An electrical equipment remote state detection and fault early warning system based on Internet of Things

Abstract

This invention discloses a remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things (IoT), belonging to the field of electrical equipment monitoring technology. It includes a status acquisition module that collects operational status information from multiple electrical devices in a target electrical system to obtain operational status data. This invention, through the configuration of a status acquisition module, a data processing module, a topology modeling module, a correlation analysis module, and an early warning judgment module, jointly collects and analyzes electrical, thermal, mechanical, and environmental status data. It combines equipment topology relationships, anomaly timing, and propagation correlations to perform root cause identification and alarm reconstruction. This overcomes the limitations of existing systems that rely on single-parameter diagnosis, improves the ability to identify anomalies, perform causal analysis, and locate the source of complex faults, distinguishes between root cause devices and affected devices, reduces fragmented alarms and the burden of manual troubleshooting, avoids misdiagnosis and misrepresentation, and improves the accuracy, systematic nature, and efficiency of remote diagnosis.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment monitoring technology, specifically to an Internet of Things-based remote status detection and fault early warning system for electrical equipment. Background Technology

[0002] Electrical equipment refers to all types of equipment used in the processes of power generation, transmission, transformation, distribution, and consumption to produce, convert, distribute, control, protect, and consume electrical energy. It encompasses a wide range of equipment, including motors, frequency converters, and lighting equipment from the high-voltage side, medium- and low-voltage side, and end-user side. The safe and reliable operation of electrical equipment is the fundamental guarantee for the stable power supply of the entire power system. Any fault or performance degradation of electrical equipment can trigger local or even large-scale power outages, causing production interruptions, economic losses, and even casualties. Therefore, full life-cycle condition monitoring and effective maintenance management of electrical equipment has always been the most critical issue in the field of power operation and maintenance.

[0003] Patent publication number CN119689144A discloses a remote monitoring system and method for high-voltage electrical equipment based on the Internet of Things (IoT). The system involves sensors collecting data, a data acquisition and transmission module acquiring and processing the data, generating an audit certificate, and transmitting it to a data processing module. The data processing module performs auditing operations to avoid misjudgments of data packet loss. Subsequently, the data undergoes denoising and standardization to improve data quality. A data analysis and prediction module constructs a fault prediction model and analyzes the processed data to predict equipment status, calculate the probability of fault occurrence, and generate a health status score. This score is compared with a preset threshold to assess the equipment's health status. Based on the assessment results, a fault warning and alarm module issues warning commands of different levels, notifying the remote monitoring and control module to take necessary measures. This invention achieves the beneficial technical effects of real-time monitoring of equipment status, early detection of potential faults, and ensuring the safe and stable operation of equipment.

[0004] Existing remote diagnostic systems typically have limited monitoring and diagnostic dimensions. Their knowledge bases and diagnostic rules are mostly preset for single electrical parameters such as temperature, current harmonics, and partial discharge, essentially falling under known fault mode matching. However, due to the complex operating conditions of electrical equipment in the field, faults often evolve from the coupling of multiple factors such as heat, mechanics, environment, and load. While relying solely on single-parameter monitoring can trigger anomaly warnings, it is difficult to accurately distinguish the actual physical root cause of the fault, thus resulting in insufficient diagnostic depth.

[0005] Furthermore, there are obvious electrical couplings and upstream-downstream transmission relationships between electrical equipment. An anomaly in an upstream device can trigger simultaneous alarms in multiple downstream devices, creating a chain reaction of "one cause, multiple effects." Existing systems typically treat each monitoring node in isolation, lacking the ability to analyze the interrelationships and anomaly propagation relationships across the entire "supply-transmission-consumption" chain. This makes it difficult to quickly identify the root cause device and affected devices when multiple devices malfunction simultaneously. This not only increases the cost of manual troubleshooting and reduces the efficiency of remote diagnosis but also easily leads to misdiagnosis and misrepair. In severe cases, the failure to address the root cause of the fault in a timely manner can cause the fault scope to expand. Therefore, this invention is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things, the system comprising: The status acquisition module collects the operating status information of multiple electrical devices in the target electrical system to obtain operating status data; The data processing module processes the operating status data using status data preprocessing and feature construction methods to generate status feature data for each electrical device. The topology modeling module establishes a device association topology model between multiple electrical devices by executing the device association topology modeling method based on the state feature data corresponding to each electrical device. The correlation analysis module constructs cross-device anomaly propagation correlation chains based on the device correlation topology model and the anomaly propagation correlation analysis method. The early warning judgment module determines the root cause device and affected devices based on the cross-device anomaly propagation association chain and uses root cause identification and early warning judgment methods to generate fault early warning results. The alarm reconstruction module generates a unified early warning event based on the fault early warning results through the associated alarm merging and reconstruction method; The interactive output module obtains the target address and outputs unified early warning events in real time based on the target address.

[0008] Furthermore, the state data preprocessing and feature construction method includes: time alignment of operating state data of different electrical equipment; outlier processing, missing value compensation, noise filtering and dimension unification of the time-aligned operating state data; feature extraction of the processed operating state data; and formation of state feature data corresponding to each electrical equipment based on the extracted features. The operating state data includes at least two of electrical state data, thermal state data, mechanical state data and environmental state data. The extracted features include at least one of time domain features, frequency domain features, trend features and coupling features.

[0009] Furthermore, the device association topology modeling method includes: obtaining at least one of the power supply connection relationship, upstream and downstream relationship, branch relationship, busbar affiliation relationship and load association relationship between each electrical device; constructing a device association topology graph with each electrical device as a node and the association relationship between devices as an edge; and assigning directional attributes and weights to the edges based on the association relationship to form a device association topology model. The directional attributes are used to characterize at least one of the energy transfer direction, fault disturbance propagation direction and control action direction, and the weights are used to characterize the association strength between device nodes.

[0010] Furthermore, the anomaly propagation correlation analysis method includes: setting a preset time window, identifying anomaly events within the time window, generating anomaly event sequence according to the occurrence time of the anomaly events, and filtering candidate propagation paths based on the device correlation topology model; determining whether there is a propagation correlation between anomaly events based on the time difference between the anomaly events, the correspondence between anomaly types, the consistency of change trends, and the topological reachability; and connecting anomaly events with propagation correlation to form a cross-device anomaly propagation correlation chain.

[0011] Furthermore, the process of determining whether there is a propagation correlation between abnormal events is as follows: when an upstream electrical device exhibits at least one of the following abnormalities: voltage abnormality, current surge, frequency deviation, insulation abnormality, or partial discharge abnormality, and a downstream electrical device exhibits at least one of the following abnormalities within a preset delay time: harmonic increase, temperature increase, vibration enhancement, efficiency decrease, or increased operational fluctuation, it is determined that there is a propagation correlation between the corresponding abnormal events; when multiple electrical devices exhibit the same type of abnormality, but the direction of abnormality change, duration, or abnormality evolution trend are inconsistent, it is determined that there is no common source propagation correlation between the corresponding abnormal events.

