Electrical complete equipment state monitoring and early warning system based on multi-parameter intelligent sensing

Abstract

The invention discloses an electrical complete equipment state monitoring and early warning system based on multi-parameter intelligent sensing, belongs to the technical field of electrical equipment state monitoring, and can clearly point out a specific fault mode (such as contact resistance increasing overheating) and a possible position (such as an A-phase bus connection point). Through multi-parameter time sequence feature fusion and an intelligent diagnosis model, deep mining of a fault source is realized, the operation and maintenance efficiency is greatly improved, and the maintenance work is turned from blind troubleshooting to accurate disposal. Weak precursor signals of early and slow faults can be captured by extracting depth time sequence characteristics (such as temperature rise rate and dominant frequency offset rate) strongly related to a fault evolution mechanism. The AI model is utilized to learn complex modes of these precursor, and early warning can be given out before the equipment performance is obviously degraded, so that predictive maintenance of'nipping in advance 'is realized, and unplanned shutdown and major accidents are effectively avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical equipment state monitoring, and in particular to an electrical complete equipment state monitoring and early warning system based on multi-parameter intelligent sensing. BACKGROUND

[0002] Electrical complete equipment, such as power distribution cabinets, control cabinets, etc., is the key equipment for realizing power distribution, control and protection in the power system, and its operation reliability is directly related to the safety and stability of the entire power supply system. At present, the state monitoring technology in this field is developing from single parameter monitoring to multi-parameter sensing, but the existing technical solutions still have significant deficiencies in realizing true multi-parameter intelligent sensing and intelligent early warning.

[0003] For example, existing technologies, such as the invention patent with publication number CN120652964A, provide a multi-sensor fusion method, the core of which is to improve the reliability of data by dynamically assigning sensor weights and performing trend consistency verification. Although this method takes into account the reliability of the sensor itself, the final output is a single state indicator obtained by weighted averaging, which lacks the ability to diagnose specific fault modes of the equipment and cannot inform the operation and maintenance personnel of the problem with the equipment, and is not sensitive to the weak signs of early and slow faults.

[0004] For another example, another existing technology, such as the invention patent with publication number CN120498129A, provides an edge computing and cloud collaboration system, which focuses on improving the real-time and reliability of data acquisition through hardware filtering and architecture optimization. However, the core risk assessment module is essentially a deviation degree calculation based on preset thresholds, and it fails to train an intelligent model that can learn the complex patterns of "normal" and "pre-fault signs" using historical data, so it is difficult to achieve early warning based on pattern recognition.

[0005] In summary, existing technologies mostly focus on data level fusion and optimization, or focus on improving the computing architecture, and fail to effectively solve the core problem of "how to mine the early evolution rules of faults from multi-parameter time series data and achieve accurate fault mode identification and early warning". Therefore, there is an urgent need for a new intelligent monitoring solution that can deeply integrate multi-parameter time series information and has the ability of early fault diagnosis and positioning. SUMMARY

[0006] The present application aims to overcome the deficiencies of the prior art, and provides a state monitoring and early warning system and method for electrical complete equipment based on multi-parameter intelligent sensing.

[0007] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides a state monitoring and early warning method for electrical complete equipment based on multi-parameter intelligent sensing, comprising the following steps: S1: synchronously collecting original time series data of electrical parameters, temperature parameters, partial discharge parameters, mechanical vibration parameters and environmental parameters in a multi-mode adaptive acquisition manner through a multi-sensor array deployed at key nodes of the electrical complete equipment; S2: pre-processing, filtering and time stamp alignment based on a hardware clock synchronization protocol on the edge computing gateway side for the original time series data; S3: extracting a deep time series feature set strongly related to the fault mechanism from the pre-processed time series data of each parameter; S4: fusing the deep time series features from different parameters to construct a unified multi-parameter fusion feature vector; S5: inputting the multi-parameter fusion feature vector into a pre-trained intelligent diagnosis model to output the real-time health index of the equipment and the probability of each main fault mode, and evaluating the confidence level of the diagnosis result; S6: based on the diagnosis result and its confidence level, performing a hierarchical early warning and decision degradation process to generate and publish early warning information with clear fault directionality.

[0008] Preferably, in step S3, the deep time series feature set includes: extracting the temperature rise trend slope based on linear regression and the dynamic time warping distance from historical data from the temperature parameters; extracting the energy of a specific fault sensitive frequency band, the main frequency component and its shift rate, and the envelope spectrum analysis feature from the vibration parameters; extracting the pulse repetition rate and the statistical moment feature of the discharge phase distribution map from the partial discharge parameters; extracting the three-phase unbalance degree, current harmonic distortion rate and load rate fluctuation feature from the electrical parameters.

