Multi-modal sensor fused high-low voltage switch cabinet monitoring system

Abstract

The invention relates to the technical field of intelligent electric power equipment state monitoring, in particular to a multi-modal sensor fused high-low voltage switch cabinet monitoring system, which obtains multi-modal operation data of a high-low voltage switch cabinet through an acquisition module, a first feature modeling module extracts cabinet body response features, and a second feature modeling module extracts multi-modal operation data of the high-low voltage switch cabinet through a second feature modeling module; the second feature modeling module establishes an external condition model, and the self-tuning processing module dynamically corrects the signal features based on the structure and environment parameters and outputs a tuned fusion signal set; the state analysis module generates a high-low voltage switch cabinet operation state data set based on the fusion signal set; the parameter evolution management module records the time sequence change of the tuning parameters and maintains the stability of the signal mapping relation; according to the invention, the consistency of cross-cabinet monitoring data and the self-calibration capability of long-term operation can be realized, and the accuracy and reliability of the operation state monitoring of the high-low voltage switch cabinet are greatly improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent power equipment condition monitoring technology, and more specifically to a high and low voltage switchgear monitoring system that integrates multimodal sensors. Background Technology

[0002] High and low voltage switchgear monitoring systems are crucial for ensuring the safe operation of power systems, enabling real-time monitoring, anomaly identification, and fault early warning of switchgear operation. Existing high and low voltage switchgear monitoring systems primarily employ multi-source sensor data acquisition and intelligent analysis technologies. By fusing data on physical quantities such as temperature, current, voltage, partial discharge, and humidity, they comprehensively assess the equipment's operating status. For example, CN119696190A discloses a high and low voltage switchgear monitoring method based on adaptive sampling and data fusion algorithms, achieving early fault probability identification through signal processing and a state prediction model. CN120046088A proposes a state prediction method based on multi-feature data fusion, combining long short-term memory networks with a data fusion model to improve prediction accuracy. CN120377488A discloses a distributed intelligent monitoring system based on the Internet of Things (IoT), utilizing local model training at each edge node and cloud aggregation to achieve collaborative and intelligent monitoring. These solutions collectively improve monitoring accuracy and automation, forming a complete system from data acquisition and model training to state prediction.

[0003] However, in practical engineering applications, the aforementioned technical solutions often struggle to maintain consistent detection results across different cabinets. For example, when the same monitoring system is deployed in two substations, differences in cabinet manufacturers, heat dissipation structures, metal shielding thickness, and internal component layouts lead to inconsistencies between the temperature and electromagnetic signal baseline values ​​collected by the same sensors in different cabinets. When analyzing this data, the algorithm may misinterpret these structural differences as abnormal conditions, resulting in false alarms or missed alarms. In some existing field tests, older switchgear cabinets, due to insufficient local ventilation, consistently exhibited excessively high temperatures, leading to continuous system-level anomaly detection; while newer cabinets, with their stronger signal shielding capabilities, failed to accurately identify actual discharge hazards. This phenomenon causes significant performance differences in the same monitoring algorithm under different environments, often requiring frequent parameter recalibration, increasing maintenance workload, reducing the comparability and reliability of monitoring results, and hindering the widespread application of intelligent monitoring systems in the power industry. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention discloses a high- and low-voltage switchgear monitoring system with multi-modal sensor fusion, aiming to solve the problems mentioned in the background art.

[0005] To achieve the above-mentioned technical effects, the present invention adopts the following technical solution: A high- and low-voltage switchgear monitoring system with multimodal sensor fusion includes: The acquisition module is used to collect temperature, current, voltage, partial discharge, humidity and vibration signal data of high and low voltage switchgear in real time and generate multimodal signal datasets. The first feature modeling module is used to perform feature modeling on the signal response distribution of high and low voltage switchgear under standard operating conditions based on multimodal signal datasets, extract the structural mapping parameters of the cabinet, and generate a structural feature matrix. The second feature modeling module is used to acquire external operating condition data of high and low voltage switchgear, model the variation law of environmental parameters, form an environmental factor matrix, and output environmental compensation parameters. The self-tuning processing module is used to perform dual-channel difference analysis based on the structural feature matrix and the environmental factor matrix, calculate the structural mapping correction coefficient and the environmental offset compensation coefficient, and obtain the tuned fusion signal set. The status analysis module is used to evaluate the status of signal distribution in the feature space based on the fused signal set, extract operating status features, and generate a high and low voltage switchgear operating status dataset. The parameter evolution management module is used to store the structural mapping correction coefficients and environmental offset compensation coefficients in time series, establish a tuning parameter evolution model, and output a reference parameter set for subsequent periodic tuning calculations.

[0006] Furthermore, the working process of the first feature modeling module includes: Time alignment and cross-correlation function analysis are performed on the multimodal signal dataset to determine the response time difference and amplitude correlation between each signal, and to generate a nodal response feature set. Based on the node response feature set, the vibration power spectral density, current fluctuation rate and temperature gradient of each structural node in the steady state stage are calculated as the structural response parameter set. The set of structural response parameters is arranged in the order of node spatial distribution to form a structural feature matrix, and the comprehensive response characteristics of each node under steady-state operation are recorded.

[0007] Furthermore, the working process of the second feature modeling module includes: Collect external meteorological temperature, relative humidity and atmospheric pressure signals of high and low voltage switchgear, and simultaneously collect signals of air inlet and outlet temperature, internal humidity and airflow speed of the cabinet; Calculate the temperature difference, humidity difference, and airflow velocity difference between the outside and inside based on the same time window to obtain temperature gradient parameters, humidity gradient parameters, and airflow coupling parameters. The above parameters are normalized and time-weighted to generate an environmental factor matrix that reflects changes in the external environment and the heat and moisture transfer characteristics of the internal cavity. Temperature difference correction coefficient, humidity correction coefficient, and airflow correction coefficient are extracted from the environmental factor matrix to obtain the environmental compensation parameter set.

[0008] Furthermore, the self-tuning processing module includes a two-factor constraint unit, a multi-modal determination unit, a self-consistent tuning control unit, a two-factor joint solution unit, and a tuning stability constraint unit; The two-factor constraint unit is used to establish the mutual constraint relationship between the structural mapping characteristics and the environmental compensation characteristics within the same tuning period. By limiting the joint variation range of the structural response correction amplitude and the environmental compensation gain, a two-factor feasible tuning domain is formed. The multimodal determination unit is used to dynamically determine the tuning participation level of each signal channel in the current operating stage based on the statistical stability and cross-modal correlation consistency of the multimodal signal before and after tuning, generate a set of tunable channels and a set of frozen channels, and use the determination result as a constraint condition for subsequent parameter solving. The self-consistent tuning control unit is used to adaptively adjust the update step size of the structure mapping correction coefficient and the environmental offset compensation coefficient according to the signal self-consistency change trend, and generate two-factor tuning intensity and direction control quantities. The dual-factor joint solution unit is used to perform joint parameter solution of the structural mapping correction coefficient and the environmental offset compensation coefficient under the common constraints of the dual-factor feasible tuning domain and the tunable channel set, and generate a self-tuning parameter set corresponding to the tuning period. The tuning stability constraint unit is used to calculate the rate of change of the structure mapping correction coefficient and the environmental offset compensation coefficient within a continuous tuning cycle. When the rate of change exceeds a preset threshold, the parameter value of the current tuning cycle is replaced with the corresponding parameter value of the previous tuning cycle.

[0009] Furthermore, the multimodal decision unit employs a joint reliability modeling method based on multi-scale time consistency, cross-modal energy coupling relationship, and tuning sensitivity to perform dynamic reliability assessment of each modal signal channel. The assessment process includes: Within multiple consecutive tuning cycles, for the first A confidence state vector is constructed from each modal signal channel. The state vector includes a short-time stability component, a long-time drift component, a cross-modal coupling component, and a tuning-sensitive component; Based on the above components, a comprehensive reliability function is constructed, and the function expression is: in, Indicates the first The overall reliability of each modal signal channel within the current tuning cycle; The first element in the credibility state vector represents the... One component; These are the weighting coefficients for each component; This is a nonlinear mapping function used to suppress outliers; When the overall credibility When the modal signal channel exhibits a monotonically decreasing trend over multiple consecutive tuning cycles, it is divided into a frozen channel set to reduce its participation weight in the two-factor collaborative solution process or freeze the tuning participation state.

[0010] Furthermore, the process by which the self-consistent tuning control unit generates the two-factor tuning intensity and direction control quantities includes: Within a continuous tuning period, the consistency state quantity is calculated for the multimodal fused signal after being subjected to the current tuning parameters. And a consistency driving quantity is constructed based on the consistency difference between adjacent tuning cycles. The expression is: in, The consistency enhancement weighting coefficient is used to reflect the extent of consistency improvement brought about by the current tuning. This is the uniform curvature weighting coefficient, used to reflect the accelerating or decaying trend of uniformity changes; The tuning period time interval; Based on the consistency driving quantity A two-factor tuning allocation function is constructed to adaptively allocate the tuning contribution ratios of the structure mapping factor and the environmental compensation factor. The function expression is as follows: in, The output of the tuning control is the structural mapping correction coefficient; The output of the tuning control is the environmental offset compensation coefficient; These are the basic tuning gains for structural factors and environmental factors, respectively. The modulation factor is a constraint on the structural and environmental tuning sensitivity based on the current operating state, and satisfies... ; When the consistency driving quantity When the absolute value is lower than the preset minimum drive threshold, it is determined that the current tuning has entered the uniform convergence interval, and the output of the tuning control quantity is stopped; otherwise, the output includes... and The set of tuning control parameters.