[0012] Furthermore, the root cause identification and early warning determination method includes: determining candidate root cause devices based on cross-device abnormal propagation association chains, calculating the root cause score value of each candidate root cause device; determining the target root cause device based on the root cause score value, determining the affected devices based on the propagation range of the target root cause device in the cross-device abnormal propagation association chain, and generating a fault early warning result based on the state feature data corresponding to the target root cause device.

[0013] Furthermore, the root cause score is determined based on at least two of the following: the order of anomaly occurrence, topological location, propagation coverage, degree of coupling of anomaly features, and anomaly persistence stability.

[0014] Furthermore, the associated alarm merging and reconstruction method includes: merging multiple associated alarms caused by the same target root cause device; classifying the associated alarms into primary and secondary alarms based on the propagation sequence, propagation distance, impact degree, and risk level between each associated alarm and the target root cause device; integrating the primary alarm and secondary alarms into a unified early warning event; and outputting a unified early warning event containing the root cause device, propagation path, affected devices, and risk level.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This IoT-based remote status monitoring and fault early warning system for electrical equipment integrates electrical, thermal, mechanical, and environmental status data into a unified monitoring and analysis system through the coordinated configuration of status acquisition, data processing, and correlation analysis modules. It further extracts time-domain, frequency-domain, trend, and coupling characteristics from the raw data, enabling the system to identify equipment malfunctions from multi-dimensional status changes, rather than simply judging whether a single parameter exceeds limits. Compared to systems that rely solely on rule matching based on single electrical parameters such as temperature, current harmonics, or partial discharge, this invention more comprehensively reflects the fault evolution process under the combined influence of multiple physical factors, including thermal stress, mechanical vibration, environmental disturbances, and load fluctuations, improving the completeness of anomaly identification and its physical interpretation capabilities. Therefore, for scenarios where "the same temperature anomaly manifests but with different root causes," this invention does not merely output a general temperature rise alarm but comprehensively analyzes the formation background by combining relevant current, vibration, load, and environmental characteristics, thus overcoming the problems of misjudgment, missed judgment, and root cause confusion caused by single-dimensional monitoring and diagnosis.

[0016] Simultaneously, the topology modeling module acquires the power supply connections, upstream and downstream relationships, branch relationships, busbar affiliations, and load relationships among various electrical devices. Combined with the timing of abnormal events, the correspondence between abnormal types, the consistency of change trends, and topological reachability, a cross-device abnormal propagation chain is constructed, enabling the system to model and analyze the causal relationships between devices across the entire "supply-transmission-consumption" chain. In this way, when multiple devices such as transformers, motors, frequency converters, busbars, or switchgear experience abnormalities simultaneously within a similar timeframe, the system no longer treats them as isolated alarm points. Instead, it can identify which abnormalities belong to different manifestations of the same propagation process, which are primary abnormalities, and which are accompanying abnormalities caused by propagation. Based on this, the present invention effectively solves the shortcomings of existing systems where each monitoring node operates independently and is difficult to handle "one cause, multiple effects" problems. It enables maintenance personnel to quickly grasp the abnormal propagation path and causal logic from a large number of fragmented alarm signals, significantly enhancing the diagnostic depth and system-level cognitive ability under complex operating conditions.

[0017] Simultaneously, through the early warning judgment module and alarm reconstruction module, after forming an anomaly propagation correlation chain, candidate root cause devices are scored to determine the target root cause device and affected devices. Multiple related alarms triggered by the same root cause device are merged and reconstructed, ultimately outputting a unified early warning event containing the root cause device, propagation path, affected devices, and risk level. Thus, the system can not only indicate "where the anomaly is," but also further answer "where the anomaly originates, where it affects, and what should be prioritized," thereby avoiding maintenance personnel being bogged down in extensive manual cross-comparison and repetitive troubleshooting when multiple devices simultaneously trigger alarms. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system structure of the present invention; Figure 2 This is a schematic diagram of the state data preprocessing and feature construction method of the present invention; Figure 3 This is a schematic diagram of the device association topology modeling method of the present invention; Figure 4 This is a schematic diagram of the abnormal propagation correlation analysis method of the present invention; Figure 5 This is a schematic diagram of the root cause identification and early warning determination process of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Remote condition monitoring and fault early warning for electrical equipment refers to a series of technical means that, without direct contact with the equipment itself, utilize multiple types of sensors (such as temperature, current, vibration, and partial discharge sensors) deployed at key nodes. Leveraging IoT communication technology, real-time operational data is collected and transmitted to a cloud or edge computing platform. Combined with data analysis algorithms and fault diagnosis models, this continuously assesses the equipment's health status, identifies anomalies, and predicts trends, ultimately issuing maintenance or intervention prompts before the equipment completely fails. Its core value lies in transforming the traditional "reactive repair" model into a proactive "early warning, condition-based maintenance" model. This helps maintenance personnel grasp the equipment's operational status without needing to be on-site, effectively reducing sudden power outages and economic losses, and improving the power supply reliability and maintenance efficiency of the power system. It is a key technological support for building new power systems and intelligent industrial operation and maintenance systems.

[0021] Example 1:

[0022] like Figures 1-5 As shown, the present invention provides a technical solution: a remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things, the system comprising: The status acquisition module collects the operating status information of multiple electrical devices in the target electrical system to obtain operating status data; It is important to note that the target electrical system is a system consisting of a collection of electrical devices that require remote status monitoring and fault early warning. The status acquisition module is used to collect the operating status data of multiple electrical devices in the target electrical system. In actual use, this module uses monitoring terminals, smart meters, sensors, or edge acquisition devices installed on the device side to periodically or continuously sample information such as voltage, current, power, temperature, vibration, partial discharge, and humidity of the device, and convert it into an uploadable data format. The principle is that equipment abnormalities will eventually be reflected in certain observable parameters. Only by first forming sufficiently comprehensive raw sensing data can subsequent analysis be based on this. The effect is to establish a multi-device, multi-dimensional, and sustainable status monitoring foundation for the system.