[0009] Preferably, step S4 includes: The recursive graph, Gram angle field or Markov transition field method is used for feature assimilation of the heterogeneous time sequence data, so as to convert the time sequence data to a unified feature expression space; The assimilated deep time sequence features are spliced and processed by using the Z-score standardization or Min-Max normalization method, so as to construct the multi-parameter fusion feature vector.

[0010] Preferably, in the step S5, the intelligent diagnosis model adopts the gradient boosting decision tree, the random forest, the long short-term memory network, the support vector machine or the one-dimensional convolutional neural network algorithm; and the confidence level of the evaluation diagnosis result is specifically realized by using the Monte Carlo Dropout method or by calculating the divergence of the integrated learning model prediction result.

[0011] Preferably, the decision degradation process in the step S6 includes: When the confidence level is high, a precise diagnosis early warning is triggered; When the confidence level is medium, an abnormal trend reminder is degraded, and manual review is suggested; When the confidence level is low or a serious conflict of multi-parameter information is perceived, a data quality alarm is further degraded, and a conservative control strategy is triggered.

[0012] Preferably, the system further includes a sensor health state evaluation step: The health score of each sensor is dynamically calculated by analyzing the self-noise level, output stability and time relationship with the maintenance record of the sensor data; The health score is used as the data reliability weight to dynamically participate in the feature fusion in the step S4 and the model diagnosis process in the step S5.

[0013] In a second aspect, the application provides an electrical complete equipment state monitoring and early warning system based on multi-parameter intelligent perception. The system includes: A multi-parameter intelligent perception layer includes a sensor array based on fault mode and effect analysis and fixedly deployed at key nodes of the electrical complete equipment, which is used for collecting original time sequence data of electrical parameters, temperature parameters, partial discharge parameters, mechanical vibration parameters and environmental parameters; An edge computing and data acquisition layer includes an edge computing gateway, which is used for receiving and pre-processing the original time sequence data, time synchronization based on IEEE 1588 PTP or GPS time service, and simple feature calculation on the edge side; A cloud intelligent diagnosis and early warning platform is used for storing full-amount historical data, and running a multi-parameter time sequence feature fusion module and an intelligent diagnosis and early warning module.

[0014] Preferably, the intelligent diagnosis and early warning module includes: A fault diagnosis model trained based on a machine learning algorithm, configured to receive the multi-parameter fusion feature vector and output a device comprehensive health index and a specific fault mode probability; An uncertainty quantification unit configured to evaluate a confidence level of the diagnosis result; A decision degradation unit configured to trigger a hierarchical early warning process according to the confidence level.

[0015] In a third aspect, the present application provides an electronic device, comprising a processor, a memory, a sending device, an input device and an output device, wherein the memory stores a computer program, and when the computer program is executed by the processor, the method as described above is implemented.

[0016] In a fourth aspect, the present application provides a computer readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the method as described above is implemented.

[0017] Compared with the prior art, the present application has the following advantages: Unlike the prior art which only outputs a single "abnormal" or "risk value", the present application can explicitly indicate the specific fault mode (such as "increased contact resistance overheat") and its possible location (such as "A-phase bus connection point"). Through multi-parameter time sequence feature fusion and intelligent diagnosis model, the root cause of the fault is deeply excavated, which greatly improves the operation and maintenance efficiency, and changes the maintenance work from "blind troubleshooting" to "precise disposal".

[0018] By extracting deep time sequence features (such as temperature rise rate, main frequency offset rate) that are strongly related to the fault evolution mechanism, the weak precursor signals of early and slow faults can be captured. Using AI model to learn the complex patterns of these precursors can issue an early warning before the device performance deteriorates significantly, realizing predictive maintenance to "nip in the bud", effectively avoiding unplanned shutdown and major accidents.

[0019] The sensor health state evaluation module is innovatively introduced, which can dynamically evaluate the reliability of the sensor itself and participate in data fusion and diagnosis decision as a weight. When the performance of part of the sensors degrades or the data is abnormal, the system can automatically reduce the data weight of the sensors, rely on other healthy sensors to maintain effective diagnosis, and significantly improve the stability of long-term operation in complex and harsh industrial environments.

[0020] The uncertainty quantification and decision degradation mechanism is introduced, so that the system has "self-knowledge". When the diagnosis confidence is high, the accurate conclusion is output; when the data conflict or the confidence is low, the more conservative early warning strategy (such as triggering manual review) is automatically degraded, rather than forcibly outputting unreliable results. This effectively balances the accuracy of the early warning and the overall safety of the system, and has high engineering practical value.

[0021] This invention provides a complete solution integrating the perception layer, edge layer, and cloud platform, forming an intelligent closed loop from data acquisition to operation and maintenance decision-making. The output results are intuitive and operable, and can be seamlessly integrated into existing power distribution automation systems, significantly improving the safety, reliability, and availability of electrical equipment, and possessing immense value for engineering applications and market promotion. Attached Figure Description

[0022] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0023] 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, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the overall architecture of the system according to Embodiment 1 of the present invention.