[0011] Furthermore, the processing procedure of the two-factor collaborative solution unit includes: Within the current tuning cycle, candidate updated solutions are constructed based on the tuning control parameter set, using the structure mapping correction coefficients and environmental offset compensation coefficients, to generate a candidate multimodal fusion feature set. ; based on Computational structural environment coherence cost function The calculation formula is: Where N is the number of effective modal channels participating in the collaborative solution; The center vector of the candidate fusion feature; These are the candidate structure mapping correction coefficient and the environmental offset compensation coefficient, respectively. These are the weighting coefficients of structural factors and environmental factors in the time evolution constraint, respectively; The tuning period time interval; The When compared with the corresponding generation value of the previous tuning cycle, when When the cost is less than the historical best, the current candidate updated solution is confirmed as a valid collaborative solution, and the structural mapping correction coefficient and environmental offset compensation coefficient are updated. when When the current solution is higher than the upper limit of the historical stable range, it is determined that the current dual-factor collaborative update violates the multimodal consistency or parameter evolution stability, and the current candidate solution is abandoned while maintaining the parameter state of the previous tuning cycle. After the collaborative solution is completed, the confirmed structural mapping correction coefficients and environmental offset compensation coefficients are output as the self-tuning parameter solution set.

[0012] Furthermore, the processing procedure of the state analysis module is as follows: Based on the fused set of tuned signals, a multimodal operation feature space is constructed under a unified feature mapping rule, and the mapping result of each tuned signal in this feature space is represented as a set of feature points. Statistical analysis is performed on the spatial distribution pattern of the feature point set to extract distribution feature quantities that represent the operating status of high and low voltage switchgear. The distribution feature quantities include feature center location, dispersion degree and distribution directionality parameters. Based on the distribution feature quantity, the state offset within the current operating cycle is obtained and compared with the feature distribution benchmark under the standard operating state to evaluate the degree of deviation of the current operating state from the standard state. When the state offset is within a preset stable range, the high and low voltage switchgear is determined to be in normal operation; when the state offset exceeds the corresponding threshold and shows a continuous evolution trend, the high and low voltage switchgear is determined to enter an abnormal operation state or a deteriorated operation state. The state assessment results and the corresponding distribution characteristics are combined to form the high and low voltage switchgear operating state feature set, which is output as the operating state dataset.

[0013] Furthermore, the state offset is calculated from the statistical difference between the multimodal fusion feature distribution under the current operating cycle and the standard operating state feature distribution. The calculation process includes: Based on the set of multimodal fusion feature points in the current operating cycle, calculate its feature center vector and covariance matrix to form a statistical distribution description of the current operating state; Based on the feature center vector and covariance matrix pre-established under standard operating conditions, a benchmark distribution model for standard operating conditions is formed. The overall deviation of the current operating state distribution from the standard operating state distribution is quantitatively calculated using a covariance-weighted distance metric, and the resulting weighted distance value is used as the state deviation. The state deviation is used to represent the degree of deviation of the current operating state of the high and low voltage switchgear from the standard operating state in the feature space.

[0014] Furthermore, the parameter evolution management module manages the variation law of the tuning parameters across operating cycles through a management and constraint mechanism based on the time evolution characteristics of the tuning parameters. The process includes: For the structural mapping correction coefficients and environmental offset compensation coefficients output by the self-tuning processing module in each operating cycle, a parameter time series is constructed in chronological order. The validity of the parameter time series is then screened, and parameter samples corresponding to unsteady operating phases or abnormal operating states are removed to form a valid tuning parameter series. Based on the effective tuning parameter sequence, the variation amplitude analysis and variation rate analysis of the structural mapping correction coefficient and the environmental offset compensation coefficient are performed respectively to identify the evolution trend characteristics of the parameters in the continuous operation cycle and to distinguish between the stationary evolution segment and the abrupt offset segment of the parameters. For the parameters of the identified stationary evolution segment, an evolutionary constraint model reflecting the long-term variation law of the parameters is constructed to limit the reasonable variation range of the tuning parameters in subsequent operating cycles; When entering a new tuning cycle, based on the evolution constraint model, a subset of parameters that meet the current operating conditions is extracted from the historical effective tuning parameter sequence, and interval constraints or weighting are applied to it to generate a reference parameter set for subsequent periodic tuning calculations.

[0015] Based on the above technical solution, the positive and beneficial effects of the present invention are as follows: 1. This invention establishes a two-factor self-tuning mechanism using a structural feature matrix and an environmental factor matrix, enabling the system to automatically correct signal response differences under varying cabinet structures, ventilation conditions, and installation locations. Compared to existing monitoring systems that use fixed normalized parameters, this solution maintains consistent signal feature distribution in operating scenarios with significant cabinet differences, thereby ensuring uniformity of fault identification thresholds across different devices and solving the engineering challenge of incomparable cross-cabinet data in traditional systems.

[0016] 2. This system employs a periodic correction mechanism combining environmental factor modeling and two-factor self-tuning to dynamically compensate for signal baseline drift caused by external disturbances such as temperature and humidity changes and load fluctuations during operation. This mechanism enables the monitoring system to maintain long-term signal response balance without requiring manual recalibration or algorithm retraining over extended operating periods, significantly reducing maintenance frequency and ensuring continuous stability of monitoring accuracy.

[0017] 3. This invention is based on a long-term evolution model constructed by the time-series recording and updating of structural mapping correction coefficients and environmental compensation parameters, forming a dynamic correspondence between historical and real-time data. This enables the system to compare and judge historical structural response patterns when identifying abnormal equipment status, thereby distinguishing between structural drift and real fault signals, and further improving the accuracy of system anomaly detection and the continuity of system response. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of 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 schematic diagram illustrating the framework principle of the present invention; Figure 2 This is a schematic diagram of the self-tuning processing module of the present invention; Figure 3 This is a schematic diagram of the principle of the multimodal determination unit of the present invention; Figure 4 This is a schematic diagram of the self-consistent tuning control unit of the present invention; The diagram is labeled as follows: 100, Acquisition Module; 200, First Feature Modeling Module; 300, Second Feature Modeling Module; 400, Self-Tuning Processing Module; 500, State Analysis Module; 600, Parameter Evolution Management Module; 401, Two-Factor Constraint Unit; 402, Multimodal Decision Unit; 403, Self-Consistent Tuning Control Unit; 404, Two-Factor Joint Solution Unit; 405, Tuning Stability Constraint Unit. Detailed Implementation

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

[0020] Unless otherwise defined, all techniques and scientific methods used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The descriptions herein are for the purpose of illustrating particular embodiments only and are not intended to limit the invention. The terms "and / or" as used herein include any and all combinations of one or more of the associated listed items.

[0021] In one possible implementation, this monitoring system is deployed using a single switchgear as the basic monitoring unit. Each switchgear is equipped with a temperature sensor, a current sensor, a voltage sensor, a partial discharge sensor, a humidity sensor, and a vibration sensor. These sensors can be installed in locations such as the busbar compartment, circuit breaker compartment, cable compartment, or cabinet frame, depending on the cabinet structure. It should be noted that the "acquisition module 100" mentioned in this application is not limited to an independent hardware module. It can be composed of several distributed sensor nodes, signal conditioning circuits, and data acquisition interfaces, or it can directly reuse some of the original measuring devices in the switchgear. There is no limitation in this regard.

[0022] Various sensors are connected via wired or wireless means to an edge computing unit located on top of the switch cabinet or in an adjacent control cabinet. This edge computing unit can employ an industrial-grade embedded processor in its hardware structure, including at least a data acquisition interface, a storage unit, and a processing unit for running tuning and analysis algorithms. After the system is powered on, the edge computing unit first enters the initialization phase, performing self-tests and time synchronization on each sensor channel and establishing a unified sampling time base. In this embodiment, the sampling period of each sensor can be set to a fixed period within the range of 1 to 10 seconds, which can be determined according to the specific needs of the site.

[0023] During the initial operation phase of the system, when the switchgear is in a known standard operating state (e.g., the initial commissioning phase or a stable operating phase confirmed by manual verification of no abnormalities), the acquisition module 100 continuously acquires multimodal signals and forms a multimodal signal dataset. The first feature modeling module 200 is triggered to run during this phase. It performs time alignment and correlation analysis on the multimodal signal dataset, extracts structural response parameters such as vibration power spectrum, current fluctuation amplitude, and temperature gradient of each acquisition node under steady-state operating conditions, and constructs a structural feature matrix according to the spatial position of the sensors in the cabinet.

[0024] Meanwhile, the second feature modeling module 300 continuously collects external environmental parameters of the cabinet, as well as environmental data from the cabinet's air inlets and outlets and internal cavities, and constructs an environmental factor matrix within the same time window as the first feature modeling module 200. This structural feature matrix describes the mapping relationship between the cabinet's inherent structural conditions and the signal response distribution. It is not a traditional geometric structural model, but rather an engineering mapping description based on signal statistical characteristics. The environmental factor matrix reflects the coupling relationship between external climatic conditions and the cabinet's internal heat transfer, moisture transfer, and ventilation characteristics, and is used for environmental compensation calculations during subsequent tuning.

[0025] After the system enters the normal operation phase, the self-tuning processing module 400 is periodically triggered to run according to a preset tuning cycle. This tuning cycle can be triggered by a timer in engineering implementation, for example, every 5 or 10 minutes, or when the state analysis module 500 detects a significant change in signal distribution; there is no limitation on this. Within each tuning cycle, the self-tuning processing module 400 performs dual-channel difference analysis on the acquired multimodal signals based on the current structural feature matrix and environmental factor matrix, calculates the structural mapping correction coefficient and environmental offset compensation coefficient, and applies these coefficients to the original multimodal signals to generate a tuned fused signal set. In the actual operation scenario of high and low voltage switchgear, the statistical consistency between multimodal signals is itself the result of the combined effect of structural state and environmental conditions; optimizing any single factor alone may disrupt the inherent coordination relationship between multimodal signals. This invention, through the dual-factor constraint unit 401, establishes for the first time a joint feasible tuning domain between the structural mapping correction coefficient and the environmental offset compensation coefficient within the same tuning cycle, so that the updating of the two types of factors is no longer a simple superposition or parallel correction, but a collaborative adjustment process subject to common constraints and mutual checks and balances. The resulting technical effect is that, even with both structural and environmental models already in place, this system can effectively avoid the problem of multimodal signal consistency degradation caused by excessive correction of single factors, thereby improving the overall reliability of the fusion results in a statistical sense.