[0023] The data processing module processes the operating status data using status data preprocessing and feature construction methods to generate status feature data for each electrical device. It is important to note that the data processing module performs time alignment, outlier handling, missing value compensation, noise filtering, and dimensional standardization on the collected operational status data. It further extracts characteristic parameters reflecting the equipment's operational status, forming status characteristic data for each electrical device. Here, "time alignment" refers to mapping data from different devices and sampling frequencies to a unified time axis; "outlier handling" removes erroneous data that clearly violates physical continuity; "missing value compensation" repairs a small number of missing sampling points; "noise filtering" reduces the impact of random disturbances on judgment; and "dimensional standardization" allows data from different units to be used in the analysis. Further extraction of status characteristics transforms the raw data into a more representative expression of equipment status changes and abnormal trends. The principle behind this is that when raw data is directly used for cross-device comparisons, it often suffers from high noise, inconsistent scales, and asynchronous time sequences. Standardization is necessary beforehand. This module improves data usability and analytical reliability, providing a consistent data foundation for subsequent propagation and correlation analysis.

[0024] The topology modeling module establishes a device association topology model between multiple electrical devices by executing the device association topology modeling method based on the state feature data corresponding to each electrical device. It is important to note that the topology modeling module is used to obtain at least one of the following relationships among electrical devices: power supply connection relationships, upstream and downstream relationships, branch relationships, busbar affiliation relationships, and load association relationships. Using each electrical device as a node and the association relationships between devices as edges, it constructs a device association topology graph and assigns directional attributes and weights to the edges based on the association relationships to form a device association topology model. The specific process is as follows: first, it reads the wiring relationships, device installation locations, or system configuration relationships; then, it abstracts each device as a node in the graph; subsequently, it abstracts the actual association relationships as edges; finally, it uses directional attributes to represent the propagation direction and weights to represent the strength of the association. The principle behind this is that anomaly propagation does not occur randomly but is dependent on the system's structural relationships. If the connections between devices are unknown, it is impossible to determine whether an anomaly can be transmitted from one device to another. This module establishes the structural foundation for anomaly propagation analysis, enabling the system to infer propagation paths from network relationships.

[0025] The correlation analysis module constructs cross-device anomaly propagation correlation chains based on the device correlation topology model and the anomaly propagation correlation analysis method. It's important to note that the correlation analysis module identifies anomalous events within a time window, generates anomaly event sequences based on their occurrence time, and filters candidate propagation paths using the device correlation topology model. Then, based on the time difference between anomalous events, the correspondence between anomaly types, the consistency of change trends, and topological reachability, it determines whether propagation correlations exist between them. Anomalous events with propagation correlations are connected to form cross-device anomaly propagation correlation chains. Here, an "anomaly event" refers to a state change unit of a device whose state characteristic deviates from the normal range within a certain time and meets the requirement of continuity; an "anomaly event sequence" refers to a set of anomalies from multiple devices ordered by time; a "candidate propagation path" refers to a path in the topology graph that may connect consecutive anomaly nodes; and a "propagation correlation chain" refers to anomaly connection results that satisfy structural, temporal, and physical logic requirements. The principle is that true fault propagation requires multiple conditions to be met: structural connectivity, temporal sequence, typological consistency, and trend correspondence. This module organizes previously fragmented multiple anomaly alarms into a propagation chain with causal logic, improving diagnostic depth.

[0026] The early warning judgment module determines the root cause device and affected devices based on the cross-device anomaly propagation association chain and uses root cause identification and early warning judgment methods to generate fault early warning results. It is important to note that the early warning determination module is used to identify candidate root cause devices based on the cross-device anomaly propagation chain, calculate the root cause score for each candidate root cause device, determine the target root cause device based on the root cause score, and identify affected devices based on the propagation range of the target root cause device in the anomaly propagation chain. Finally, it generates a fault early warning result by combining the status characteristic data corresponding to the target root cause device. The specific process is as follows: first, root cause candidates are screened from the front end or key nodes of the propagation chain; then, scores are given based on factors such as occurrence order, topological location, coverage, coupling degree, and continuous stability; the device with the highest score is determined as the target root cause device; subsequently, the set of devices affected by it is identified, and a comprehensive early warning conclusion is generated. The principle is that when multiple devices malfunction simultaneously, the device that first alarms is not necessarily the root cause source; a comprehensive judgment based on the propagation structure and anomaly intensity is also necessary. This module accurately identifies the source of the fault, avoiding mistaking intermediate nodes in the propagation process as the root cause device.

[0027] The alarm reconstruction module generates a unified early warning event based on the fault early warning results through the associated alarm merging and reconstruction method; It's important to note that the alarm reconstruction module is used to merge multiple related alarms triggered by the same root cause device. It prioritizes alarms based on their propagation sequence, distance, impact, and risk level relative to the root cause device, integrating primary and secondary alarms into a unified early warning event. "Merging" means grouping alarms belonging to the same propagation chain into a single event; "priority classification" distinguishes between primary alarms that truly reflect the root cause and accompanying alarms generated by propagation. The principle is that when faced with a large number of simultaneous alarms, on-site maintenance personnel need a core conclusion that clearly explains the root cause, path, and risk, rather than several independent alerts. This module reduces information interference caused by alarm storms and improves the structure and operability of alarm information.

[0028] The interactive output module obtains the target address and outputs unified early warning events in real time based on the target address.

[0029] It is important to note that the interactive output module is used to output the unified early warning event. The unified early warning event includes at least the root cause device, propagation path, affected devices, and risk level. Specifically, the process can involve displaying or pushing the early warning results to the on-duty personnel through the monitoring platform interface, mobile terminal, SMS, email, audible and visual alarms, or the maintenance work order system. The target address is the address where the on-duty personnel receive the message. The principle is that only when the analysis results are clearly expressed and delivered in a timely manner can they truly be transformed into operational decision-making value. The output content includes the root cause device, which helps to quickly locate the source of the problem; it includes the propagation path, which helps to understand the anomaly evolution logic; it includes the affected devices, which helps to determine the scope of investigation; and it includes the risk level, which helps to arrange the priority of handling. This module transforms complex analysis results into easily understood and executable operational information, improving on-site response efficiency.

[0030] like Figure 2 As shown, the state data preprocessing and feature construction method includes: time alignment of the operating state data of different electrical equipment; outlier processing, missing value compensation, noise filtering and dimension unification of the time-aligned operating state data; feature extraction of the processed operating state data; and formation of state feature data corresponding to each electrical equipment based on the extracted features. The operating state data includes at least two of electrical state data, thermal state data, mechanical state data and environmental state data. The extracted features include at least one of time domain features, frequency domain features, trend features and coupling features.