[0025] Figure 2 This is a flowchart of the method according to Embodiment 1 of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0027] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0028] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" and "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. When the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0029] Example 1 like Figure 1 As shown, this invention provides a condition monitoring and early warning system for electrical complete sets of equipment based on multi-parameter intelligent sensing. The system includes: The multi-parameter intelligent sensing layer includes sensor arrays deployed at key nodes of electrical complete equipment (such as circuit breaker contacts, busbar connections, and cable joints) to collect raw time-series data of electrical parameters, temperature parameters, partial discharge parameters, mechanical vibration parameters, and environmental parameters. The edge computing and data acquisition layer, including the edge computing gateway, is used to receive, preprocess, synchronize time, and perform simple feature calculations on the edge side of the perception layer data. The cloud-based intelligent diagnosis and early warning platform is used to store all historical data and run a multi-parameter time-series feature fusion module and an intelligent diagnosis and early warning module.

[0030] The multi-parameter intelligent sensing layer refers to the comprehensive and multi-faceted perception of the equipment's operating status through a collection of various types and quantities of sensors deployed in key parts of the equipment. Its core lies in the diversity and synergy of perception, rather than the monitoring of a single parameter.

[0031] The deployment of sensor arrays can adopt a "fixed-point deployment scheme based on Failure Mode and Effects Analysis (FMEA)," which calculates the optimal installation location and number of sensors precisely based on historical fault data and equipment physical models to achieve the most effective capture of target fault modes.

[0032] The acquisition of raw time-series data can be conducted in a "multi-mode adaptive acquisition" mode. Specifically, this includes: continuous sampling (for electrical parameters and vibration), low-speed periodic sampling (for temperature and environmental parameters), and event-triggered acquisition (for transient signals such as partial discharge; high-speed recording is initiated when the signal exceeds a threshold). The edge computing gateway can coordinate the acquisition rhythm of different sensors to achieve synchronous high-density acquisition centered on key events.

[0033] The edge computing and data acquisition layer, as a localized computing unit close to the equipment site, is responsible for the initial processing of data, aiming to reduce the pressure on the cloud and improve the system's real-time performance. Its core function is to "bridge" the gap between the upper and lower layers, connecting with sensors and serving the cloud platform.

[0034] In addition to cleaning and filtering, the preprocessing may also include data interpolation (to compensate for missing data points), outlier removal, etc., which will not be elaborated here.

[0035] The time synchronization can be achieved using a hardware clock synchronization protocol (such as IEEE 1588 PTP) or GPS timing to achieve microsecond-level synchronization accuracy, providing a solid foundation for multi-parameter data alignment.

[0036] In the simplified feature calculation at the edge, the simplified features may specifically include: moving average, effective value (RMS), peak value, fundamental frequency energy under short-time Fourier transform (STFT), etc. The purpose is to complete data compression and initial feature screening at the edge, and to achieve millisecond-level rapid local alarm for abnormalities exceeding thresholds (such as instantaneous temperature exceeding limits), forming the system's first line of defense for rapid response.

[0037] Preferably, the multi-parameter time-series feature fusion module is responsible for transforming the original, heterogeneous time-series data into a high-dimensional feature vector that can comprehensively characterize the health status of the device, and is configured to perform the following operations: T1. Extract a deep time-series feature set that characterizes the evolution of equipment state from the preprocessed time-series data of various parameters. This deep time-series feature set includes not only conventional statistical features (such as mean and maximum values) but also higher-order features strongly correlated with the extraction and fault mechanisms, such as, but not limited to: Extract the rate of temperature rise and the temperature difference between the same period and the same phase from temperature parameters, such as "the slope of the temperature rise trend based on linear regression" and "the dynamic time warping (DTW) distance from historical data of the same period"; Extract the energy, dominant frequency components, and offset rates of specific fault-sensitive frequency bands from vibration parameters; for example, "envelope spectrum analysis features" can be used to more effectively extract the fault frequencies of components such as bearings and gears.

[0038] Statistical features of pulse repetition rate and discharge phase distribution (PRPD spectrum) are extracted from partial discharge parameters; for example, statistical moments (such as skewness and kurtosis) of the discharge quantity-phase-number (φ-qn) spectrum are used to quantify the morphological changes of discharge modes.

[0039] Extract three-phase unbalance, current harmonic distortion rate, and load rate fluctuation characteristics from electrical parameters.