[0026] Furthermore, this invention introduces "multimodal consistency stability" into the tuning decision layer through the multimodal determination unit 402, rather than merely using it as an evaluation indicator for monitoring results. Based on this, the system can dynamically determine which modal signals in the current operating phase have actual contribution value to structure-environment joint tuning, and accordingly limit or freeze the influence of unstable modes on parameter updates. This significantly reduces the risk of abnormal modal signals interfering with the overall parameter evolution direction. In addition, the self-consistent tuning control unit 403 introduces the multimodal consistency evolution trend as the driving basis for tuning intensity and direction, so that structure mapping correction and environmental offset compensation no longer rely on fixed update step size or a single error index, but adaptively adjust according to the consistency improvement or degradation trend of the multimodal fusion results in the time dimension.

[0027] Then, the multimodal consistency constraint and parameter time continuity constraint are simultaneously incorporated into the joint solution framework through the two-factor joint solution unit 404, so that the updates of structural and environmental parameters not only meet the consistency requirements in the current cycle, but also maintain smoothness and predictability in the cross-cycle evolution process.

[0028] It should be noted that in this embodiment, both the structure mapping correction coefficients and the environmental offset compensation coefficients are stored as parameter vectors in the parameter area of ​​the edge computing unit, with the dimension consistent with the number of signal channels participating in the tuning. After each tuning cycle, the updated parameters are written to the storage unit and used as the initial reference for the tuning calculation of the next cycle.

[0029] The tuned fused signal set is then input to the state analysis module 500. Under a unified feature mapping rule, the state analysis module 500 maps the multimodal fused signal to the same feature space and extracts distribution features such as the center position, dispersion, and directionality of the feature distribution through statistical analysis methods. The state analysis module 500 compares the feature distribution under the current operating cycle with a pre-established standard operating state distribution, calculates the state offset, and determines the current operating state of the switchgear accordingly.

[0030] During long-term operation, the parameter evolution management module 600 performs time-series storage and analysis of the structural mapping correction coefficients and environmental offset compensation coefficients output in each tuning cycle. This module identifies the evolution trend of parameters during stable operation and constructs a parameter evolution constraint model to limit the variation of tuning parameters in subsequent cycles. Through this method, even when the system is deployed among switchgear from different manufacturers and with different structural forms, the stability of the tuning process can be maintained without frequent manual calibration.

[0031] Compared with the "uniform threshold + fixed model" monitoring method commonly used in existing projects, this embodiment does not simply regard the structural differences between different cabinets as the source of anomalies. Instead, it explicitly incorporates the above differences into the tuning and analysis process through structural mapping modeling and environmental compensation modeling, thereby avoiding false alarms or missed alarms caused by differences in cabinets or changes in environmental conditions.

[0032] In actual operation testing, three high- and low-voltage switchgear units (three of each) from two different manufacturers, A and B, were selected in the 110 kV substation. Type A switchgear uses a single-sided ventilation structure with a metal shield thickness of 2.5 mm; Type B switchgear uses a double ventilation channel structure with a shield thickness of 3.8 mm, and the busbar layout differs from Type A by about 15%.

[0033] The experiment was divided into a control group (using a traditional multimodal monitoring system without structural modeling and environmental compensation functions) and an experimental group (using the system of this invention, including structural feature modeling, environmental factor modeling, and a two-factor self-tuning module).

[0034] Environmental conditions were controlled as follows: room temperature fluctuating between 20 and 40 °C; relative humidity between 30 and 85%; operating current between 400 and 800 A; and partial discharge intensity varying between 5 and 25 pC. Each experiment lasted for 144 hours, with a sampling frequency of 5 kHz. The measurement parameters are shown in the table below: Serial Number Test parameters definition unit Measurement methods P1 Temperature signal baseline deviation Average difference in temperature signal between different cabinets under the same load ℃ Comparison of Infrared Thermal Imaging and Temperature Sensors P2 Zero drift of current signal Signal mean shift in open circuit state mA Comparison of sampled averages P3 Poor sensitivity of discharge signal Same power supply signal amplitude difference % Standard discharge power supply simulation P4 Signal false alarm rate The proportion of cases where no abnormalities were found to be abnormal. % System alarm record statistics P5 Correction parameter convergence period The time required for the self-tuning coefficient to reach a stable value s A coefficient change rate of <1% indicates stability. P6 Parameter volatility steady-state stage coefficient variance — Variance calculation P7 System recalibration times Number of times the algorithm is automatically recalibrated within 144 hours Second-rate Control log recording P8 Data consistency coefficient Standard deviation / mean of signals in different cabinets % Multi-cabinet signal statistics P9 Long-term drift rate 24-hour temperature signal baseline change rate % Time series fitting analysis P10 Overall monitoring reliability Percentage of correctly identified anomalies % Anomaly Simulation Detection The experimental data results are shown in the table below: parameter Control group (traditional system) Experimental group (system of this invention) Improvement range P1 Temperature signal baseline deviation 5.8 ℃ 0.9 ℃ ↓ 84.5% P2 current signal zero drift 18 mA 4 mA ↓ 77.8% P3 has poor discharge signal sensitivity 26% 6% ↓ 76.9% P4 Signal False Alarm Rate 14.7% 2.1% ↓ 85.7% P5 Correction parameter convergence period 320 s 68 s ↓ 78.8% P6 Parameter Fluctuation 0.021 0.006 ↓ 71.4% P7 System Recalibration Times 6 times / 144 hours 1 time / 144 hours ↓ 83.3% P8 Data Consistency Coefficient 11.5% 2.3% ↓ 80.0% P9 Long-term drift rate 4.6% 0.7% ↓ 84.8% P10 Monitoring Reliability 85.4% 98.2% ↑ 15.1% Experimental results show that the system of this invention exhibits significantly better signal consistency than traditional solutions under cross-cabinet and cross-environment conditions. When the same system is deployed on high and low voltage cabinets from different manufacturers, the first feature modeling module 200 automatically identifies structural differences and forms different structural feature matrices, while the second feature modeling module 300 forms independent environmental factor matrices based on feedback from environmental sensors. After the two sets of data are input into the self-tuning processing module 400, the system completes coefficient iteration convergence in approximately 10 sampling cycles. This reduces the deviation of the two sets of cabinet signals, which originally had a 6°C temperature baseline difference, to less than 1°C after fusion, and suppresses the difference in discharge detection sensitivity to an acceptable range. Therefore, this system maintains stable and consistent detection capabilities under cross-cabinet and cross-environment application scenarios.

[0035] To facilitate a deeper understanding of the technology in this invention, a detailed description of a high- and low-voltage switchgear monitoring system with multi-modal sensor fusion disclosed in the embodiments of this application is provided below. Please refer to [link to relevant documentation]. Figure 1 The system framework diagram shown indicates that this system includes: The acquisition module 100 is used to acquire temperature, current, voltage, partial discharge, humidity and vibration signal data of high and low voltage switchgear in real time and generate multimodal signal datasets. In practice, temperature sensors are typically installed at locations prone to heat generation, such as busbar connection points, circuit breaker contacts, or cable terminals, to reflect the thermal state of critical components of the cabinet; current and voltage sensors can be installed in the main circuit or measurement circuit to collect operating electrical parameters; partial discharge sensors, which can be ultra-high frequency sensors or transient ground voltage sensors, are installed in insulation gaps or areas of concentrated electric field within the cabinet; humidity sensors are placed in air circulation areas inside the cabinet to reflect changes in internal humidity; and vibration sensors are fixed to cabinet structural components or mounting brackets to collect the mechanical vibration response of the cabinet during operation.

[0036] As one possible implementation, the acquisition module 100 can be configured with multi-level sampling buffers to temporarily store raw signal data obtained at different sampling frequencies, and to organize and encapsulate the data within a preset time window to form a multimodal signal data frame. The multimodal signal dataset is not a simple data splicing, but a set of information including signal type identifiers, sampling timestamps, physical quantity dimensions, and sampling location identifiers, used to clarify the correspondence between different signals in the cabinet structure and operating environment.

[0037] It should be noted that the configuration of the acquisition module 100 is based on meeting the requirements of multimodal signal fusion and subsequent structural feature modeling and self-tuning processing, but is not limited to a fixed sensor combination or installation method. For different types of high and low voltage switchgear, the acquisition conditions can be differentiated into necessary and unnecessary restrictions based on the cabinet structure, operating environment, and on-site implementation conditions.

[0038] Firstly, from the perspective of necessary data acquisition conditions, in order to support the establishment of the structural feature matrix and environmental factor matrix, the acquisition module 100 needs to acquire signal data that reflects the thermal state, electrical operating state, and structural response characteristics of the cabinet. Temperature, current, and voltage signals are the basic acquisition objects. Among them, the temperature signal is used to characterize the heat dissipation and heat distribution characteristics of the cabinet, while the current and voltage signals are used to reflect the operating conditions and load changes.