[0031] It is important to note that state data preprocessing and feature construction methods refer to the process of uniformly processing raw operating state data and extracting representative parameters that can be used for state identification under conditions of different electrical equipment acquisition sources, different sampling frequencies, and different data quality. This process typically includes steps such as time alignment, outlier handling, missing value compensation, noise filtering, and dimensional unification. Among these, time alignment refers to mapping data from different devices such as distribution transformers, low-voltage incoming cabinets, busbars, frequency converters, and motors onto the same time axis according to a unified time reference. For example, if a transformer monitoring terminal uploads data once every 5 seconds and a motor vibration sensor uploads data once every 2 seconds, then the two types of data can be resampled and aligned according to a 5-second granularity. The principle behind this is that anomaly propagation analysis depends on the chronological order of anomaly events. Only by ensuring that the data from different devices are comparable in time can it be determined whether an upstream anomaly occurred before a downstream anomaly. Time alignment can eliminate timing errors caused by inconsistent sampling periods between different devices. Outlier handling refers to the identification and correction of outlier data caused by communication jitter, sensor transient interference, or acquisition failures. For example, if the bus voltage is recorded as 0V at a certain moment, but the two preceding and following sampling points are 381V and 379V respectively, this value can be determined as an invalid outlier and corrected using the sliding median or neighborhood interpolation method. The principle is that the actual electrical operating state usually has a certain continuity, and data that deviates significantly from the neighboring distribution often lacks physical rationality. Outlier handling can prevent false alarms triggered by pseudo-anomalies. Missing value compensation refers to restoring data continuity when a small number of sampling points are lost, using linear interpolation, neighboring value filling, or historical compensation based on the same operating condition. For example, if a frequency converter loses 3 sampling points within 15 seconds, and its preceding and following input currents are 152A and 156A respectively, the intermediate values ​​can be filled by interpolation according to the time ratio. The principle is that short-term continuous signals usually change smoothly, and local missing values ​​can be estimated through neighboring trends. Missing value compensation can maintain the integrity of subsequent feature extraction. Noise filtering refers to removing high-frequency random disturbances or non-state fluctuations from the measured signal. For example, the original vibration velocity curve of an electric motor may contain spikes caused by electromagnetic interference. These spikes can be reduced by using moving average filtering, wavelet denoising, or low-pass filtering. The principle is to separate stable operating components from random noise components, thus highlighting the true trend of changes. Noise filtering improves the stability of anomaly identification. Dimensional unification refers to converting data from different units such as voltage, current, temperature, and vibration into a unified and comparable data representation, such as normalized values, standard scores, or deviation rates. The principle is that the absolute value ranges of data with different dimensions vary significantly; direct comparison can mask the changing characteristics of smaller variables. Dimensional unification makes multi-feature fusion analysis more objective. After preprocessing, the feature construction method further extracts time-domain features, frequency-domain features, trend features, and coupling features from the original data.For example, the root mean square value, fluctuation amplitude, and peak factor can be extracted from three-phase current; the 5th and 7th harmonic content can be extracted from the harmonic spectrum; the temperature rise rate and duration can be extracted from the temperature sequence; and the correlation coefficient between current and temperature rise can be extracted. The principle is that different fault modes exhibit stable differences across different characteristic dimensions. By constructing state characteristic data, raw monitoring quantities can be transformed into state expressions more suitable for anomaly identification and propagation analysis, thereby improving the accuracy of early warning. "Electrical state data, thermal state data, mechanical state data, and environmental state data" further define the sources and categories of operating state data. Electrical state data mainly reflects the power quality and load changes of equipment, such as voltage, current, frequency, power factor, and harmonic content; thermal state data mainly reflects the heat generation and dissipation of equipment, such as winding temperature, bus temperature, casing temperature, and bearing temperature; mechanical state data mainly reflects the operational stability of rotating parts or structural components, such as vibration velocity, vibration acceleration, and impact value; environmental state data mainly reflects the external operating environment of the equipment, such as ambient temperature, ambient humidity, and dust concentration. The specific process involves simultaneously incorporating different categories of data into the condition assessment. For example, when upstream harmonics increase, it may first manifest as current waveform distortion, then lead to increased inverter temperature, and subsequently increased motor vibration. Monitoring only electrical data can identify front-end anomalies, but it's difficult to confirm whether the anomaly has had a substantial impact on the load. Combining thermal and mechanical conditions provides a more complete picture of the anomaly propagation process. "Time-domain characteristics, frequency-domain characteristics, trend characteristics, and coupling characteristics" further define the composition of condition characteristics. Time-domain characteristics reflect the direct statistical properties of the signal on the time axis, such as mean, root mean square value, and standard deviation; frequency-domain characteristics reflect the distribution characteristics of the signal in the frequency space, such as specific harmonic amplitudes and band energy proportions; trend characteristics reflect the direction and speed of change of condition quantities over time, such as temperature rise rate and fluctuation slope; coupling characteristics reflect the correlation between different parameters, such as the correlation coefficient between current changes and temperature changes. The principle is that electrical equipment faults are usually not manifested in a single dimension, but are accompanied by changes in multiple characteristics simultaneously. By combining at least two types of state data and at least one feature form, the operating status of equipment can be more comprehensively characterized, improving the ability to identify early anomalies and complex faults. The holistic approach overcomes the limitations of single monitoring parameters, enabling the system to not only identify electrical disturbances themselves but also observe their subsequent impact on thermal and mechanical conditions, thereby enhancing the physical reliability of propagation correlation judgments.

[0032] like Figure 3As shown, the equipment association topology modeling method includes: obtaining at least one of the power supply connection relationship, upstream and downstream relationship, branch relationship, busbar affiliation relationship and load association relationship between each electrical device; constructing an equipment association topology graph with each electrical device as a node and the association relationship between devices as an edge; assigning directional attributes and weights to the edges based on the association relationship to form an equipment association topology model; the directional attribute is used to characterize at least one of the energy transfer direction, fault disturbance propagation direction and control action direction; and the weight is used to characterize the association strength between equipment nodes.