[0040] T2. A feature assimilation mechanism is introduced. For heterogeneous data (such as structured time-series data and unstructured image data), methods such as recurrence plots are used to transform them into a unified feature representation space, laying the foundation for subsequent deep fusion. A recurrence plot is a method that visualizes time series data as a two-dimensional image, allowing feature extraction using image processing techniques. Besides recurrence plots, Gramian Angular Fields, Markov Transition Fields, or Wavelet Transforms can also be used to extract joint time-frequency features. The essence of these methods is to transform the time-series signal into a new feature space so that it can be fused with data from other modalities (such as images) or features with different physical meanings.

[0041] T3. The deep temporal features extracted from different physical quantities are concatenated and standardized to construct a unified, high-dimensional multi-parameter fusion feature vector. This vector comprehensively represents the overall health status of the device within a specific time window. The standardization can be "Z-score standardization" or "Min-Max normalization" to eliminate the influence of different dimensions and orders of magnitude.

[0042] Preferably, the intelligent diagnosis and early warning module serves as the "brain" of the system, utilizing an artificial intelligence model to analyze the fused feature vectors to achieve status assessment, fault diagnosis, and risk warning. In this embodiment, the intelligent diagnosis and early warning module is a fault diagnosis model trained based on machine learning algorithms (including but not limited to: Gradient Boosting Decision Tree (GBDT), Long Short-Term Memory Network (LSTM), Support Vector Machine (SVM), Random Forest, One-Dimensional Convolutional Neural Network (1D-CNN), etc., to better capture key information in long sequences), and is configured as follows: P1. Taking the multi-parameter fusion feature vector as input, the output includes diagnostic results including the overall equipment health index and the probability of specific fault modes. The fault diagnosis model is trained with a large amount of historical data (including normal state and various fault states) and can identify early feature patterns of various typical faults such as "overheating due to increased contact resistance", "deterioration of insulation dielectric strength" and "loose mechanical connection".

[0043] P2. An uncertainty quantification method is introduced, where the model assesses the confidence level of diagnostic results while outputting them. When the system detects serious conflicts among multiple parameters (i.e., inconsistencies in perception) or when the confidence level falls below a preset threshold, a pre-defined tiered warning and decision downgrade process is triggered. For example, the warning is downgraded from "accurate diagnosis warning" to "abnormal trend alert," and a more conservative operational strategy is recommended to prioritize system security in complex and uncertain scenarios. This could involve increasing the frequency of data feedback, triggering manual review, or implementing conservative security measures (such as recommending load reduction), rather than blindly trusting low-confidence diagnostic conclusions.

[0044] Uncertainty quantification is key to improving system reliability. Its specific implementation can be as follows: For deep learning models, the "Monte Carlo Dropout" method can be used to estimate the confidence level by performing multiple forward propagations during inference and calculating the variance of the output.

[0045] For ensemble learning models, uncertainty can be quantified by calculating the divergence (e.g., standard deviation) of the predictions made by base learners (such as multiple trees in GBDT).

[0046] The specific steps of the decision-making downgrade process are as follows: High confidence: Outputs accurate diagnostic conclusions and triggers "precise treatment warnings" (such as "A-phase contact is loose, tightening is recommended").

[0047] Medium confidence level: downgraded to "abnormal trend alert" (e.g., "Multi-parameter analysis shows an abnormality in phase A, but the diagnostic conclusion is uncertain. It is recommended to strengthen monitoring and perform infrared verification").

[0048] Low confidence / data conflict: Further downgrade to "data quality alarm" and trigger "conservative control strategy" (such as "sensor data is unreliable, manual on-site inspection is recommended, and the operating load of the device is automatically limited to a safe range").

[0049] Preferably, this system also includes a built-in sensor health status assessment module. This module assesses the sensor's own condition to ensure the reliability of the system's input data and continuously monitors the operating status of each sensor. By analyzing the sensor data's inherent noise level, output stability, and its relationship to maintenance records over time, a health score is dynamically calculated for each sensor (e.g., sensors that have been recently maintained have higher scores). This health score serves as a data reliability weight and dynamically participates in the aforementioned feature fusion and model diagnostic processes. For sensors with low health scores, their data weight in the fusion process is automatically reduced, thereby improving the system's overall robustness in the event of individual sensor performance degradation.

[0050] This involves dynamically calculating the health score of each sensor by analyzing its intrinsic noise level, output stability, and relationship to maintenance records. Examples of the scores include: Self-noise level: Evaluated by calculating the signal-to-noise ratio (SNR) or variance of the output signal when there is no signal input or a steady-state signal.

[0051] Output stability: This is determined by analyzing whether the output value exceeds a reasonable physical range or by the magnitude of the residual between the output value and the predicted value based on the device's physical model.

[0052] Maintenance log: Design a function that decays over time, with a lower score for this item the longer the time since the last maintenance.

[0053] Specifically, the dynamic participation includes: during feature fusion, dynamic weights are used as coefficients for weighted averaging; during model diagnosis, dynamic weights are used as part of the input features, or to construct a weighted loss function that emphasizes health sensor data during model training.

