[0018] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0020] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0021] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0022] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0023] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0024] Example 1 Reference Figure 1 This is the first embodiment of the present invention, which provides a fault prediction method for low-voltage systems based on digital twins, including: S1. Based on a pre-defined domain-specific language, create twin definition code that describes the topology and equipment attributes of the low-voltage system, and generate and deploy a digital twin graph model based on the code; Furthermore, a domain-specific language is pre-designed and implemented, referred to in this embodiment as the twin definition language. This language adopts a declarative syntax, allowing maintenance personnel or system engineers to accurately describe each component of the low-voltage system and its internal relationships in a manner similar to writing configuration files. The design goal of this twin definition language is to become the "sole true source" of the state of the low-voltage system. It should be noted that the design of the domain-specific language follows these principles: Structured syntax based on YAML or JSON: To facilitate reading, writing, and machine parsing, the language can be extended based on common data serialization formats (such as YAML or JSON), using key-value pairs and hierarchical structures to define entities, attributes, and relationships; Extensible type system: Allows users to define new device types (such as type:"L2_PoE_Switch") and inherit the attributes of the base type to adapt to low-voltage devices from different manufacturers and models; Embedded Expression Language: To support the definition of physical rule constraints, this language integrates a simple expression engine for parsing and calculating logic such as sum(ports.filter(...).map(...)). This expression language supports basic arithmetic operations, logical operations, and aggregate functions for sets (such as sum, avg, count). S101. Define all device entities in the low-voltage system and the connection relationships between all devices. For example, you can define the core switch on a rack in a data center, the access switches connected to it, and the wireless APs or surveillance cameras connected to the access switches. Each entity has a unique identifier (ID). S102. Describe the attributes of each device entity in detail. In this embodiment, the attributes are divided into two categories: one is static configuration attributes (expected state): which are the configurations that the physical device should have and conform to the design specifications, such as device model, hostname, management IP address, firmware version, VLAN division of each port, PoE (Power over Ethernet) power budget, etc., and these configuration attributes are assigned initial values ​​during the initialization of the twin definition language; the other is dynamic telemetry attributes: which are designed to define the data points that need to be collected from the physical device in real time, such as CPU utilization, memory usage, device temperature, port traffic, actual power consumption of PoE ports, etc. In the twin definition language, these dynamic telemetry attributes usually only define their names and data sources, such as an SNMP OID (Object Identifier) ​​or a specific API endpoint. S103. For this twin definition language, it is permissible to directly embed rule constraints based on the first principles of physics. It should be noted that these constraints are not simple thresholds in the traditional sense, but rather describe the inherent physical or engineering laws that must be followed by the low-voltage system. In this embodiment, the rule constraint may be: defining a total power consumption constraint for a PoE switch entity, and explicitly stating that the total real-time power consumption of all PoE-enabled ports should not exceed a certain safe percentage (e.g., 90%) of the switch's rated power budget; the rule constraint may also be: defining a temperature-related constraint, i.e., the internal temperature of the device should not rise abnormally when the CPU load is low. Furthermore, for greater clarity, this embodiment provides a simplified code snippet example of the twin definition language, which describes a PoE switch: #PoE-Switch-01.tdl id:"PoE-Switch-01"; type:"L2_PoE_Switch"; metadata: location:"DataCenter-A,Rack-A,U-24"; attributes: model: "HPE-Aruba-2930F"; firmware_version: "16.10.0016"; # Expected firmware version poe_power_budget:370.0; #Unit: Watts ports: id:"Ge0 / 1"; description:"Connect-to-AP-101"; poe_enabled:true; #...More port definitions constraints: name:"PowerBudgetIntegrity"; description: "Total PoE power draw must not exceed 90% of the budget"; severity: "critical"; #This is the logical expression for the rule, which will be parsed by the compiler. expression:"sum(ports.filter(p->p.poe_enabled).map(p->p.telemetry.poe_power_draw)) <attributes.poe_power_budget*0.9"; S104. Based on the code of the created twin definition language, a digital twin graph model is generated through a specialized twin compiler. The process is as follows: The twin compiler first reads all relevant code snippets or code files composed of code snippets in the twin definition language, analyzes them according to the original syntax of the twin definition language, and performs a static verification process. It is important to note that this process must be performed before the digital twin graph model is generated and does not interact with any physical devices; it purely checks the completeness and compliance of the logic