[0033] It is important to note that the equipment association topology modeling method refers to the process of constructing a graph model reflecting the structural connections and propagation paths between devices based on the actual power supply connections, upstream and downstream relationships, branch relationships, busbar affiliation relationships, and load association relationships between devices in a power distribution system. The specific process includes: first, obtaining basic information such as primary wiring diagrams, equipment installation locations, power supply branches, and control relationships; then, treating each electrical device as a node; next, abstracting the power supply, affiliation, or drive relationships between devices as edges to form an equipment association topology graph; finally, assigning directional attributes and weights to each edge according to the relationship type to obtain an equipment association topology model that can be used for anomaly propagation analysis. For example, in a 10kV / 0.4kV power distribution system, transformers, low-voltage cabinets, buses, frequency converters, and motors can be defined as nodes; if a transformer supplies power to a low-voltage cabinet, a directed edge is established from the transformer to the low-voltage cabinet; if a busbar distributes power to a frequency converter, a directed edge is established from the busbar to the frequency converter. The principle lies in graphically and structurally representing the physical connections, enabling subsequent algorithms to determine whether anomalies have the potential for propagation within the graph structure. The overall effect is to transform the originally scattered device information into a topological relationship model with path meaning, enabling the system to shift from "viewing anomalies from a single device perspective" to "viewing anomaly propagation from a system structure perspective," thus providing a structural foundation for cross-device correlation analysis. "Directional attributes" are used to characterize at least one of the following: energy transfer direction, fault disturbance propagation direction, and control action direction. This means that the correlation edges between different devices are not directionless, but reflect a certain preferred propagation or action direction. For example, the direction from the transformer to the busbar usually represents the power supply direction; the direction from the frequency converter to the motor usually represents the drive direction; some control signal links can also represent the action direction of control commands. The principle is that many anomalies do not propagate with equal probability in any direction, but usually spread downstream along the power supply path or coupling path. For example, an upstream voltage drop is more likely to cause undervoltage or harmonic anomalies in downstream devices, while a slight vibration in a downstream motor generally will not directly cause an upstream transformer voltage anomaly. By setting directional attributes, the accuracy of propagation path screening can be improved. "Weight" is used to characterize the correlation strength between device nodes. Its specific determination can be based on the tightness of electrical connections, power coupling strength, frequency of historical linkage anomalies, or operational dependence. For example, the weight of edges with direct power supply relationships can be set to 0.9 or higher, while the weight of edges with indirect branch relationships can be set to 0.5 to 0.7. The principle is that although different paths all have connectivity, their actual propagation probabilities and influence strengths differ. By introducing weights, the propagation chain that best reflects actual operational patterns can be prioritized among multiple candidate paths. The effect is that the directional attribute solves the problem of "where to propagate," while the weights solve the problem of "how strong the propagation is and which path is more reliable," both of which together improve the accuracy and engineering applicability of anomaly propagation analysis.

[0034] like Figure 4As shown, the anomaly propagation correlation analysis method includes: setting a preset time window, identifying abnormal events within the time window, generating anomaly event sequences according to the occurrence time of the abnormal events, and filtering candidate propagation paths based on the device correlation topology model; determining whether there is a propagation correlation between abnormal events based on the time difference between abnormal events, the correspondence between abnormal types, the consistency of change trends, and the topological reachability; and connecting abnormal events with propagation correlation to form a cross-device anomaly propagation correlation chain.

[0035] It is important to note that the anomaly propagation correlation analysis method refers to the process of identifying abnormal events within a time window and analyzing whether the anomaly propagates along the device association path based on the event sequence and topological relationship. The specific process includes: identifying abnormal events for each device within a preset time window; sorting the abnormal events by occurrence time to form an abnormal event sequence; searching for candidate propagation paths from the node corresponding to the preceding abnormal event to the node corresponding to the following abnormal event based on the device association topology model; and determining whether the events constitute a propagation correlation based on time difference, anomaly type correspondence, consistency of change trends, and topological reachability; if a propagation correlation exists, connecting them to form a cross-device anomaly propagation correlation chain. The principle is that real fault propagation often manifests as "upstream anomaly first, downstream response later, path connectivity, and explainable changes." Therefore, this invention does not employ a simple simultaneous alarm merging method, but rather improves the reliability of propagation judgment through multiple conditional constraints. For example, mere time proximity is insufficient to demonstrate a propagation relationship; there must also be physical causal rationality between the anomaly types. The effect it achieves is that it can identify causally related propagation chains from discrete anomalies across multiple devices and time points, reducing isolated alarms, duplicate alarms, and false merging.

[0036] like Figure 4 As shown, the process for determining whether there is a propagation correlation between abnormal events is as follows: when an upstream electrical device experiences at least one of the following abnormalities: voltage abnormality, current surge, frequency deviation, insulation abnormality, or partial discharge abnormality, and a downstream electrical device experiences at least one of the following abnormalities within a preset delay time: harmonic increase, temperature increase, vibration enhancement, efficiency decrease, or increased operational fluctuation, it is determined that there is a propagation correlation between the corresponding abnormal events; when multiple electrical devices experience the same type of abnormality, but the direction of abnormality change, duration, or abnormality evolution trend are inconsistent, it is determined that there is no common source propagation correlation between the corresponding abnormal events.

[0037] It is important to note that "at least one of the following anomalies occurs in upstream electrical equipment: voltage anomaly, current surge, frequency deviation, insulation abnormality symptoms, or partial discharge anomaly" is an example limiting the type of anomaly at the source end; while "at least one of the following anomalies occurs in downstream electrical equipment within a preset delay time: harmonic increase, temperature rise, vibration enhancement, efficiency decrease, or increased operational fluctuation" is an example limiting the form of anomaly response at the affected end. The specific process involves the system first detecting whether an upstream node has a source-end anomaly, and then searching for a matching response anomaly at the downstream node within the preset delay time. For example, if an upstream transformer experiences partial discharge anomaly and voltage fluctuation, and a downstream frequency converter experiences increased harmonic content and current fluctuation within 40 seconds, and a motor experiences increased temperature rise within 90 seconds, then a propagation correlation can be determined. The principle is that many electrical anomalies physically exhibit typical downstream response patterns, thus a mapping relationship between source-end anomalies and affected-end anomalies can be established through a rule base. The statement "when multiple electrical devices exhibit the same type of anomaly, but the direction of anomaly change, duration, or evolution trend is inconsistent, it is determined that there is no common-source propagation correlation" is used to eliminate false correlations. For example, if two frequency converters both experience an increase in current, but one experiences a sustained increase for 30 seconds while the other only experiences a momentary increase for 5 seconds before returning to normal, and downstream equipment does not show a consistent response, they should not be simply identified as belonging to the same propagation chain. The principle is that homologous propagation typically exhibits a relatively continuous and mutually reinforcing trend, rather than merely belonging to the same anomaly in name. The effect achieved is that by combining forward matching rules and reverse exclusion rules, it can both enhance the ability to identify homologous propagation and reduce the false positive rate.

[0038] like Figure 5 As shown, the root cause identification and early warning determination method includes: determining candidate root cause devices based on cross-device abnormal propagation association chains, calculating the root cause score value of each candidate root cause device; determining the target root cause device based on the root cause score value, determining the affected devices based on the propagation range of the target root cause device in the cross-device abnormal propagation association chain, and generating a fault early warning result based on the status feature data corresponding to the target root cause device.