[0054] The system operates as follows: First, the sensor array of the multi-parameter intelligent sensing layer acts as the system's "sensory nerves," continuously collecting multi-dimensional raw time-series data from the device. This data is first received by the edge computing and data acquisition layer, where it undergoes preprocessing such as data cleaning, filtering, and unified timestamps to ensure data quality and time-series alignment. The edge side can also perform simple feature calculations (such as 1-minute average temperature rise) to achieve millisecond-level rapid response to emergency anomalies (such as sudden temperature increases).

[0055] Then, the preprocessed data is uploaded to the cloud-based intelligent diagnosis and early warning platform. The multi-parameter time-series feature fusion module then begins its core work: Deep time-series feature sets are extracted from various parameter data. These features are no longer raw readings, but refined state indicators that are strongly correlated with the physical failure mechanism of the equipment (such as the rate of temperature rise characterizing the severity of heating, and the frequency offset rate characterizing changes in mechanical structure).

[0056] By using a feature assimilation mechanism, different types of data (such as vibration waveforms and infrared thermograms) are mapped to a unified mathematical space, solving the problem of heterogeneous data such as "apples and oranges cannot be directly compared".

[0057] All features are concatenated and standardized to construct a multi-parameter fusion feature vector. This vector is a "multi-dimensional portrait" depicting the health status of the device from multiple perspectives.

[0058] Furthermore, the fault diagnosis model of the intelligent diagnosis and early warning module receives the fused feature vector and performs the following operations: The model, based on failure modes learned from historical data, outputs quantitative diagnostic results, including a comprehensive health index (overall score) and the probability of specific failure modes (precise localization). This achieves a leap from "feeling unwell" to "diagnosing the cause of the illness."

[0059] Uncertainty quantification and decision downgrading mechanisms are introduced. The model will "self-awarely" assess the confidence level of the diagnosis. When the confidence level is low (such as due to sensor data conflicts or encountering unknown operating conditions), the system will not force an unreliable "accurate diagnosis," but will automatically downgrade to a more conservative "anomaly alert" and suggest manual review, which greatly improves the system's engineering practicality and safety.

[0060] In addition, the sensor health assessment module runs continuously in the background, acting like a "physician" to monitor the operational status of all sensors and generate a health score for each sensor. This score serves as a dynamic weight, feeding back into the feature fusion and model diagnostic processes in real time. When a sensor's performance degrades, the system automatically reduces its data weight, relying more on other healthy sensors, thereby significantly improving the long-term robustness of the entire system in real industrial environments.

[0061] like Figure 2 As shown, the present invention also provides a method for monitoring and early warning of the condition of electrical complete sets of equipment based on multi-parameter intelligent sensing. The method includes the following steps: S1: A multi-sensor array deployed at key nodes of the equipment synchronously collects raw time-series data of multiple parameters. This array, comprised of various types of sensors deployed at critical locations to address different fault modes of the electrical equipment, forms a distributed sensing network. Synchronous acquisition, using a unified clock reference, ensures that different sensors collect data at the same point in time, laying the foundation for subsequent multi-parameter correlation analysis. The raw time-series data is a sequence of unprocessed data directly output from the sensors, arranged chronologically.

[0062] S2: Preprocess, filter, and align the original time-series data. The timestamp alignment is based on a unified time axis. Time-series data with different sampling rates are resampled or interpolated so that they have corresponding data points at the same time, forming "data slices".

[0063] S3: Extract deep time series features from the preprocessed time series data of each parameter to form a feature set; deep time series features go beyond simple statistics (such as mean and maximum value) and can reveal the inherent patterns, trends and high-order features that are strongly correlated with the evolution of the physical state of the equipment.

[0064] S4: Deep time-series features from different parameters are fused to construct a comprehensive multi-parameter fusion feature vector. Feature fusion combines features from different sources and with different physical meanings to form a unified, high-dimensional data structure to provide a comprehensive description of the device's status. The single vector formed after the fusion of the multi-parameter fusion feature vector is a "digital portrait" of the device's health status within a specific time window.

[0065] S5: Input the multi-parameter fusion feature vector into the pre-trained intelligent diagnostic model, and the model outputs the real-time health index of the device and the probability of each major fault mode; S6: Based on the diagnostic results output by the model, generate and issue graded early warnings with clear fault indications.

[0066] The specific implementation process of step S1 includes: Deployment planning of multi-sensor array: Based on fault mode and effect analysis, wireless temperature sensors are deployed at circuit breaker contacts, ultra-high frequency partial discharge sensors are deployed in the switch cabinet space, vibration acceleration sensors are installed in the circuit breaker operating mechanism, and circuit current is collected through smart meters.