[0039] Based on this, partial discharge signals, humidity signals, and vibration signals are considered enhanced acquisition targets, used to improve the system's ability to identify insulation degradation, environmental changes, and structural response differences. It should be noted that this invention does not require the simultaneous acquisition of all enhanced signals in all implementation scenarios. For example, in low-voltage switchgear or applications with no significant discharge risk, partial discharge sensors may not be required; in dry environments or cabinets with good sealing structures, humidity signal acquisition can be simplified; in cabinets with high structural rigidity and minimal vibration impact, vibration signals can be used as an optional acquisition item. Such adjustments do not affect the basic technical concept of this invention, but will correspondingly adjust the dimensions of the structural feature matrix and the environmental factor matrix.

[0040] The arrangement of the acquisition module 100 can be adapted to different cabinet structures. For traditional metal enclosed cabinets, the acquisition module 100 is usually centrally installed in the secondary compartment or on the side wall of the cabinet, and connected to each sensor through internal wiring. For modular cabinets, the acquisition module 100 can be divided into multiple sub-acquisition units, which are placed close to different functional compartments to shorten the length of the sensor leads and reduce interference. For compact cabinets with limited space, some acquisition functions can be integrated into the sensor body or a distributed acquisition method can be adopted, and data can be exchanged with the main acquisition module 100 through a communication bus.

[0041] In some implementation scenarios, the switchgear itself already integrates monitoring devices for current, voltage, or temperature. For such cases, the acquisition module 100 of this invention can be used as a data aggregation and unified encapsulation unit, directly acquiring data output from the original monitoring devices in the cabinet via a communication interface, without the need to repeatedly deploy corresponding sensors. As one possible implementation, the acquisition module 100 can interface with the original monitoring system of the cabinet via an industrial communication protocol. The data after interface connection is uniformly converted into the multimodal signal data format defined in this invention before entering subsequent processing, ensuring consistency in time signature, units, and signal type.

[0042] It should be noted that the “acquisition module 100” mentioned in this application is not limited to a single hardware entity. It can be composed of a centralized acquisition device, a distributed acquisition unit, or a combination of the original acquisition device in the cabinet and the data aggregation module. The “acquisition” is not limited to direct physical measurement. Under the premise of not affecting the authenticity of the signal and the consistency of the timing, the equivalent monitoring data obtained through the communication interface is also regarded as the acquisition behavior in the sense of this invention.

[0043] The first feature modeling module 200 is used to perform feature modeling on the signal response distribution of high and low voltage switchgear under standard operating conditions based on a multimodal signal dataset, extracting structural mapping parameters of the cabinet, and generating a structural feature matrix. Specifically, the core function of the first feature modeling module 200 is to quantitatively describe the inherent structural response characteristics of different high and low voltage switchgear under normal operating conditions, and to transform these response characteristics into a structural feature matrix that can be directly used by subsequent algorithms. The output of this module is used to characterize the mapping relationship between the cabinet structure itself and changes in multimodal physical quantities, rather than to evaluate or judge the operating state.

[0044] In specific implementation, the first feature modeling module 200 can be implemented by an industrial computing unit or an embedded processing unit. Its calculation process can be completed locally in the switch cabinet or in the station server or centralized monitoring platform. The specific deployment method does not constitute a limitation on the present invention.

[0045] The first feature modeling module 200 first receives the multimodal signal dataset output by the acquisition module 100. The multimodal signal dataset includes at least two of the following: temperature, current, voltage, and vibration signals. Partial discharge and humidity signals can be selectively introduced according to the specific cabinet configuration. All signals are stored in time-series format and include the corresponding sampling timestamp, sampling channel number, and sensor installation node identifier.

[0046] Before proceeding with feature modeling calculations, this module performs time alignment processing on the multimodal signal dataset. This eliminates differences in sampling frequency, data buffering, and communication latency between different sensors, ensuring comparability of different modal signals from the same structural node under a unified time reference. As one possible implementation, for analog signals with a fixed sampling frequency, a unified sampling sequence can be reconstructed using linear interpolation or spline interpolation; for event-triggered signals, alignment correction can be performed based on hardware trigger timestamps.

[0047] Next, cross-correlation function analysis is performed between different modal signals. The cross-correlation analysis is limited to different signal channels under the same structural node, and is used to extract the response time difference and correlation degree between signals. This analysis can determine whether there is a stable temporal relationship between the changes of various physical quantities under structural excitation conditions, thereby eliminating accidental correlations caused by asynchronous sampling or local interference. Notably, the cross-correlation function analysis is not used to identify anomalous events, but rather to characterize the inherent response consistency of structural nodes under multimodal observations. Its calculation results are used to generate a node response feature set, and do not directly participate in subsequent state determination.

[0048] After obtaining the node response feature set, the analysis interval is further defined as the steady-state operation phase. In this embodiment, the steady-state operation phase is defined as: within a preset time window, the effective rate of change of current and the voltage fluctuation amplitude are both lower than a set threshold, and no significant load switching or operational events occur. This interval can be automatically identified by the system or configured by maintenance personnel according to the on-site operation plan.

[0049] During the steady-state operation phase, this module calculates structural response parameters for each structural node. Specifically, it calculates the power spectral density of the vibration signal to describe the vibration energy distribution characteristics of the structural node under steady-state conditions; it calculates the volatility of the current signal to reflect the sensitivity of the structural response to small load changes; and it calculates the temporal or spatial gradient of the temperature signal to characterize the structural characteristics of heat conduction and heat dissipation paths within the cabinet. All of these parameters are calculated based on statistical results within the steady-state range to avoid interference from transient disturbances.

[0050] It should be noted that the "structural response parameters" mentioned in this application are different from the operating state characteristics. They reflect the response pattern of the structure itself under normal conditions, rather than the characteristic performance under fault or abnormal conditions. Therefore, they have strong time stability and cross-scenario transferability.

[0051] Finally, based on the physical distribution order of the structural nodes within the cabinet, the structural response parameters of each node are arranged to form a structural feature matrix. These structural nodes can correspond to sensor installation locations, functional zones, or predefined structural units. Their arrangement order is determined during system initialization and remains consistent throughout subsequent operation.

[0052] The resulting structural feature matrix serves as the foundational data describing the cabinet's structural mapping relationship. It is stored and output to the self-tuning processing module 400 for subsequent calculation of structural mapping correction coefficients. Since this matrix reflects the inherent structural characteristics of the cabinet, its update frequency can be lower than the real-time acquisition frequency. Typically, remodeling is only required when the cabinet is modified, sensors are rearranged, or there are significant long-term structural evolutions.

[0053] The second feature modeling module 300 is used to acquire external operating condition data of high and low voltage switchgear, model the variation law of environmental parameters, form an environmental factor matrix, and output environmental compensation parameters. In specific implementation, the second feature modeling module 300 receives synchronous data from external environmental sensors and internal cabinet environmental sensors. The external environmental data includes at least the meteorological temperature and relative humidity signals at the location of the cabinet; when conditions permit, atmospheric pressure signals can also be collected as auxiliary parameters. The internal cabinet environmental data includes at least the temperature signals at the air inlet and outlet, the relative humidity signal inside the cabinet, and the airflow velocity signal along the ventilation path. The aforementioned sensors can be independently installed environmental monitoring units or provided by the existing environmental monitoring system of the substation; their data interface format does not constitute a limitation of the present invention.

[0054] During the data acquisition phase, the second feature modeling module 300 timestamps both external and internal environmental signals and processes them in segments according to a preset time window. The length of the time window can be configured according to the rate of environmental change, and is generally longer than the multimodal signal acquisition period, used to reflect the slow impact of environmental changes on the internal conditions of the cabinet.

[0055] Within each time window, this module calculates the corresponding difference and gradient indices based on external and internal environmental parameters. Specifically, by comparing the external meteorological temperature with the temperatures at the cabinet's inlet and outlet, a temperature difference gradient parameter is calculated to describe the intensity and direction of the transmission of external temperature changes to the interior of the cabinet; by comparing the external relative humidity with the humidity inside the cabinet, a humidity gradient parameter is calculated to characterize the degree of influence of environmental humidity changes on the internal humid environment of the cabinet; and by comparing the airflow velocities at the inlet and outlet, an airflow coupling parameter is calculated to reflect the cabinet's ventilation structure's responsiveness to external air exchange conditions.

[0056] It should be noted that the "temperature gradient parameter" and "humidity gradient parameter" mentioned in this application are different from simple temperature difference or humidity difference. They emphasize the transmission relationship between external changes and internal conditions within a certain time window, rather than instantaneous absolute values. Therefore, they can more accurately reflect the trend of environmental changes.

[0057] After obtaining the aforementioned gradient and coupling parameters, the second feature modeling module 300 normalizes them to eliminate differences in the dimensions and numerical ranges of different physical quantities. The normalization process can be set based on historical statistical ranges, rated operating intervals, or empirical thresholds. Simultaneously, to enhance the model's adaptability to long-term environmental changes, this module introduces a time-weighted mechanism into the normalized parameters, making the influence of recent environmental changes on the environmental factor matrix more significant than that of historical data.

[0058] Through the above processing, this module generates an environmental factor matrix, which comprehensively describes the mapping relationship between changes in the external environment and the heat transfer, moisture transfer, and ventilation characteristics of the cabinet's internal cavity. Each element in the environmental factor matrix has a clear physical meaning, and their arrangement order and parameter type are determined during the system initialization phase and remain consistent during operation.

[0059] The module then further extracts the environmental compensation parameter set from the environmental factor matrix. Specifically: it calculates the temperature difference correction coefficient based on the temperature gradient parameter to correct the environmental offset of temperature-related signals; it calculates the humidity correction coefficient based on the humidity gradient parameter to compensate for the influence of humidity on partial discharge and insulation performance-related signals; and it calculates the airflow correction coefficient based on the airflow coupling parameter to reflect the regulatory effect of changes in ventilation conditions on heat and moisture accumulation.