[0039] It is important to note that root cause identification and early warning determination methods refer to the process of further identifying the originating device that truly triggers the anomaly propagation after a cross-device anomaly propagation chain has been formed, and determining the scope of the anomaly's impact and the early warning result. The specific process includes: first, screening candidate root cause devices based on the propagation chain, typically prioritizing nodes at the front end of the propagation chain, upstream nodes, or nodes with high anomaly intensity; then calculating the root cause score for each candidate device; next, determining the target root cause device based on the score; subsequently, identifying affected devices based on the coverage area of ​​the target root cause device in the propagation chain; and finally, generating a fault early warning result by combining the state characteristic data corresponding to the target root cause device. The principle is that in a propagation chain, the device that first exhibits an anomaly is not necessarily the root cause device; a comprehensive judgment is needed considering topological location, scope of impact, and anomaly stability. For example, a bus node may experience slight fluctuations early on, but if its upstream transformer has a more stable anomaly and a wider impact range, the transformer is more likely to be the true root cause. Through comprehensive scoring, intermediate propagation nodes can be avoided from being misjudged as the source of the root cause. Its effect is to accurately pinpoint the source of the fault from multiple abnormal devices, and to provide clear information on the affected objects and risks, thereby improving the efficiency of fault handling.

[0040] like Figure 5 As shown, the root cause score is determined based on at least two of the following: the order of anomaly occurrence, topological location, propagation coverage, degree of coupling of anomaly features, and anomaly persistence stability.

[0041] It is important to note that "sequence of anomalies, topological location, propagation coverage, coupling degree of anomaly features, and persistence stability" further define the criteria for the root cause score. The sequence of anomalies reflects whether a device is at the forefront of the anomaly chain; generally, the earlier the anomaly occurs, the more likely it is to be the source of propagation. Topological location reflects the device's upstream / downstream position in the system structure; upstream core power supply nodes typically have a higher probability of being the root cause. Propagation coverage reflects whether the device's anomaly triggers a number of downstream device anomalies. The coupling degree of anomaly features reflects the consistency in type and trend between the device's anomaly and subsequent device anomalies. Persistence stability reflects whether the device's anomaly is persistent and not sporadic. The specific process can be achieved by calculating the score using a weighted summation method. For example, if the weight of the sequence of anomalies is 0.30, the weight of the topological location is 0.20, the weight of the propagation coverage is 0.25, the weight of the coupling degree is 0.15, and the weight of persistence stability is 0.10, then the score of a candidate device can be expressed as the sum of the products of each of the above sub-items and their corresponding weights. The principle behind this approach is to replace single-factor judgment with multi-factor fusion, thereby reducing the interference of random factors on the root cause identification results. The effect achieved is to make the root cause identification results more robust and objective, avoid making biased judgments based on only a single feature, and improve the accuracy of judgments in complex fault scenarios.

[0042] like Figure 5 As shown, the associated alarm merging and reconstruction method includes: merging multiple associated alarms caused by the same target root cause device; classifying the associated alarms into primary and secondary alarms based on the propagation sequence, propagation distance, impact degree and risk level between each associated alarm and the target root cause device; integrating the primary alarm and secondary alarms into a unified early warning event; and outputting a unified early warning event containing the root cause device, propagation path, affected devices and risk level.

[0043] It's important to note that the "Associated Alarm Merging and Reconstruction Method" refers to the process of aggregating, classifying, and recombining multiple local alarms caused by the same root cause device to form a unified early warning event. The specific process includes: first, identifying which alarms are caused by the same root cause device; then, classifying the alarms into primary and secondary categories based on propagation sequence, propagation distance, impact degree, and risk level; next, designating alarms with source indication significance as primary alarms and accompanying alarms generated during propagation as secondary alarms; finally, integrating the primary and secondary alarms into a unified early warning event output. For example, transformer voltage anomalies can be considered primary alarms, while low-voltage switchgear harmonic increases, busbar temperature increases, inverter input distortion, and increased motor vibration can be considered secondary alarms, ultimately outputting a unified early warning event: "Transformer power supply anomalies lead to risk of multi-device linkage downstream." The principle is that maintenance personnel are truly concerned with "what is the source of the problem, where is it affecting, and how severe is it," rather than manually piecing together logic from a large number of fragmented alarms. The effects achieved are: significantly reducing duplicate alarms and alarm storms, improving the structure and manageability of early warning information, and enabling on-duty personnel to complete fault location and handling decisions more quickly.

[0044] Example 2:

[0045] An industrial power distribution system includes transformer T1, low-voltage busbar B1, incoming line cabinet C1, frequency converters VFD1 and VFD2, motors M1 and M2, and switchgear SW1. The power supply relationships are: T1→B1→C1→VFD1→M1, T1→B1→C1→VFD2→M2, and T1→B1→C1→SW1. Status acquisition units are installed on each device. T1 acquires three-phase output voltage, current, total harmonic distortion (THDu) of the voltage, and winding temperature; VFD1 and VFD2 acquire input voltage, input current, and THDu of the input current; M1 and M2 acquire three-phase current, bearing temperature, and the root mean square value of vibration velocity; and SW1 acquires contact temperature. The sampling period is 1 second, and the data is uploaded to remote diagnostics via an edge gateway.

[0046] During the monitoring period from 14:05:00 to 14:09:00, the following data was received: THDu of T1 was 4.8% at 14:05:12, 5.6% at 14:05:24, 6.2% at 14:05:36, and 6.1% at 14:05:48; the output line voltage dropped from 401 V to 392 V during the same period. The total harmonic distortion (THD) of the input current of VFD1 was 11.7% at 14:06:02, previously stable at 5.4%; VFD2 was 10.9% at 14:06:15, previously stable at 5.1%. At 14:06:48, the three-phase currents of M1 were 186 A, 178 A, and 181 A, respectively, with a root mean square vibration velocity of 3.6 mm / s, compared to a baseline of 2.1 mm / s. At 14:07:05, the three-phase currents of M2 were 171 A, 164 A, and 168 A, respectively, with a bearing temperature of 73.6 ℃, compared to a baseline of 68.2 ℃, and rising to 76.4 ℃ at 14:09:00. The contact temperature of SW1 was 84.3 ℃ at 14:08:10, compared to 71.5 ℃ previously.

[0047] First, data preprocessing is performed. For missing data at the same time point, time alignment is achieved by filling in adjacent time points; data that clearly exceed the limits and only show single-point spikes are discarded; then, data of different dimensions are normalized. Taking THDu of T1 as an example, assuming its normal reference range is 0%~8%, the normalized value at 14:05:36 is 6.2 / 8 = 0.775; taking the bearing temperature of M2 as an example, assuming the reference range is 0~120 ℃, the normalized value at 14:07:05 is 73.6 / 120 = 0.613. After completing the preprocessing, state features are extracted.