[0067] Establish a synchronization mechanism: Integrate a high-precision clock source into the edge computing gateway, or adopt the IEEE 1588 precision clock protocol. All sensors are assigned a unified microsecond-level timestamp during data acquisition.

[0068] Multi-mode acquisition, such as: For current signals, continuous high-speed sampling (e.g., 6.4kHz) is used to capture harmonics and transient processes; For temperature signals, use low-speed periodic sampling (e.g., 1Hz). For partial discharge signals, event-triggered acquisition is used, and a complete waveform data is recorded only when a pulse exceeding the threshold is detected.

[0069] Step S1 is the data source of the entire method, and the comprehensiveness and synchronization accuracy of the collected data directly determine the accuracy and reliability of all subsequent analyses.

[0070] The specific implementation process of step S2 includes: Edge-side preprocessing: The edge computing gateway performs a moving average filter on the temperature data to smooth random fluctuations; and uses a band-stop filter on the vibration data to eliminate specific mechanical resonance interference.

[0071] Time alignment: After receiving the data, the cloud platform performs linear interpolation on the 1Hz temperature data and the 6.4kHz current data at 1-millisecond intervals to generate a data sequence with perfectly aligned timestamps, which is used for subsequent calculation of the "current-temperature" joint feature.

[0072] Step S2 is the "refinery" of the data. It transforms the raw, heterogeneous, and noisy data collected in step S1 into clean, well-organized, and spatiotemporally aligned high-quality data, providing qualified input for feature extraction in step S3.

[0073] In step S3, the extraction of deep temporal features is tailored to the physical characteristics of different parameters. For example, for temperature parameters, the focus is on capturing their dynamic trends rather than instantaneous absolute values; for vibration parameters, frequency domain analysis is used to uncover their periodic impact or resonance characteristics related to the mechanical state. Specifically, this includes: Temperature parameters: Focus on capturing dynamic trends. For example, extract the linear regression slope over the past 5 minutes as the temperature rise rate, and calculate the temperature difference between the contact points of phase A and phase B at the same moment.

[0074] Vibration parameters: Extracting mechanical state information through frequency domain analysis. For example, performing a Fast Fourier Transform (FFT) on the vibration signal to extract the total energy within the 100Hz-1kHz frequency band, and calculating the percentage offset between the current dominant frequency (e.g., 100Hz) and the historical reference dominant frequency.

[0075] Partial discharge parameters: Extract the number of discharge pulses per minute (pulse repetition rate) and generate a phase-resolved partial discharge spectrum (PRPD), and then extract its statistical characteristics, such as the skewness of the discharge quantity distribution.

[0076] Electrical parameters: Calculate the unbalance of the three-phase current, the total harmonic distortion (THD), and the rolling standard deviation of the load rate.

[0077] Step S3 elevates the data from "data" to "information." It utilizes the clean data provided in step S2 to extract measurable quantitative indicators directly related to the device's health status. These characteristics serve as the "language" and "evidence" for subsequent intelligent diagnostics.

[0078] The specific implementation process of step S4 includes, for example: Feature assimilation: For heterogeneous features that are difficult to fuse directly, such as vibration time series signals and infrared thermograms, a recurrence plot can be used to convert them into features in image format, thereby achieving fusion under the same modality.

[0079] Concatenation and Standardization: All features extracted in step S3 (such as temperature rise rate, dominant frequency offset rate, pulse repetition rate, harmonic distortion rate, etc.) are concatenated into a long vector. Then, this vector is Z-score standardized to bring all features to the same order of magnitude. For example, the final fused feature vector might look like this: [Temperature rise rate: 0.85, Simultaneous temperature difference: 1.2, Vibration band energy: -0.3, Dominant frequency offset rate: 2.1, Pulse repetition rate: 0.75, Three-phase imbalance: -1.1, ...].

[0080] Step S4 is the integration of "information" into "knowledge." It consolidates the scattered features reflecting the device status from different perspectives in step S3 into a comprehensive evaluation criterion. This fused feature vector eliminates the one-sidedness of a single parameter, providing a comprehensive and unified input for the intelligent diagnosis in step S5.

[0081] In step S5, the intelligent diagnostic model may employ ensemble learning algorithms such as Gradient Boosting Decision Tree (GBDT) or Random Forest, or Long Short-Term Memory (LSTM) networks suitable for processing time series. The model training process focuses on learning the collaborative change patterns among multi-parameter features under different fault modes.

[0082] The specific implementation process of step S5 includes, for example: Model inference: Input the fused feature vector generated in step S4 into the model.

[0083] Output result: Health index: For example, output 85.

[0084] Fault mode probabilities: For example, output {'Increased contact resistance': 0.72, 'Moisture absorption': 0.05, 'Mechanical loosening': 0.20, ...}.