[0060] It should be noted that the "environmental compensation parameters" mentioned in this application are different from environmental state description parameters. They do not directly reflect the environmental conditions themselves, but are used as input parameters for the subsequent self-tuning processing module 400 to perform baseline correction on multimodal signals, in order to eliminate the systematic shift caused by environmental changes.

[0061] The self-tuning processing module 400 is used to perform dual-channel difference analysis based on the structural feature matrix and the environmental factor matrix, calculate the structural mapping correction coefficient and the environmental offset compensation coefficient, and obtain the tuned fusion signal set. The self-tuning processing module 400 can be deployed in the local industrial processing unit of the switchgear, or on a station-side server or cloud platform. Its calculation process uses the tuning cycle as the basic operating unit. The length of the tuning cycle can be configured according to operational requirements, and is generally longer than the sampling period of the multi-mode signal to ensure sufficient statistical stability in parameter calculation.

[0062] Please see Figure 2 As shown, within each tuning cycle, the self-tuning processing module 400 first calls the two-factor constraint unit 401 to establish a joint constraint relationship between the structural mapping correction coefficient and the environmental offset compensation coefficient. To ensure the process has clear engineering feasibility, in this embodiment, the two-factor constraint unit 401 uses the structural feature matrix and the environmental factor matrix as the only input data source. The function of the two-factor constraint unit 401 is not to directly calculate parameters, but to establish a feasible tuning domain for the subsequent solution process. In this embodiment, the structural mapping correction coefficient is used to describe the degree of response offset of the multimodal signal under different cabinet structures, and the environmental offset compensation coefficient is used to describe the magnitude of the impact of external environmental changes on the signal baseline. The dimensions of the above matrices are defined during the system initialization phase, and the matrix structure remains unchanged during subsequent operation, only the numerical content is updated.

[0063] The two-factor constraint unit 401 first calculates the magnitude of change of the structural feature matrix within the current tuning period. Specifically, for the structural response parameters corresponding to each structural node, it calculates the offset of the parameters relative to the historical reference period within the current tuning period, and determines the maximum allowable structural correction ratio for that node based on the historical statistical interval or the reference range provided by the parameter evolution management module 600. This ratio is used to limit the variable range of the structural mapping correction coefficients within the current tuning period, thereby avoiding over-correction of structural parameters due to a single abnormal data event.

[0064] Simultaneously, the two-factor constraint unit 401 performs the same amplitude evaluation processing on each environmental gradient parameter in the environmental factor matrix. By calculating the rate of change of the temperature gradient, humidity gradient, and airflow coupling parameters within the current tuning cycle, and combining this with the ventilation capacity, heat dissipation structure, and moisture-proof rating parameters in the cabinet design, the maximum gain range of the environmental offset compensation coefficient is limited. This limitation ensures that environmental compensation only responds to reasonable environmental changes and does not overcompensate for sudden, non-continuous environmental disturbances.

[0065] Based on the two constraints mentioned above, the two-factor constraint unit 401 further establishes a joint constraint relationship between the structural mapping correction coefficient and the environmental offset compensation coefficient within the same parameter space. Specifically, when the environmental compensation coefficient approaches its allowable gain upper limit, the adjustable amplitude of the structural mapping correction coefficient is correspondingly compressed; conversely, when the structural mapping correction coefficient is close to its maximum correction ratio, the gain change of the environmental compensation coefficient is restricted. In this way, a two-factor feasible tuning domain in which structure and environment mutually restrain each other is formed. It should be noted that this "two-factor feasible tuning domain" introduces the coupling constraints of structural factors and environmental factors, ensuring that the tuning result is always within a parameter space that conforms to the physical characteristics of the cabinet and the logic of environmental response, thus avoiding distortion of the tuning result from an engineering perspective.

[0066] After completing the two-factor constraint, the multimodal decision unit 402 is used to dynamically evaluate the reliability of each modal signal channel in the current operating phase. For details on the working principle, please refer to [link to relevant documentation]. Figure 3 As shown, its participation level in the tuning process is determined accordingly. Its input accepts multimodal signal channel data, and its output is connected to the data input of the two-factor joint solution unit 404. It should be noted that a modal signal channel refers to a similar signal data stream after uniform time alignment, scale normalization, and basic preprocessing. For example, temperature data from multiple temperature sensors can be considered as a single temperature modal signal channel before entering this unit.

[0067] During implementation, the multimodal determination unit 402 does not make judgments based on data from a single tuning cycle, but rather performs a comprehensive evaluation based on the evolutionary characteristics of signal behavior over multiple consecutive tuning cycles. Therefore, the system provides the first... A confidence state vector is constructed from each modal signal channel. This state vector is used to describe the stability and consistency of the signal at different time scales and different physical correlation dimensions. The confidence state vector includes at least short-time stability components, long-time drift components, cross-modal coupling components, and tuning sensitivity components.

[0068] Among them, the short-time stability component This component reflects the instantaneous fluctuation of the i-th mode signal within a single tuning cycle. It can be obtained by calculating the mean square fluctuation amplitude or energy dispersion of the sampled values ​​of the mode signal within the current tuning cycle, with the calculation window consistent with the tuning cycle. The engineering significance of this component lies in distinguishing between "explainable variations caused by structure or environment" and "random noise or occasional interference." When this component exceeds a preset threshold, it indicates that the signal does not possess stable statistical characteristics within the current cycle.

[0069] Long-term drift components This component is used to characterize the baseline variation trend of the i-th modal signal across multiple tuning cycles. It is obtained by calculating the rate of change of the statistical mean or characteristic quantity of the signal over consecutive tuning cycles. It should be noted that drift is not directly equivalent to equipment degradation, but rather refers to the continuous deviation of the signal from the expected response relationship described by the structural feature matrix and the environmental factor matrix.

[0070] Cross-modal coupling components This component is used to reflect the cooperative relationship between the i-th modal signal and the other modal signals. It is obtained by calculating the correlation index between the i-th modal signal and at least one other modal signal within the same tuning period. Its physical meaning lies in determining whether the signal still conforms to the structural or environmental change mechanism jointly reflected by multiple modes. If the change trend of a certain modal signal cannot reasonably correspond to other modes over a long period, this component will decrease significantly.

[0071] Tuning Sensitive Components This component is used to characterize the response of the i-th modal signal to changes in the structure mapping correction coefficient and the environmental offset compensation coefficient. It is obtained by comparing the proportional relationship between the signal's amplitude change before and after tuning and the parameter's amplitude change, and is used to determine whether the signal has a reasonable linear or weakly nonlinear response to changes in the tuning parameters. If a signal exhibits excessive amplification or almost no response during parameter fine-tuning, this component is considered abnormal.

[0072] Based on the above components, the multimodal determination unit 402 constructs the comprehensive reliability function of the i-th modal signal channel. ,in, This indicates the overall reliability evaluation result of the modal signal within the current tuning cycle; These are weighting coefficients used to balance the influence of each confidence component; It is a nonlinear mapping function, which aims to suppress the dominant influence of outliers on the overall credibility and avoid the distortion of judgment results caused by a single outlier.

[0073] During implementation, when the overall credibility is... If a signal exhibits a continuous downward trend over multiple tuning cycles, or falls below a preset minimum confidence threshold, the multimodal determination unit 402 marks the modal signal as a low-confidence channel. For marked low-confidence channels, the system does not stop signal acquisition; instead, it reduces the channel's weight in subsequent two-factor joint solutions, or excludes it from the tuning parameter solution constraints within a preset period. It should be noted that the "frozen channel" described in this application refers to freezing at the parameter solution level, not disabling at the hardware level. Its purpose is to prevent structural differences or abnormal environmental signals from being mistakenly treated as learnable features during self-tuning, thereby compromising the stability of the tuning results.

[0074] Next, the self-consistent tuning control unit 403 adaptively generates tuning intensity and tuning direction control quantities for the structure mapping correction factor and environmental offset compensation factor based on the consistency evolution characteristics of the multimodal fusion signal before and after tuning during a continuous tuning cycle; please refer to [link to details]. Figure 4 The schematic diagram shows that the input terminal of the self-consistent tuning control unit 403 is connected to the data output terminals of the multimodal determination unit 402 and the two-factor constraint unit 401, and its output terminal is connected to the parameter input terminal of the two-factor joint solution unit 404. It should be noted that "consistency" refers to whether the joint distribution of the multimodal fused signal in the structural response space and environmental compensation space exhibits a higher degree of internal coordination after being subjected to the current tuning parameters. This consistency is reflected in the enhanced synergy of different modal signals in terms of change direction, change amplitude, and change rhythm.

[0075] During implementation, at the end of each tuning cycle, the self-consistent tuning control unit 403 calculates the consistency state quantity based on the fused signal after the current tuning parameters are applied. . The cross-modal consistency index of the multimodal fused signal within the current tuning cycle is calculated as follows: First, time window statistics are performed on the tuned multimodal fusion signal within the current tuning period to obtain the normalized feature sequence of each modal signal. Second, the correlation consistency coefficient between any two modal signals is calculated, and a weighted average is performed on all modal pairs. Finally, this weighted average is used as the consistency state quantity K(t) for the current tuning period.

[0076] In the program implementation, K(t) is usually stored as a double-precision floating-point variable, and its value range is limited to the interval [0,1] or [-1,1]. It should be noted that this application does not limit the specific mathematical construction form of K(t), as long as it can monotonically reflect the enhancement or weakening of the consistency of the multimodal fusion signal.