at the code level. This static verification process includes, but is not limited to: Check for network address conflicts, i.e., whether two devices have been assigned the same IP address; Check the validity of the connection, that is, whether an optical port is incorrectly connected to an electrical port; Verify the logical consistency of physical rule constraints. For example, if the total static port power allocation defined in the twin definition language code exceeds the budget, the compiler will report an error before the digital twin graph model is generated. Specifically, the workflow of this twin compiler is as follows: First, the twin definition language code file is taken as input and decomposed into a series of tokens by the lexical analyzer. Then, the parser constructs an abstract syntax tree (AST) based on the predefined language grammar. Secondly, the syntax tree is traversed to perform type checking, reference resolution, and static validation processes, such as checking whether the IP address is within the defined subnet and checking the syntactic and semantic correctness of the physical rule constraint expressions. The compiler converts the validated syntax tree into specific instructions for the target graph database. For example, if the target is the Neo4j graph database, the compiler will generate a series of Cypher query statements. The definition of a device entity will be converted into a CREATE or MERGE node statement, the join relationship will be converted into a CREATE or MERGE relationship statement, and the attributes will be set in the statement through the SET clause. The output of the entire compilation process is one or more executable Cypher script files. S105. After all static checks pass, the twin compiler will instantiate a complete digital twin graph model in memory (or persist it to the graph database). In this model, each device entity is a node, and the attributes of the node are the attributes defined in the twin definition language. The connection between devices is an edge. Furthermore, the inherent physical verification logic for the aforementioned total power consumption constraint can be represented by the following mathematical formula, and in both static verification and subsequent dynamic monitoring, it is necessary to ensure that the following inequality always holds: in, This represents a port on a PoE switch. This refers to the set of all ports on the switch that have PoE enabled; Port The real-time power consumption value, which will be dynamically obtained from the physical device; This refers to the total PoE power budget defined for the switch in the twin definition language code, which is a static configuration value; It is a preset safety factor, ranging from For example, 0.9 means that the maximum allowable budget percentage is 90%; Similarly, the aforementioned temperature-related constraints can be represented by the following simplified heat dissipation model: in, It is a remotely measured value of the internal temperature of the equipment. It is the ambient temperature of the computer room (which can be used as a telemetry value for another node). It is the telemetry value of CPU utilization, u is the heat dissipation coefficient, and C is a constant offset. During the static verification process, it can check whether the settings of k and C are within a reasonable range. During runtime, this constraint is used to capture abnormal temperature rise caused by heat dissipation system (such as fan) failure. S2. When the twin definition code changes, the deployed digital twin graph model is automatically updated according to the changes to maintain consistency between the model and the code definition. It should be noted that this step aims to achieve dynamic and adaptive evolution of the digital twin model, so as to ensure that the digital world and the code of the twin definition language, which is the "sole source of truth," always remain synchronized, thereby solving the pain point that traditional digital twin models quickly become ineffective due to changes in the physical system. Furthermore, in order to achieve accurate tracking of the code of the twin definition language, the first step of the present invention is to store all code files of the twin definition language (such as the aforementioned PoE-Switch-01.tdl) in a distributed version control system (VCS). Preferably, the distributed version control system uses the Git system. As an open-source system, every modification to the static configuration attributes (expected state) of the low-voltage system, whether it is adding a device, changing an IP address, or adjusting a physical rule constraint, must be done by modifying the code file and recording the modification as a new commit in the Git repository. This provides an immutable historical record with timestamps and operator information for each system change, thereby enhancing the traceability and operational security of the low-voltage system. Furthermore, the present invention is deeply integrated with the control system. Specifically, by configuring a webhook or similar automated trigger, when the version control system (such as GitLab or GitHub) detects a new code commit (git push) on the main branch (e.g., the Main branch), it automatically triggers a predefined compilation and verification process. This process is equivalent to a CI / CD pipeline operation, designed to ensure that upcoming changes do not compromise the integrity and correctness of the low-voltage system. This compilation and verification process includes at least the following steps: Syntax check: Run the lexical and syntactic analysis used in step S104 again to ensure that the submitted twin definition language code is formatted correctly; Static validation: The static