[0085] Uncertainty quantification: The model simultaneously assesses the confidence level of the diagnosis. For example, using the Monte Carlo Dropout technique, if the variance of multiple inference results is large, the confidence level is low.

[0086] Step S5 is the "brain" and "decision-making core" of the method in this embodiment. It receives the comprehensive "knowledge" (fusion feature vector) provided in step S4, uses the "experience" learned from history (trained model), and finally outputs a quantitative and interpretable diagnostic conclusion, providing direct evidence for the early warning in step S6.

[0087] The specific implementation process of step S6 includes, for example: Generate warning content: Combine with the output of S5. For example, based on the high probability of increased contact resistance, generate the warning: "Alarm (Level II): There is a risk of excessive contact resistance at the A-phase circuit breaker contact of XX switch cabinet. It is recommended to check and tighten it." Decision downgrade: If the uncertainty quantification in S5 shows a low confidence level, an automatic downgrade warning will be issued. For example, the warning may be issued as follows: "Attention (Level I): The multi-parameter data of phase A of the XX switchgear is abnormal, but the diagnostic conclusion has high uncertainty. It is recommended to conduct infrared temperature measurement verification." Release: Alerts are pushed to operations and maintenance personnel through multiple channels such as web interface, mobile APP, and SMS.

[0088] Step S6 is the "value output" and "action closed loop" of the method in this embodiment. It transforms the intelligent diagnostic conclusion of step S5 into executable operation and maintenance instructions, and finally completes the complete closed loop from "data perception" to "operation and maintenance decision-making", achieving the ultimate goal of status monitoring.

[0089] The six interconnected steps of this invention form a complete, intelligent, and adaptive technological chain from the physical world to the digital world, and then to decision-making and action. Each step builds upon the output of the previous step, progressively enhancing the value of information and the intelligence of decision-making, ultimately achieving accurate perception, early warning, and intelligent operation and maintenance of electrical equipment.

[0090] Example 2 Based on Example 2, taking a low-voltage distribution cabinet at a project site as the monitoring object, the technical solution of the present invention is specifically described, including: In the multi-parameter intelligent sensing layer, we install wireless temperature sensors at the busbar connection points inside the cabinet, vibration sensors near the circuit breaker operating mechanism, ultra-high frequency (UHF) partial discharge sensors inside the cabinet, and collect circuit current and voltage signals through intelligent power meters.

[0091] The edge computing gateway collects all sensor data at high speed, performs preliminary filtering, and assigns a uniform timestamp to all data. It also calculates some basic real-time characteristics, such as the 1-minute moving average of the temperature sensor.

[0092] The data is encrypted and transmitted to the cloud-based intelligent diagnostic and early warning platform. The platform's multi-parameter time-series feature fusion module uses a 10-minute sliding window as the analysis unit to extract the following deep time-series features from the raw data: Extract the following from the temperature data: real-time temperature, maximum temperature rise rate in the last 10 minutes, and real-time temperature difference between this measuring point and the adjacent B-phase measuring point.

[0093] Extract the following from the vibration data: effective vibration value in the 50Hz-500Hz frequency band (related to electromagnetic force and mechanical resonance), and percentage deviation of the current dominant vibration frequency from the historical baseline dominant frequency.

[0094] Extract from partial discharge data: the ratio of the number of discharge pulses in the positive half-cycle to the number of discharge pulses in the negative half-cycle within the power frequency cycle.

[0095] Extract the following from the electrical parameter data: the effective value of the current in the circuit and the third harmonic content of the current.

[0096] All these features, after being normalized, are combined into a multidimensional fusion feature vector, which is then input into the trained Gradient Boosting Tree (GBDT) fault diagnosis model.

[0097] After performing a comprehensive analysis of the feature vector, the model outputs the following diagnostic results: Overall equipment health index: 75.

[0098] Fault mode probability analysis: "Increased contact resistance at busbar connection point": 70%; "Mechanical jamming of circuit breaker mechanism": 8%; "Contamination of internal insulation surface": 15%; Others: 7%.

[0099] Since the probability of "increased contact resistance at busbar connection point" is significantly higher than other types and exceeds the set warning threshold (60%), the system immediately generates a Level II warning, which clearly states: "Warning: There is a risk of increased contact resistance at the connection point of phase A busbar of XX distribution cabinet. The characteristic is an abnormal temperature rise rate accompanied by specific electromagnetic vibration. It is recommended to check and tighten the connection bolts as soon as possible." Based on the accurate early warning information, the maintenance personnel focused on checking the connection point during subsequent planned power outages and found that the bolts were slightly loose. They addressed the issue promptly, successfully preventing a potential power outage caused by overheating of the connection point.

[0100] This embodiment demonstrates that the present invention, through the effective combination of multi-parameter time-series feature fusion and intelligent diagnostic model, achieves early and accurate diagnosis of potential faults, and upgrades the operation and maintenance mode from "passive maintenance" to "proactive early warning and accurate maintenance", fully reflecting the technological progress and practical value of the present invention.