[0077] To avoid making tuning decisions based solely on changes in a single cycle, the previous tuning cycle is read from the local cache. and the first two tuning cycles The consistency state quantity is calculated, and the consistency driving quantity is calculated item by item according to the following formula. : in, This represents the consistency state quantity of the current tuning cycle; This represents the consistency state quantity of the previous tuning cycle. This represents the consistency state quantity for the first two tuning cycles. The consistency enhancement weight coefficient is used to adjust the system's response sensitivity to changes in the current consistency level. When the value is large, the system is more sensitive to immediate changes in consistency, such as increases or decreases, and it is usually taken as a positive value; This is the uniform curvature weighting coefficient, used to suppress oscillations or overtuning during the uniformity change process. It smooths the tuning process by introducing uniformity change curvature constraints. Tuning period time interval. It is determined by the system's operating cycle or sampling period and is set during the system initialization phase; Driven quantity It consists of first-order consistency difference terms and second-order consistency difference terms. The first-order terms reflect the immediate trend of consistency improvement or deterioration, while the second-order terms reflect the acceleration or decay characteristics of the consistency change rate.

[0078] Obtaining Consistency Drivers Subsequently, a two-factor tuning allocation function is constructed to determine the relative tuning contribution ratios of the structure mapping correction factor and the environmental offset compensation factor within the current tuning cycle. This allocation process is not a fixed-ratio allocation but rather a dynamic adjustment based on the current operating state. The formula for calculating the tuning control quantity is as follows: , in: The tuning control output is the structure mapping correction coefficient. Its sign is used to indicate the tuning direction, and its absolute value is used to characterize the tuning intensity. This indicates the corresponding adjustment amount for the environmental offset compensation coefficient; , These are the basic tuning gains for structural and environmental factors, respectively, which are used to limit the maximum magnitude of parameter variation under unit consistency driving quantity; A modulation factor is assigned to the tuning, which describes the relative proportion of structural tuning and environmental tuning under the current operating state, and satisfies the following conditions: This ensures the tuning energy is conserved and distributed between the two factors. As a possible implementation, when the system detects large fluctuations in external environmental parameters while the structural response remains stable, the energy level is increased. The proportion of values; conversely, when structural characteristics deviate significantly while environmental factors change relatively little, increasing The proportion of values.

[0079] To prevent the tuning process from continuing to output minute adjustments even when consistency has largely converged, thus causing parameter jitter, the self-consistency tuning control unit 403 sets a minimum consistency drive threshold. When the consistency drive amount... When the absolute value is lower than the threshold, the system determines that the current tuning has entered the uniform convergence interval, stops outputting the tuning control quantity, and maintains the parameter value of the previous tuning cycle unchanged.

[0080] Next, under the constraints of the tuned control parameter set, a two-factor collaborative solution unit is used to map the correction coefficients of the structure. Environmental offset compensation coefficient Joint updates and validity assessments are performed. This unit uses the tuning cycle as the basic execution unit. It should be noted that the two-factor collaborative solution differs from the traditional single-parameter update or independent optimization process. Its core lies in the fact that structural factors and environmental factors are not optimized separately, but are updated simultaneously within the same candidate solution space, and are evaluated as a whole through a unified collaborative consistency cost function.

[0081] Within the current tuning cycle, the two-factor collaborative solution unit first bases its solution on the structural tuning control quantity. Environmental tuning control quantity The candidate structure mapping correction coefficients and candidate environment offset compensation coefficients are constructed and can be expressed as follows: in, These are the parameter values ​​that have been confirmed to be effective in the previous tuning cycle. It should be noted that this application does not limit the generation of only one candidate solution; provided computational resources permit, multiple candidate solutions can be generated based on the above. Multiple sets of candidate solutions with limited amplitudes are generated, without any restrictions.

[0082] Obtain candidate parameter pairs ( After that, the system applies this parameter transiently to the original multimodal feature data within the current tuning cycle, and obtains a candidate multimodal fusion feature set through the fusion signal generation process. Where i represents the effective modal channel number participating in the collaborative solution, and the "effective modal channel" refers to a signal channel that is not frozen by the multimodal determination unit 402 and whose confidence level is higher than a preset threshold in the current tuning cycle. The number of effective modal channels is denoted as N.

[0083] It should be noted that the candidate multimodal fusion features are different from the final output fusion features. They are only used for internal consistency and stability evaluation within this unit and are not directly output externally.

[0084] For the candidate multimodal fusion feature set, the two-factor collaborative solution unit calculates the structure-environment collaborative consistency cost function, the expression of which is: Where N is the number of effective modal channels participating in the collaborative solution; This represents the fused feature vector obtained by the i-th modal channel under the action of the candidate parameters; Let be the center vector of the candidate fusion features, which is all The arithmetic mean of the terms; the first term of the formula is used to characterize the dispersion of multimodal fusion features under candidate parameters. The smaller the value, the higher the cross-modal consistency; the second and third terms are parameter evolution penalty terms, used to limit the variation of structural parameters and environmental parameters in adjacent tuning cycles; The time evolution constraint weight coefficient is used to balance the relationship between consistency improvement and parameter smoothness, and its value range is usually a non-negative real number.

[0085] It should also be noted that the “cooperative consistency cost function” is not a general loss function. Its characteristic is that it simultaneously introduces a cross-modal consistency term and a two-parameter time evolution term to prevent a single factor from dominating the tuning process.

[0086] Next The historical cost value is compared with the historical cost value stored in the system. The historical cost value includes at least the confirmation cost value of the previous tuning cycle and the upper limit of the stable interval obtained statistically over a period of time.

[0087] When satisfied When conditions are met, the system confirms the current candidate solution as a valid cooperative solution and sets it as such. Write the current parameter status, where This represents the historical best cost. When the following conditions are met... When the condition is met, the system determines that the current candidate update violates multimodal consistency or parameter evolution stability, discards the current candidate solution, and maintains the parameter state of the previous tuning cycle unchanged; whereby... This represents the upper limit of the historical stable range. In the event of the aforementioned rejection, the system simultaneously outputs a "parameter freeze" flag to the parameter evolution management module 600 to suppress subsequent continuous tuning attempts within a short period.

[0088] After the collaborative solution is completed, the two-factor collaborative solution unit will finally confirm the structure mapping correction coefficients. Environmental offset compensation coefficient The fused signal is normalized and the historical records in the parameter evolution management module 600 are updated simultaneously for differential calculation and stability constraints in subsequent tuning cycles.

[0089] Simultaneously, based on the obtained tuning parameter set, the tuning stability constraint unit 405 is invoked to verify the temporal continuity of the tuning results. The tuning stability constraint unit 405 takes the parameter sequence within multiple consecutive tuning cycles as input and calculates the rate of change for the structural mapping correction coefficient and the environmental offset compensation coefficient, respectively. The calculation of the rate of change is based on the ratio of the difference between parameter values ​​in adjacent tuning cycles to the reference parameter value, and its domain is limited to the parameter update range allowed by the system.

[0090] When the rate of change of any parameter exceeds a preset threshold within the current tuning cycle, the tuning stability constraint unit 405 determines that the parameter update lacks sufficient stability. This threshold can be configured based on cabinet type, operating voltage level, or historical operating experience to distinguish between normal gradual tuning and abnormal jump updates. In the event of instability, the tuning stability constraint unit 405 directly calls the parameter value from the previous tuning cycle stored in the parameter evolution management module 600 to replace the corresponding parameter in the current cycle, thereby blocking the impact of abnormal tuning results on the fused signal. It should be clarified that this parameter replacement is a time-continuous effectiveness screening method, aiming to ensure the traceability and engineering controllability of tuning parameters during long-term operation.

[0091] The status analysis module 500 is used to assess the status of signal distribution in the feature space based on the fused signal set, extract operating status features, and generate a high- and low-voltage switchgear operating status dataset. It should be noted that the purpose of this module is not to make limit judgments for a single physical quantity, but to make a statistical assessment of the operating status of the high- and low-voltage switchgear based on the overall distribution of multi-modal signals in a unified feature space, thereby reducing false alarms and missed alarms caused by differences in cabinet structure or operating environment. In implementation, the status analysis module 500 is deployed in the edge computing unit or centralized analysis server of the monitoring system, and its input is connected to the data output interface of the self-tuning processing module 400.

[0092] In the specific implementation process, the state analysis module 500 first maps the tuned multimodal fusion signal to the same operating feature space based on a pre-set unified feature mapping rule. It should be noted that the purpose of the "unified feature mapping rule" is not to eliminate numerical differences, but to ensure that different modal signals have comparable statistical semantics after mapping. As one possible implementation, the mapping rule may include a combination of amplitude normalization, energy scale transformation, and statistical feature extraction within a time window. For example, within a fixed operating time window, the mean, variance, energy density, or rate of change of each modal signal may be extracted and arranged into a feature vector in a predetermined order. The specific mapping method used can be determined according to actual monitoring needs and computing resources, and is not limited to a single method.

[0093] After completing the feature mapping, the state analysis module 500 represents the set of multimodal feature vectors within the current operating cycle as a set of feature points, where each feature point corresponds to a fused operating state description within a time window. This set of feature points is not used for instantaneous judgment, but rather serves as an input sample set for statistical analysis to characterize the overall distribution of the current operating state in the feature space.

[0094] Subsequently, the state analysis module 500 performs spatial distribution statistical analysis on the feature point set to extract distribution feature quantities describing the operating state of the high and low voltage switchgear. In implementation, these distribution feature quantities include at least the feature center location, dispersion, and distribution directionality parameter. The feature center location reflects the overall bias level of the current operating state in the feature space; the dispersion reflects the stability or fluctuation intensity of the operating state; and the distribution directionality parameter reflects whether there is a consistent evolutionary trend among the multimodal features. It should be noted that the "distribution directionality" differs from traditional signal trend analysis; it emphasizes the main directional change of multidimensional features under the covariance structure, used to identify systematic shifts rather than random disturbances.