validation process is executed again. If a single commit attempts to increase the sum of the PoE budgets of all switches in a rack to a level exceeding the load limit of the rack's power supply lines (a higher-level physical constraint), the process will fail at this stage and prevent the changes from taking effect. Furthermore, the CI / CD pipeline enters the deployment phase only after the compilation and verification process has successfully passed: S201. Simultaneously obtain code snippets / files for two versions of the twin definition language, namely, the old version corresponding to the currently deployed model and the newly submitted version. Subsequently, the twin compiler will generate temporary digital twin graph model representations in memory for these two versions of code, denoted as... and Next, a graph difference comparison algorithm is executed, outputting a change set, denoted as . : in, This is represented as a graph difference comparison algorithm. This indicates the older version corresponding to the currently deployed model. This indicates the version of the currently deployed model that has been newly submitted. Specifically, in this embodiment, the graph difference comparison algorithm can be implemented through the following steps: First, iterate through the new version of the digital twin graph model. All nodes in the old version of the digital twin graph model, for each node, are identified by their unique identifier (ID). Search within; If in If not found, the operation of that node is recorded as ADD_NODE and added to the changeset. middle; If in If found, compare all its static configuration attributes one by one. If any attribute value changes, record the node and the changed attribute as UPDATE_NODE_ATTRIBUTE in the changeset. middle; Secondly, the new version of the graph model is traversed using the same logic as the nodes mentioned above. All edges in the old version of the graph model, for each edge, are represented by the IDs of its source and destination nodes. Find the corresponding edge in the middle; If in If not found, record it as ADD_EDGE; If found, compare the edge's attributes (if any) and record UPDATE_EDGE_ATTRIBUTE; Finally, the old version graph model is traversed in reverse. All nodes and edges in, if in If the corresponding entity cannot be found, record it as REMOVE_NODE or REMOVE_EDGE to the changeset, respectively. middle; It should be noted that, through the above traversal and comparison, a change set containing all atomic operations can be generated precisely. This provides deterministic input for subsequent transactional updates; It should be noted that this change set Includes from Evolved to The required series of atomic operations is as follows: ADD_NODE(node_id, attributes): Indicates the addition of a new device node; REMOVE_NODE(node_id): Indicates the deletion of a device node; UPDATE_NODE_ATTRIBUTE(node_id, attr_name, new_value): Represents updating a specific attribute of a device node; ADD_EDGE(source_id, target_id, attributes): Indicates the addition of a new connection edge; REMOVE_EDGE(source_id, target_id): indicates that a connection edge is deleted; S202. In order to ensure that the running digital twin graph model does not enter an intermediate inconsistent state during the update process (affected by newly submitted code) (e.g., a device node is deleted, but the edge connected to it still exists), the update process must be executed as a transactional request. Specifically, the change set Packing all atomic operations into a single transaction means that this series of atomic operations either all executes successfully or all roll back if any step fails, thus allowing the digital twin graph model to move from a valid state (assuming it is...). Directly transition to another valid state. (and will not expose any intermediate, potentially illegal state); Specifically, the transactional request is sent to a graph database (e.g., Neo4j, JanusGraph) that has deployed the digital twin graph model. The graph database, relying on its own characteristics, can guarantee the atomicity, consistency, isolation, and durability of the transactional request, i.e., the ACID properties. It should be noted that through the above processing steps, the present invention realizes the "self-evolution" capability of the digital twin model, so that operation and maintenance personnel only need to focus on maintaining the code of the twin definition language, and can safely and reliably reflect the planned changes in the physical world into the digital twin graph model, thereby providing a continuous, effective and highly reliable model foundation for subsequent accurate drift quantification and fault prediction. S3. By comparing the actual data collected from physical devices with the expected state represented by the twin definition code or digital twin graph model, the configuration drift and state drift of the system are quantified respectively. It should be noted that this step is a bridge connecting the physical world and the digital world. Its core task is to periodically and automatically compare the "actual state" of the low-voltage system with the "expected state" defined by the code of the twin definition language in a two-dimensional and quantitative manner, so as to accurately capture subtle deviations that may indicate future failures. The two dimensions are "configuration drift" and "state drift". Specifically, configuration drift refers to the inconsistency between the actual configuration parameters of a physical device and the configuration attributes defined in the code of the twin definition language. This inconsistency is usually caused by unauthorized manual modification, configuration failure or faulty automated scripts, and is the direct root cause of