[0048] For the harmonic content in the electrical characteristics, the collected values ​​are used directly. Therefore, the characteristic value of the total harmonic distortion (THD) of the voltage at T1 at 14:05:36 is 6.2%, the characteristic value of the THD of the input current at VFD1 at 14:06:02 is 11.7%, and the characteristic value of the THD of the input current at VFD2 at 14:06:15 is 10.9%. For the current imbalance, the maximum value of the relative average deviation of the three-phase current is used for calculation. Taking M1 as an example, its three-phase current average value is Iavg = (186 + 178 + 181) / 3 = 181.67 A; the absolute values ​​of the deviations of each phase are 4.33 A, 3.67 A, and 0.67 A, respectively; then the current imbalance is 4.33 / 181.67 × 100% = 2.38%. Taking M2 as an example, its average current is Iavg = (171 + 164 + 168) / 3 = 167.67 A; the maximum deviation is 3.67 A; therefore, the current imbalance is... 3.67 / 167.67×100%=2.19%. For the temperature rise characteristic, the heating rate is calculated by dividing the temperature change by the time difference. The M2 bearing temperature rises from 68.2 ℃ at 14:05:00 to 73.6 ℃ at 14:07:05, a time difference of 125 s. Therefore, the heating rate is (73.6-68.2) / 125=0.0432 ℃ / s, or approximately 2.59 ℃ / min. The SW1 contact temperature rises from 71.5 ℃ to 84.3 ℃. If the corresponding time difference is 183 s, then the heating rate is (84.3-71.5) / 183=0.0699 ℃ / s, or approximately 4.19 ℃ / min. For vibration characteristics, the root mean square value of vibration velocity was directly used. Therefore, the vibration characteristic value of M1 was 3.6 mm / s, an increase of 1.5 mm / s compared to the baseline of 2.1 mm / s. Based on the above characteristics, abnormal events were identified according to preset anomaly criteria. The identification results are as follows: E1: T1 first showed voltage harmonic anomaly at 14:05:12; E2: VFD1 showed input current harmonic anomaly at 14:06:02; E3: VFD2 showed input current harmonic anomaly at 14:06:15; E4: M1 showed current imbalance and vibration enhancement anomaly at 14:06:48; E5: M2 showed current imbalance and temperature rise anomaly at 14:07:05; E6: SW1 showed contact temperature rise anomaly at 14:08:10. Subsequently, a device association topology model was constructed based on the power supply connection relationship. In this topology, T1 is located upstream, VFD1, VFD2, and SW1 are located in the middle and downstream, and M1 and M2 are located downstream of VFD1 and VFD2, respectively. Within a preset 5-minute time window, propagation correlation analysis is performed on each anomalous event, such as... Figure 4As shown, the judgment criteria include the time difference between abnormal events, the correspondence between abnormal types, the consistency of change trends, and the topological reachability. For E1 and E2, the time difference is 14:06:02-14:05:12=50 s, and T1→VFD1 is topologically reachable. The abnormal types satisfy the correspondence of "upstream voltage harmonic anomaly—downstream input current harmonic anomaly". Further calculation of the consistency of change trends is performed. The THDu sequence of T1 [4.8, 5.6, 6.2, 6.1] is taken, and the input harmonic sequence of VFD1 in the corresponding time period [8.9, 10.2, 11.7, ...] is taken. [11.5] Calculated using the Pearson correlation coefficient: the means of the two groups are as follows: mean x = (4.8 + 5.6 + 6.2 + 6.1) / 4 = 5.675, mean y = (8.9 + 10.2 + 11.7 + 11.5) / 4 = 10.575; the covariance numerator is (4.8 - 5.675)(8.9 - 10.575) + (5.6 - 5.675)(10.2 - 10.575) + (6.2 - 5.675)(11.7 - 10.575) + (6.1 - 5.675)(11.5 - 10.575) = 1.465; the product of the standard deviations is approximately 1.611; therefore, the correlation coefficient r = 1.465 / 1.611 ≈ 0.91.

[0049] Since the time difference is within the time window, the types correspond, the trend correlation coefficient is 0.91, and the topology is reachable, it is determined that there is a propagation correlation between E1 and E2. Similarly, for E1 and E3, the time difference is 14:06:15-14:05:12=63 s. Taking the input harmonic sequence of VFD2 [8.5, 9.6, 10.9, 10.7], the trend correlation coefficient calculated in the same way is approximately 0.88, therefore it is determined that there is a propagation correlation between E1 and E3. For E2 and E4, the time difference is 14:06:48-14:06:02=46 s. VFD1 and M1 have a direct upstream and downstream topological relationship, and the anomaly type corresponds to "inverter input anomaly - motor current / vibration anomaly". The calculated trend correlation coefficient is 0.84, therefore it is determined that there is a propagation correlation between E2 and E4. For E3 and E5, the time difference is 14:07:05-14:06:15=50 s, and the trend correlation coefficient is 0.86. Therefore, it is determined that there is a propagation correlation between E3 and E5. This forms two cross-device abnormal propagation correlation chains: the first correlation chain is T1(E1)→VFD1(E2)→M1(E4); the second correlation chain is T1(E1)→VFD2(E3)→M2(E5).

[0050] For E6, although it occurred after E1 at 14:08:10, with a time difference of 178 seconds, and T1→SW1 is topologically reachable, the anomaly types are "voltage harmonic anomaly" and "contact local temperature rise anomaly," respectively, which do not meet the preset type correspondence. Furthermore, the SW1 contact temperature showed a daily upward trend in the previous week; for example, the recorded temperatures for the same time period in the last three days were 74.2 ℃, 76.8 ℃, and 79.4 ℃, indicating that it is more consistent with the characteristics of an independent thermal fault caused by local contact resistance degradation. Therefore, E6 is not included in the above propagation chain but is retained as an independent anomaly event.

[0051] In the root cause determination stage, root cause scores are calculated for candidate root cause devices in the propagation chain. The scoring is based on the anomaly occurrence timing, topological upstream degree, propagation coverage, anomaly intensity, and sustained stability as described in the claims. For ease of explanation, this embodiment normalizes and weights the above five indicators, with weights set to 0.25, 0.20, 0.20, 0.20, and 0.15, respectively. Among them: T1 is the earliest to occur in both chains, so its timing is 1.00; it is at the upstream end, so its topological upstream degree is 1.00; the propagation covers 4 affected devices, which is normalized to 1.00 under the maximum value in this scenario; its THDu increases from 2.1% to 6.2%, and the relative increase in intensity is (6.2-2.1) / 2.1=1.95, which is normalized to 0.90; the sustained stability is 0.88 because the anomaly lasts for about 3 minutes and the fluctuation is small. The root cause score for T1 is: 0.25×1.00+0.20×1.00+0.20×1.00+0.20×0.90+0.15×0.88=0.25+0.20+0.20+0.18+0.132=0.962. Based on the system scoring calibration, the output in this embodiment is approximately 0.93.