[0095] After obtaining the aforementioned distribution characteristics, the state analysis module 500 calculates the corresponding state offset based on the statistical distribution description of the current operating cycle. Specifically, the module first calculates the feature center vector and covariance matrix based on the feature point set within the current operating cycle to characterize the statistical distribution structure of the current operating state. Simultaneously, it reads the pre-established standard operating state feature center vector and covariance matrix from the system initialization phase or historical stable operation phase as the benchmark distribution model of the standard state. The "standard operating state" is not a fixed threshold state, but a statistical benchmark formed for a specific cabinet under normal operating conditions, which can be gradually corrected as the parameter evolution management module 600 updates.

[0096] Based on this, the module uses a covariance-weighted distance metric to quantify the overall deviation of the current operating state distribution from the standard operating state distribution. Unlike simple Euclidean distance, the covariance-weighted distance introduces correlation constraints between various feature dimensions during the calculation process. This ensures that changes in highly volatile but consistent directions are not misjudged as abnormal, while changes in low-volatile directions that deviate from the main distribution structure are effectively amplified, thus better reflecting the operating state evolution characteristics in actual engineering. The obtained weighted distance value is defined as the state offset, used to represent the degree of deviation of the current operating state of the high- and low-voltage switchgear from the standard operating state in the feature space.

[0097] During the status determination phase, the status analysis module 500 does not directly draw conclusions based on the status offset of a single operating cycle. Instead, it comprehensively judges the changing trend of the status offset in conjunction with the time dimension. When the status offset is within a preset stable range and does not show a continuously increasing evolution trend in consecutive operating cycles, the high and low voltage switchgear is determined to be in normal operating condition. When the status offset exceeds the corresponding threshold and shows a consistent cumulative change trend in multiple consecutive operating cycles, the high and low voltage switchgear is determined to have entered an abnormal operating state or a deteriorated operating state.

[0098] After completing the status assessment, the status analysis module 500 encapsulates the status determination result and the corresponding distribution characteristic quantity into a high- and low-voltage switchgear operating status feature set, which is output as an operating status dataset. This operating status dataset can not only be used for alarm display or operation and maintenance decisions, but also serve as input for the parameter evolution management module 600 and subsequent analysis modules, providing data support for the long-term stable operation and adaptive capability of the system. As one possible implementation, the operating status dataset may include a status level identifier, a status offset value, a distribution characteristic summary, and corresponding timestamp information. The specific field settings can be adjusted according to actual application requirements and are not limited thereto.

[0099] It should be further clarified that the state analysis module 500 described in this invention is not a simple application of existing statistical analysis or pattern recognition methods. Its technical purpose, processing objects, and judgment logic are fundamentally different from existing technologies. In traditional high and low voltage switchgear monitoring systems, state judgment is usually based on the absolute numerical changes of several physical quantities, or by directly fusing and analyzing multi-source signals without structural and environmental corrections. This approach is easily affected by differences in cabinet manufacturing, heat dissipation structures, and operating environments in actual engineering, thus misjudging structural or environmental deviations as equipment anomalies. The state analysis module 500 of this invention operates after the multimodal signals have undergone structural mapping correction and environmental offset compensation. Based on two-factor self-tuning signal processing, it addresses the engineering challenge of maintaining consistent judgment standards for multimodal monitoring under different cabinet and environmental conditions. It proposes a state assessment technical solution with clear engineering orientation and feasibility, thereby significantly improving the stability, comparability, and long-term operational reliability of monitoring results.

[0100] The parameter evolution management module 600 is used to store the structural mapping correction coefficients and environmental offset compensation coefficients in time series, establish a tuning parameter evolution model, and output a set of reference parameters for subsequent periodic tuning calculations. It should be noted that the "parameter evolution" focuses not on the result of a single optimization, but on whether the trajectory of parameter changes over a continuous operating cycle conforms to the objective laws of equipment operation and environmental changes, and can reflect the long-term stable state of the system.

[0101] In practical implementation, the parameter evolution management module 600 first performs time-series management on the structural mapping correction coefficients and environmental offset compensation coefficients output by the self-tuning processing module 400 within each operating cycle. The time series uses the operating cycle as the basic time granularity, and each parameter sample is associated with a corresponding operating state label. These operating state labels include at least three categories: steady-state operation, transitional operation, and abnormal operation. The operating state labels can directly reuse the output results of the state analysis module 500, and the specific details can be determined according to the actual system configuration.

[0102] After constructing the parameter time series, not all parameter samples participate in subsequent evolutionary modeling. The parameter evolution management module 600 performs validity screening on the parameter samples to remove tuning parameters generated during non-steady-state operation or abnormal operating conditions. As one possible implementation, when the switchgear is in a state of rapid load switching, startup, shutdown, or has been determined by the state analysis module 500 to be in an abnormal operating state, the tuning parameter samples in the corresponding period are marked as invalid samples and are not included in the valid tuning parameter sequence. This screening mechanism can prevent short-term disturbances or abnormal states from contaminating the long-term parameter evolution model.

[0103] After obtaining the effective tuning parameter sequence, the parameter evolution management module 600 performs variation amplitude analysis and variation rate analysis on the structural mapping correction coefficient and the environmental offset compensation coefficient, respectively. Variation amplitude analysis is used to characterize the absolute range of parameter value changes within adjacent operating cycles, while variation rate analysis is used to characterize the speed and continuity of parameter changes. The variation rate is a discrete variation index based on the operating cycle, and its domain is defined between adjacent effective operating cycles.

[0104] Based on the above analysis results, the parameter evolution management module 600 segments the parameter time series to identify stationary evolution segments and abrupt shift segments. The stationary evolution segment refers to the time interval during which the parameter's amplitude and rate of change remain within a preset reasonable range over multiple consecutive operating cycles. This reasonable range can be set based on equipment type, cabinet structure differences, and historical operating experience; the specific value is not limited. The abrupt shift segment refers to the time interval during which the parameter's amplitude or rate of change exceeds this reasonable range. Through this differentiation process, occasional abnormal tuning results can be effectively separated from representative long-term tuning trends.

[0105] For the parameters of the identified stationary evolution phase, the parameter evolution management module 600 constructs a parameter evolution constraint model to reflect the reasonable variation boundaries of the parameters under long-term operating conditions. This evolution constraint model does not require the parameters to remain constant, but rather limits their allowable variation range in subsequent operating cycles, ensuring the tuning process is "adjustable but not out of control." The evolution constraint model can take the form of interval boundaries or a constraint form based on historical weights; the specific implementation method can be determined according to the system complexity.

[0106] Upon entering a new tuning cycle, the parameter evolution management module 600 does not directly output a historical parameter value. Instead, based on the current operating conditions, it extracts a subset of parameters from the historical valid tuning parameter sequence that matches the current operating state. These operating conditions include at least the current load level, environmental state category, and operating stability indicator. For the extracted parameter subset, the parameter evolution management module 600 performs interval constraint or weighting processing to generate a reference parameter set for subsequent tuning calculations. It should be noted that the reference parameter set is essentially a set of evolution-constrained feasible parameter ranges or parameter distributions, used to guide the subsequent tuning process to search within a reasonable solution space.

[0107] Finally, it should be noted that the mathematical formulas, derivations, symbol definitions, and parameter calculation methods used in this specification are all for the purpose of further clarifying and verifying the technical content of this invention, so that those skilled in the art can more intuitively and accurately understand the working mechanism and technical effects of this invention. These formulas are only used as quantitative expressions or illustrative examples of technical features and do not constitute limiting conditions of the claims of this invention. Those skilled in the art should understand that, without changing the core idea of ​​this invention, the parameter forms, calculation methods, numerical ranges, and even symbol representations involved in the formulas can be equivalently replaced or simplified in engineering according to the actual application environment. The specifics can be determined according to the actual situation, and no limitation is imposed. It should also be emphasized that the formulas in this specification are not theoretical derivations in the style of academic research papers, but rather an engineering description of the embodiments of this invention. Their purpose is to enhance the understandability and implementability of this invention, rather than to increase redundancy and complexity. Those skilled in the art can choose whether to use such quantitative tools when reading this specification, or can achieve the same technical effects through other equivalent methods.

[0108] Furthermore, while specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A high- and low-voltage switchgear monitoring system with multi-modal sensor fusion; characterized in that: include: The acquisition module is used to collect temperature, current, voltage, partial discharge, humidity and vibration signal data of high and low voltage switchgear in real time and generate multimodal signal datasets. The first feature modeling module is used to perform feature modeling on the signal response distribution of high and low voltage switchgear under standard operating conditions based on multimodal signal datasets, extract the structural mapping parameters of the cabinet, and generate a structural feature matrix. The second feature modeling module is used to acquire external operating condition data of high and low voltage switchgear, model the variation law of environmental parameters, form an environmental factor matrix, and output environmental compensation parameters. The self-tuning processing module is used to perform dual-channel difference analysis based on the structural feature matrix and the environmental factor matrix, calculate the structural mapping correction coefficient and the environmental offset compensation coefficient, and obtain the tuned fusion signal set. The status analysis module is used to evaluate the status of signal distribution in the feature space based on the fused signal set, extract operating status features, and generate a high and low voltage switchgear operating status dataset. The parameter evolution management module is used to store the structural mapping correction coefficients and environmental offset compensation coefficients in time series, establish a tuning parameter evolution model, and output a reference parameter set for subsequent periodic tuning calculations.