low-voltage system failures. Furthermore, a data acquisition engine periodically (e.g., once per hour) connects to every physical device in the low-voltage system. For devices such as network switches and routers, the complete running configuration can be obtained by logging in and executing commands (such as show running-config) via protocols such as SSH or NETCONF. For terminal devices such as cameras and sensors, their configuration information is obtained through proprietary APIs or standard protocols such as ONVIF. At the same time, the actual configuration parameters collected are parsed and stored in a structured manner. Furthermore, the actual configuration parameters collected from the physical device are compared item by item with the configuration attributes defined for that device in the twin definition language code. In this case, the twin definition language code is the "sole true source" of configuration compliance. For example, for the aforementioned PoE switch PoE-Switch-01, the comparison items might include: Firmware version: Check if the actual firmware version of the device matches the firmware_version:"16.10.0016" in the twin definition language code; Hostname: Check if the device's hostname is PoE-Switch-01; Port configuration: Check if the description of port Ge0 / 1 is Connect-to-AP-101 and if its PoE status is enabled; Security Configuration: Check if the device's server configuration matches the expected values; Furthermore, to integrate the discrete configuration differences into a unified, quantifiable metric, this invention introduces the concept of the Configuration Drift Index (CDI), which is obtained by weighted summation of the differences: in, It is the calculated configuration drift index. This is the total number of configuration items being checked. It is the first Each configuration item has a preset weight, which is predefined by domain experts based on the importance of the configuration item. For example, the weight of firmware version differences is often much higher than the weight of differences in port descriptions. It is important to emphasize that for critical configurations such as firmware version that involve system security and stability, their weights are... It will be set very high to highlight the severity of unauthorized firmware changes; The data was collected from physical devices. The actual value of each configuration item; The first one is defined in the code of the twin definition language. The expected value of each configuration item; It is a difference function; its value is 1 when the two inputs are not equal and 0 when they are equal. in, It refers to the above. , It refers to the above. ; It should be noted that the calculated configuration drift index is a dimensionless value, which intuitively reflects the severity of the entire low-voltage system's deviation from its "ideal state" at the configuration level. Specifically, state drift refers to the degree to which the real-time operating performance indicators (telemetry data) of a device deviate from its normal and healthy operating mode. Unlike configuration drift, state drift reflects the abnormality in the "behavior" of the low-voltage system, rather than an error in the "definition". Furthermore, it is necessary to collect high-frequency time-series data for a period of time (e.g., one week) covering all devices and all defined telemetry attributes (such as CPU utilization, memory utilization, port traffic, temperature, etc.) after confirming that the low-voltage system is in a normal and healthy operating state. These data constitute a high-dimensional time-series dataset. Furthermore, since there are inherent correlations among these telemetry attributes (e.g., high CPU utilization is often accompanied by high temperature), direct analysis in high-dimensional space is difficult and inefficient. Therefore, this invention employs nonlinear manifold learning techniques, such as autoencoders, to reduce the dimensionality of these high-dimensional health data and construct a low-dimensional data manifold that represents the health status of the system. Specifically, in this embodiment, the training process of the autoencoder includes: First, within the period when the low-voltage system is confirmed to be in a fault-free and stable operation, collect a sufficiently long period of high-frequency telemetry data (e.g., a week or a month). This dataset represents a sample of the "health" status of the low-voltage system. Using the prepared health status dataset as input, an autoencoder model is trained. This model consists of an encoder and a decoder, both of which are multi-layer neural networks. The training objective is to minimize the reconstruction error, which is the difference between the input data and the output data after "encoding-decoding". Through this training, the model is forced to learn a compact, low-dimensional representation of the health data sample. The space in which this representation is located is the low-dimensional data manifold. Once training is complete, the autoencoder model is saved for subsequent reconstruction of real-time telemetry data and calculation of state drift. It should be explained that this data manifold can be understood as a smooth, low-dimensional "surface" or "curve" in a high-dimensional health data space, in which all healthy low-voltage system state data points are closely distributed on or near this manifold; Furthermore, after the digital twin graph model is deployed and operational, real-time telemetry data is continuously collected, and for each collection cycle, a multi-dimensional telemetry data vector is generated. The reconstructed vector is obtained by reconstructing the data using a pre-trained autoencoder. State drift This is then quantified