[0052] For VFD1, its timing lead is lower than T1, so we take 0.62; its topology upstream degree is 0.55; its propagation coverage only affects M1, so we take 0.35; its anomaly intensity is 0.78; and its sustained stability is 0.60. Therefore, the overall score is approximately 0.25×0.62+0.20×0.55+0.20×0.35+0.20×0.78+0.15×0.60=0.581, and the calibrated output is approximately 0.54. Similarly, we can obtain a score of approximately 0.51 for VFD2, approximately 0.22 for M1, and approximately 0.25 for M2. Therefore, we determine T1 as the target root cause device, and VFD1, VFD2, M1, and M2 as affected devices, with VFD1 and VFD2 being first-level affected devices and M1 and M2 being second-level affected devices.

[0053] During the alarm reconstruction phase, multiple local alarms triggered by the same root cause device T1 are merged into a unified early warning event P1. P1 includes at least the following: root cause device is T1, propagation paths are T1→VFD1→M1 and T1→VFD2→M2, affected devices are VFD1, VFD2, M1, and M2, and the risk level is high. Simultaneously, the abnormal event E6 of SW1 is output separately as early warning event P2, which includes the fault location as SW1 contact, the abnormality type as localized temperature rise abnormality, and the risk level as medium-high. Ultimately, the original 6 scattered alarms are reconstructed into 2 unified early warning results. On-site verification results show that the root cause of P1 is the abnormal switching of the upstream parallel compensation capacitor branch of T1, causing increased harmonic distortion of the output voltage, which propagates downstream to the inverter and motor; the root cause of P2 is the increased contact resistance due to fatigue of the SW1 contact spring. With the solution in this embodiment, the root cause identification and unified warning are completed approximately 128 seconds after E1 occurs. Compared to the traditional independent alarm method, which requires maintenance personnel to check each of the 6 alarms one by one, this embodiment can directly provide the root cause device, propagation path and affected range, shortening the fault location time.

[0054] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended embodiments and their equivalents.

Claims

1. An Internet of Things-based remote status detection and fault warning system for electrical equipment, characterized in that, The system includes: The status acquisition module collects the operating status information of multiple electrical devices in the target electrical system to obtain operating status data; The data processing module processes the operating status data using status data preprocessing and feature construction methods to generate status feature data for each electrical device. The topology modeling module establishes a device association topology model between multiple electrical devices by executing the device association topology modeling method based on the state feature data corresponding to each electrical device. The correlation analysis module constructs cross-device anomaly propagation correlation chains based on the device correlation topology model and the anomaly propagation correlation analysis method. The early warning judgment module determines the root cause device and affected devices based on the cross-device anomaly propagation association chain and uses root cause identification and early warning judgment methods to generate fault early warning results. The alarm reconstruction module generates a unified early warning event based on the fault early warning results through the associated alarm merging and reconstruction method; The interactive output module obtains the target address and outputs unified early warning events in real time based on the target address. 2.The remote state detection and fault early warning system for electrical equipment based on Internet of Things according to claim 1, characterized in that: The state data preprocessing and feature construction method includes: time alignment of operating state data of different electrical equipment; outlier processing, missing value compensation, noise filtering and dimension unification of the time-aligned operating state data; feature extraction of the processed operating state data; and formation of state feature data corresponding to each electrical equipment based on the extracted features. The operating state data includes at least two of electrical state data, thermal state data, mechanical state data and environmental state data. The extracted features include at least one of time domain features, frequency domain features, trend features and coupling features. 3.The remote state detection and fault early warning system for electrical equipment based on Internet of Things according to claim 1, characterized in that: The device association topology modeling method includes: obtaining at least one of the following relationships among electrical devices: power supply connection relationship, upstream and downstream relationship, branch relationship, busbar affiliation relationship and load association relationship; constructing a device association topology graph with each electrical device as a node and the association relationship between devices as an edge; and assigning directional attributes and weights to the edges based on the association relationship to form a device association topology model. The directional attributes are used to characterize at least one of the following: energy transfer direction, fault disturbance propagation direction and control action direction, and the weights are used to characterize the association strength between device nodes.

4. The remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things as described in claim 1, characterized in that: The anomaly propagation correlation analysis method includes: setting a preset time window, identifying anomaly events within the time window, generating anomaly event sequence according to the occurrence time of the anomaly events, and filtering candidate propagation paths based on the device correlation topology model; determining whether there is a propagation correlation between anomaly events based on the time difference between anomaly events, the correspondence between anomaly types, the consistency of change trends, and the topological reachability; and connecting anomaly events with propagation correlation to form a cross-device anomaly propagation correlation chain.

5. The remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things according to claim 4, characterized in that: The process of determining whether there is a propagation correlation between abnormal events is as follows: when an upstream electrical equipment experiences at least one of the following abnormalities: voltage abnormality, current surge, frequency deviation, insulation abnormality, or partial discharge abnormality, and a downstream electrical equipment experiences at least one of the following abnormalities within a preset delay time: harmonic increase, temperature increase, vibration enhancement, efficiency decrease, or increased operational fluctuation, it is determined that there is a propagation correlation between the corresponding abnormal events. When multiple electrical devices exhibit the same type of anomaly, but the direction of the anomaly change, its duration, or its evolution trend are inconsistent, it is determined that there is no common source propagation correlation between the corresponding anomaly events.

6. The remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things according to claim 1, characterized in that: The root cause identification and early warning determination method includes: determining candidate root cause devices based on cross-device abnormal propagation association chains, calculating the root cause score value of each candidate root cause device; determining the target root cause device based on the root cause score value, determining the affected devices based on the propagation range of the target root cause device in the cross-device abnormal propagation association chain, and generating a fault early warning result based on the state feature data corresponding to the target root cause device.

7. The remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things as described in claim 6, characterized in that: The root cause score is determined based on at least two of the following: the order of anomaly occurrence, topological location, propagation coverage, degree of coupling of anomaly features, and anomaly persistence stability.

8. The remote status monitoring and fault early warning system for electrical equipment based on the Internet of Things according to claim 1, characterized in that: The associated alarm merging and reconstruction method includes: merging multiple associated alarms caused by the same target root cause device; classifying the associated alarms into primary and secondary alarms based on the propagation sequence, propagation distance, impact degree, and risk level between each associated alarm and the target root cause device; integrating the primary alarm and secondary alarms into a unified early warning event; and outputting a unified early warning event that includes the root cause device, propagation path, affected devices, and risk level.

Images