2. The high and low voltage switchgear monitoring system with multimodal sensor fusion according to claim 1, characterized in that: The working process of the first feature modeling module includes: Time alignment and cross-correlation function analysis are performed on the multimodal signal dataset to determine the response time difference and amplitude correlation between each signal, and to generate a nodal response feature set. Based on the node response feature set, the vibration power spectral density, current fluctuation rate and temperature gradient of each structural node in the steady state stage are calculated as the structural response parameter set. The set of structural response parameters is arranged in the order of node spatial distribution to form a structural feature matrix, and the comprehensive response characteristics of each node under steady-state operation are recorded.

3. The high and low voltage switchgear monitoring system with multimodal sensor fusion according to claim 1, characterized in that: The working process of the second feature modeling module includes: Collect external meteorological temperature, relative humidity and atmospheric pressure signals of high and low voltage switchgear, and simultaneously collect signals of air inlet and outlet temperature, internal humidity and airflow speed of the cabinet; Calculate the temperature difference, humidity difference, and airflow velocity difference between the outside and inside based on the same time window to obtain temperature gradient parameters, humidity gradient parameters, and airflow coupling parameters. The above parameters are normalized and time-weighted to generate an environmental factor matrix that reflects changes in the external environment and the heat and moisture transfer characteristics of the internal cavity. Temperature difference correction coefficient, humidity correction coefficient, and airflow correction coefficient are extracted from the environmental factor matrix to obtain the environmental compensation parameter set.

4. The high and low voltage switchgear monitoring system with multimodal sensor fusion according to claim 1, characterized in that: The self-tuning processing module includes a two-factor constraint unit, a multi-modal determination unit, a self-consistent tuning control unit, a two-factor joint solution unit, and a tuning stability constraint unit. The two-factor constraint unit is used to establish the mutual constraint relationship between the structural mapping characteristics and the environmental compensation characteristics within the same tuning period. By limiting the joint variation range of the structural response correction amplitude and the environmental compensation gain, a two-factor feasible tuning domain is formed. The multimodal determination unit is used to dynamically determine the tuning participation level of each signal channel in the current operating stage based on the statistical stability and cross-modal correlation consistency of the multimodal signal before and after tuning, generate a set of tunable channels and a set of frozen channels, and use the determination result as a constraint condition for subsequent parameter solving. The self-consistent tuning control unit is used to adaptively adjust the update step size of the structure mapping correction coefficient and the environmental offset compensation coefficient according to the signal self-consistency change trend, and generate two-factor tuning intensity and direction control quantities. The dual-factor joint solution unit is used to perform joint parameter solution of the structural mapping correction coefficient and the environmental offset compensation coefficient under the common constraints of the dual-factor feasible tuning domain and the tunable channel set, and generate a self-tuning parameter set corresponding to the tuning period. The tuning stability constraint unit is used to calculate the rate of change of the structure mapping correction coefficient and the environmental offset compensation coefficient within a continuous tuning cycle. When the rate of change exceeds a preset threshold, the parameter value of the current tuning cycle is replaced with the corresponding parameter value of the previous tuning cycle.

5. A high- and low-voltage switchgear monitoring system with multi-modal sensor fusion according to claim 4, characterized in that: The multimodal decision unit employs a joint reliability modeling method based on multi-scale time consistency, cross-modal energy coupling, and tuning sensitivity to dynamically evaluate the reliability of each modal signal channel. The evaluation process includes: Within multiple consecutive tuning cycles, for the first A confidence state vector is constructed from each modal signal channel. The state vector includes a short-time stability component, a long-time drift component, a cross-modal coupling component, and a tuning-sensitive component; Based on the above components, a comprehensive reliability function is constructed, and the function expression is: in, Indicates the first The overall reliability of each modal signal channel within the current tuning cycle; The first element in the credibility state vector represents the... One component; These are the weighting coefficients for each component; This is a nonlinear mapping function used to suppress outliers; When the overall credibility When the modal signal channel exhibits a monotonically decreasing trend over multiple consecutive tuning cycles, it is divided into a frozen channel set to reduce its participation weight in the two-factor collaborative solution process or freeze the tuning participation state.

6. A high- and low-voltage switchgear monitoring system with multimodal sensor fusion according to claim 4, characterized in that: The process by which the self-consistent tuning control unit generates the two-factor tuning intensity and direction control quantities includes: Within a continuous tuning period, the consistency state quantity is calculated for the multimodal fused signal after being subjected to the current tuning parameters. And a consistency driving quantity is constructed based on the consistency difference between adjacent tuning cycles. The expression is: in, The consistency enhancement weighting coefficient is used to reflect the extent of consistency improvement brought about by the current tuning. This is the uniform curvature weighting coefficient, used to reflect the accelerating or decaying trend of uniformity changes; The tuning period time interval; Based on the consistency driving quantity A two-factor tuning allocation function is constructed to adaptively allocate the tuning contribution ratios of the structure mapping factor and the environmental compensation factor. The function expression is as follows: in, The output of the tuning control is the structural mapping correction coefficient; The output of the tuning control is the environmental offset compensation coefficient; These are the basic tuning gains for structural factors and environmental factors, respectively. The modulation factor is a constraint on the structural and environmental tuning sensitivity based on the current operating state, and satisfies... ; When the consistency driving quantity When the absolute value is lower than the preset minimum drive threshold, it is determined that the current tuning has entered the uniform convergence interval, and the output of the tuning control quantity is stopped; otherwise, the output includes... and The set of tuning control parameters.

7. A high- and low-voltage switchgear monitoring system with multimodal sensor fusion according to claim 4, characterized in that: The processing procedure of the two-factor collaborative solution unit includes: Within the current tuning cycle, candidate updated solutions are constructed based on the tuning control parameter set, using the structure mapping correction coefficients and environmental offset compensation coefficients, to generate a candidate multimodal fusion feature set. ; based on Computational structural environment coherence cost function The calculation formula is: Where N is the number of effective modal channels participating in the collaborative solution; The center vector of the candidate fusion feature; These are the candidate structure mapping correction coefficient and the environmental offset compensation coefficient, respectively. These are the weighting coefficients of structural factors and environmental factors in the time evolution constraint, respectively; The tuning period time interval; The When compared with the corresponding generation value of the previous tuning cycle, when When the cost is less than the historical best, the current candidate updated solution is confirmed as a valid collaborative solution, and the structural mapping correction coefficient and environmental offset compensation coefficient are updated. when When the current solution is higher than the upper limit of the historical stable range, it is determined that the current dual-factor collaborative update violates the multimodal consistency or parameter evolution stability, and the current candidate solution is abandoned while maintaining the parameter state of the previous tuning cycle. After the collaborative solution is completed, the confirmed structural mapping correction coefficients and environmental offset compensation coefficients are output as the self-tuning parameter solution set.

8. A high- and low-voltage switchgear monitoring system with multi-modal sensor fusion according to claim 1, characterized in that: The processing procedure of the state analysis module is as follows: Based on the fused set of tuned signals, a multimodal operation feature space is constructed under a unified feature mapping rule, and the mapping result of each tuned signal in this feature space is represented as a set of feature points. Statistical analysis is performed on the spatial distribution pattern of the feature point set to extract distribution feature quantities that represent the operating status of high and low voltage switchgear. The distribution feature quantities include feature center location, dispersion degree and distribution directionality parameters. Based on the distribution feature quantity, the state offset within the current operating cycle is obtained and compared with the feature distribution benchmark under the standard operating state to evaluate the degree of deviation of the current operating state from the standard state. When the state offset is within a preset stable range, the high and low voltage switchgear is determined to be in normal operation; when the state offset exceeds the corresponding threshold and shows a continuous evolution trend, the high and low voltage switchgear is determined to enter an abnormal operation state or a deteriorated operation state. The state assessment results and the corresponding distribution characteristics are combined to form the high and low voltage switchgear operating state feature set, which is output as the operating state dataset.

9. A high- and low-voltage switchgear monitoring system with multimodal sensor fusion according to claim 8, characterized in that: The state offset is calculated from the statistical difference between the multimodal fusion feature distribution under the current operating cycle and the standard operating state feature distribution. The calculation process includes: Based on the set of multimodal fusion feature points in the current operating cycle, calculate its feature center vector and covariance matrix to form a statistical distribution description of the current operating state; Based on the feature center vector and covariance matrix pre-established under standard operating conditions, a benchmark distribution model for standard operating conditions is formed. The overall deviation of the current operating state distribution from the standard operating state distribution is quantitatively calculated using a covariance-weighted distance metric, and the resulting weighted distance value is used as the state deviation. The state deviation is used to represent the degree of deviation of the current operating state of the high and low voltage switchgear from the standard operating state in the feature space.

10. A high- and low-voltage switchgear monitoring system with multi-modal sensor fusion according to claim 1, characterized in that: The parameter evolution management module manages the variation law of tuning parameters across operating cycles through a management and constraint mechanism based on the time evolution characteristics of tuning parameters. The process includes: For the structural mapping correction coefficients and environmental offset compensation coefficients output by the self-tuning processing module in each operating cycle, a parameter time series is constructed in chronological order. The validity of the parameter time series is then screened, and parameter samples corresponding to unsteady operating phases or abnormal operating states are removed to form a valid tuning parameter series. Based on the effective tuning parameter sequence, the variation amplitude analysis and variation rate analysis of the structural mapping correction coefficient and the environmental offset compensation coefficient are performed respectively to identify the evolution trend characteristics of the parameters in the continuous operation cycle and to distinguish between the stationary evolution segment and the abrupt offset segment of the parameters. For the parameters of the identified stationary evolution segment, an evolutionary constraint model reflecting the long-term variation law of the parameters is constructed to limit the reasonable variation range of the tuning parameters in subsequent operating cycles; When entering a new tuning cycle, based on the evolution constraint model, a subset of parameters that meet the current operating conditions is extracted from the historical effective tuning parameter sequence, and interval constraints or weighting are applied to it to generate a reference parameter set for subsequent periodic tuning calculations.

Images