as the reconstruction error between the original data points and their reconstructed points, typically calculated using the square of the Euclidean distance: in, The encoder section representing the self-encoder. Representing the decoder section, when the low-voltage system is in a healthy state, Closely distributed near the manifold, the autoencoder can reconstruct it with very small errors, therefore The value is very low; when the low-voltage system exhibits abnormal behavior, This will deviate from the healthy manifold, making it impossible for the model to reconstruct it well; therefore, the reconstruction error... It will increase significantly; It is important to emphasize that this state drift quantification result provides a continuous and sensitive indicator for measuring the degree to which the real-time operating state of a low-voltage system deviates from its inherent healthy mode. It should be noted that through the above steps, the configuration drift index, which represents the deviation from the static definition, and the state drift quantification value, which represents the abnormal dynamic behavior, can be obtained. By combining these two dimensions, the early state of the fault in the weak current system can be captured. S4. Based on the quantification results of configuration drift and state drift, the future evolution trajectory of the system's health state is predicted using a state space model, and a fault warning is generated when the predicted evolution trajectory reaches a preset fault threshold. It should be noted that the goal of this step is to establish a prediction of the failure of the low-voltage system based on the quantitative results of configuration drift and state drift, so as to realize early warning operation in the early stage of failure. Compared with traditional processing, it is no longer a simple threshold judgment on the current drift value, but a mathematical model that can describe the dynamic evolution process of the health of the low-voltage system is established, thereby realizing the leap from "post-event alarm" to "pre-event prediction" in traditional fault prediction. Furthermore, in step S3 at each time step... The calculated configuration drift index and state drift quantization value together constitute a two-dimensional health state vector. This vector is then used as the observation value of the subsequent state-space model to comprehensively reflect the low-voltage system at time [time value missing]. Overall health status: in, At any moment Calculated configuration drift index; It should be noted that this invention uses a Kalman filter as the core state-space model to model and predict the potential evolution trend of the health of weak current systems. It should be explained that the Kalman filter is a powerful, recursive Bayesian filter, which is very suitable for handling linear dynamic systems with noise. In the scheme of this invention, we assume that the deterioration of the health of weak current systems (i.e., the increase of configuration drift and state drift) can be approximated as a linear dynamic process in the short term. Furthermore, the state-space model is defined by two fundamental equations: the state equation and the observation equation. Specifically, the state equation is expressed as: in, Indicates at time The real but not directly observable health status vector is, in this embodiment, set as a four-dimensional vector, including two drift values ​​themselves and their rate of change (i.e., velocity), in order to capture its dynamic trend: in, and These are the configuration drift index and the rate of change of state drift, respectively. Specifically, in the above state equation, the Represented as a state transition matrix, this matrix defines the state transition of a weak current system from... Time's up Under the assumption of a uniform velocity model, the state transition matrix can be defined as follows: in, It is the time interval between two time steps; Specifically, in the above state equation, the Represented as an observation matrix, this matrix aims to represent the true health state vector. Mapped to an actual observable vector Since the present invention directly observes the drift value itself, the observation matrix is ​​defined as follows: Specifically, in the above state equation, the Represented as process noise, it indicates the uncertainty of the state evolution model itself. In this embodiment, it is assumed to have a mean of 0 and a process noise covariance of . The Gaussian distribution, i.e. ; Specifically, the observation equation is expressed as: in, This is represented as observation noise, which indicates the noise level during observation. The error introduced by time, in this embodiment, is assumed to have a mean of 0, and the observation noise covariance is... The Gaussian distribution, i.e. ; Furthermore, the recursive process of the Kalman filter also includes two steps: prediction and update, which are performed iteratively. Prediction step: Based on the state estimate from the previous time step, predict the state and uncertainties at the current time step; Update step: Utilize the actual observations at the current time. This allows for the correction of prediction results, leading to a more accurate estimate of the current state. It should be noted that standard Kalman filters typically assume that the process noise covariance and observation noise covariance are constant. However, in actual low-voltage systems, the fluctuation of the system state and the measurement error usually change with the change of the operating mode. To solve this problem, the present invention employs an adaptive Kalman filter (AKF). Specifically, after each update step in the above recursive process, the low-voltage system calculates the residual. That is, the actual observed value Compared with the predicted observations obtained based on the predicted state The differences between them, Indicates in Time to The prediction made on the health state vector at any given time; Specifically, the residual value can be expressed as: It is important to note that the covariance of the residuals should theoretically match the parameters of the state-space model. If the statistical characteristics of the actual calculated residual sequence do not match the theoretical values ​​for a long period of time, it indicates that the parameters of the state-space model (especially the process noise covariance and the observation noise covariance) have become mismatched. Furthermore, this invention also employs a residual-based adaptive algorithm to update and adjust the process noise covariance and observation noise covariance in real time. For example, the Sage-Husa algorithm or a similar moving window estimation method can be used, based on the most recent... The statistical properties of the residuals are used to re-estimate the process noise covariance and the observation noise covariance; Specifically, calculate the residual at the current time step. The sample covariance is calculated using the residual sequence within a moving window (e.g., of size L). Where T represents the transpose operation, and m ranges from k-N+1 to k, with the same meaning as t above; then, the observation noise covariance is updated. : in, b is a decay factor, and b is a forgetting factor close to 1 (e.g., 0.95~0.99). This represents the error covariance predicted at time k, which is calculated based on time k-1. Similarly, the state correction can be used to update the process noise covariance; It should be noted that this adaptive adjustment mechanism enables the Kalman filter to automatically adapt to changes in the dynamic characteristics of the weak current system, thereby maintaining high-precision prediction performance under various operating conditions and greatly improving the robustness and reliability of the model. Furthermore, upon obtaining the current moment... Optimal state estimation After that, it is also necessary to predict the future health status of the low-voltage system; Specifically, the state equations of the Kalman filter are used to recursively predict the future. The state at each time step yields the predicted trajectory of the future health state vector: Furthermore, the configuration drift index in the predicted trajectory and state drift These factors combine to form a comprehensive predictive health index. By using a weighted summation method, we obtain: in, and These are normalized weights, and their value range is... ; and These represent the predicted values ​​for configuration drift and state drift, respectively. Furthermore, the low-voltage system will continuously track this predicted health index trajectory. Compare with one or more preset fault thresholds (e.g., warning thresholds and critical thresholds); Specifically, when a low-voltage system predicts a certain point in the future... , When the preset fault threshold is reached or exceeded for the first time, the low-voltage system will immediately generate a fault warning. This warning will not only indicate the severity of the problem, but also clearly inform the maintenance personnel: "It is expected that the fault will be reached within a certain time frame." Afterwards, the health of the low-voltage system will deteriorate to a warning level; It should be noted that by providing a clear early warning time window, valuable decision-making time is provided for maintenance personnel to arrange maintenance and avoid risks, thus achieving true predictive maintenance. Furthermore, to make the fault warning information more operable, the present invention also includes a step of automatically analyzing the root causes of increased drift, as follows: This process is executed when calculating the configuration drift index in step S3. During the weighted summation of the configuration drift index, the low-voltage system records the detailed information of the inconsistency item, including: device ID, configuration item name, expected value (from twin definition code) and actual value (from physical device). When the configuration drift increases and triggers an alert, the low-voltage system directly presents these recorded inconsistencies as the core content of the root cause analysis report to the operation and maintenance personnel. Because locating the root cause of state drift is relatively complex, because of state drift... It is a comprehensive indicator; therefore, this invention adopts an analysis method based on the contribution of reconstruction error, when state drift... When the value increases, the low-voltage system analyzes the components of each dimension of the reconstruction error vector, that is... Each dimension of the vector corresponds to an original telemetry data variable (such as CPU utilization, temperature, etc.). By calculating the absolute or squared value of each dimension component and sorting them, one or more telemetry data variables that contribute the most to the total reconstruction error can be identified. For example, if the dimension components corresponding to "CPU utilization" and "memory utilization" are found to have the largest values ​​in the error vector, the low-voltage system will report that these two telemetry variables are the main driving factors causing the current state drift to increase. This analysis result will be issued together with the fault warning to guide the operation and maintenance personnel to focus their attention on the most relevant performance indicators, thereby greatly shortening the fault troubleshooting time.

[0025] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0026] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0027] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0028] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0029] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0030] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A fault prediction method for low-voltage systems based on digital twins, characterized in that, include: Based on a pre-defined domain-specific language, create twin definition code that describes the topology and equipment attributes of a low-voltage system, and generate and deploy a digital twin graph model based on this code; When the twin definition code changes, the deployed digital twin graph model is automatically updated according to the change to maintain consistency between the model and the code definition; Quantify the configuration drift and state drift of the system separately; The quantization of the configuration drift includes: The actual configuration parameters collected from the physical device are compared with the expected configuration attributes defined for the device in the twin definition code, and the configuration drift index is obtained by weighted summation according to the preset weight of the difference items. The quantization of the state drift includes: Based on the real-time telemetry data collected during the normal operation of the system, a low-dimensional data manifold representing the health status of the weak current system is constructed. Calculate the distance from the currently acquired real-time telemetry data point to the low-dimensional data manifold, and use this distance as the quantification result of the state drift; Based on the quantification results of the configuration drift and the state drift, the future evolution trajectory of the system's health state is predicted using a state-space model, and a fault warning is generated when the predicted evolution trajectory reaches a preset fault threshold.

2. The fault prediction method for low-voltage systems based on digital twins as described in claim 1, characterized in that, The twin definition code also includes physical rule constraints based on the first principles of physics. This step further includes: Before generating the digital twin graph model, the twin definition code is statically validated according to the physical rule constraints.

3. The fault prediction method for low-voltage systems based on digital twins as described in claim 2, characterized in that, The physical rule constraints include: Total power consumption constraints based on the device power budget, or temperature-related constraints based on the device heat dissipation model.

4. The fault prediction method for low-voltage systems based on digital twins as described in claim 1, characterized in that, When the twin definition code changes, the deployed digital twin graph model is automatically updated according to the changes to maintain consistency between the model and the code definition, including: The twin definition code will be included in the version control system for management. When the code is modified and committed, the compilation and verification process is automatically triggered; After verification, the differences between the digital twin graph models corresponding to the old and new versions of the code are calculated, and the deployed digital twin graph models are updated in a transactional manner based on these differences.

5. The fault prediction method for low-voltage systems based on digital twins as described in claim 1, characterized in that, The desired configuration attributes include compliance requirements for the device firmware version; and the quantification of configuration drift also includes comparing the actual firmware version of the physical device with the compliance requirements to identify unauthorized firmware changes.

6. The fault prediction method for low-voltage systems based on digital twins as described in claim 1, characterized in that, Based on the quantification results of the configuration drift and the state drift, a state-space model is used to predict the future evolution trajectory of the system's health state. When the predicted evolution trajectory reaches a preset fault threshold, a fault warning is generated, including: The quantization results of the configuration drift and the state drift are combined to form a multidimensional health state vector, and the future evolution trajectory of the health state vector is predicted using a Kalman filter model.

7. The fault prediction method for low-voltage systems based on digital twins as described in claim 6, characterized in that, The method also includes: The noise covariance parameter in the Kalman filter model is updated and adaptively adjusted in real time using the residual between the actual observed value and the predicted value of the health state vector.

8. The fault prediction method for low-voltage systems based on digital twins as described in claim 1, characterized in that, The method also includes: After generating a fault warning, the root causes leading to the configuration drift or increased state drift are analyzed, including: When the configuration drift increases, locate the configuration parameters of the physical device that do not match the expected configuration; As state drift increases, identify one or more telemetry data variables that contribute the most to the deviation from the normal state in real-time telemetry data.

Images