[0101] Example 3 Building upon Example 2, an additional example is added to demonstrate how the system maintains robustness under sensor performance degradation scenarios, comprehensively showcasing the improved system capabilities. This example uses a high-voltage switchgear in a substation as the monitoring object to verify the system's robustness under sensor performance degradation. After long-term operation, the wireless temperature sensor at the A-phase bus connection of the high-voltage switchgear in the substation experienced intermittent signal transmission due to battery aging, resulting in an intermittent decrease in the Data Acquisition Compliance Index (DA). Specifically, this includes: Data Sensing and Health Assessment: The system detected an increase in the data packet loss rate of the temperature sensor and, through a self-test algorithm, found that its signal strength was below the threshold. The sensor health status assessment module lowered its health score from 0.95 to 0.6.

[0102] Feature fusion and weighting: When constructing the multi-parameter fusion feature vector, the system automatically reduces the weight of the original data of the temperature sensor and its derived features (such as real-time temperature values). At the same time, the multi-parameter time-series feature fusion module enhances the features from the vibration sensor (monitoring loose connections) and the infrared thermal imager (non-contact temperature measurement), and uses electrical parameter features (such as the growth of harmonics in phase A current) for cross-validation.

[0103] Intelligent Diagnosis and Decision Making: Despite incomplete temperature data, the fused feature vectors can still be effectively input into the fault diagnosis model. The model, considering both increased vibration and current harmonic characteristics, determines the probability of "increased contact resistance at the A-phase bus connection point" to be 65% (medium confidence level). Due to the unreliability of a single temperature channel, the overall confidence level of the model output is marked as "medium".

[0104] Early warning and degradation strategy: The system did not trigger the highest-level emergency warning, but instead generated a medium-level warning, which read: "Note: There is a risk of poor contact at the A-phase busbar connection of the XX switchgear, but some sensor data has low reliability. It is recommended to prioritize infrared temperature measurement verification in conjunction with the recent inspection plan." At the same time, the system automatically increased the sampling frequency of the infrared thermal imager in the cabinet to compensate for the lack of temperature data.

[0105] This embodiment demonstrates that the present invention, through dynamic weight adjustment and multi-source information complementarity, can still maintain effective state assessment and risk warning even when the performance of some sensors degrades, reflecting the system's strong engineering practicality and robustness.

[0106] Example 4 The present invention also provides an electronic device, including: a processor, a transmitting device, an input device, an output device, and a memory. The processor may be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit, or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory may be implemented using a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), and is used to store computer program code. The computer program code includes computer instructions. When the processor executes the computer instructions, the electronic device executes a method as described in any of the above possible implementation methods.

[0107] Example 5 The present invention also provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor of an electronic device, cause the processor to perform a method as described in any of the above possible implementations.

[0108] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A multi-parameter intelligent sensing-based electrical complete equipment state monitoring and early warning method, characterized in that, The method comprises the following steps: S1: synchronously collecting original time series data of electrical parameters, temperature parameters, partial discharge parameters, mechanical vibration parameters and environmental parameters in a multi-mode adaptive acquisition manner through a multi-sensor array deployed at a key node of an electrical complete set of equipment; S2: pre-processing, filtering and time stamp alignment based on a hardware clock synchronization protocol on the original time series data at the edge computing gateway side; S3: extracting a deep time series feature set strongly related to a fault mechanism from the pre-processed time series data of each parameter; S4: fusing the deep time series features from different parameters to construct a unified multi-parameter fusion feature vector; S5: inputting the multi-parameter fusion feature vector into a pre-trained intelligent diagnosis model to output a real-time health index of the equipment and probabilities of each main fault mode, and evaluating the confidence level of the diagnosis result; S6: based on the diagnosis result and its confidence level, performing a hierarchical early warning and decision degradation process to generate and issue early warning information with clear fault direction.

2. The method of claim 1, wherein, In the step S3, the deep time series feature set comprises: extracting a temperature rise trend slope based on linear regression and a dynamic time warping distance from historical data from temperature parameters; extracting energy, main frequency component and its shift rate of a specific fault sensitive frequency band, and envelope spectrum analysis features from vibration parameters; extracting pulse repetition rate and statistical moment features of discharge phase distribution atlas from partial discharge parameters; extracting three-phase unbalance degree, current harmonic distortion rate and load rate fluctuation features from electrical parameters.

3. The method of claim 1, wherein, The step S4 comprises: using a recursive graph, Gram angle field or Markov transition field method to feature assimilate heterogeneous time series data and convert them to a unified feature expression space; splicing the assimilated deep time series features and processing them by Z-score standardization or Min-Max normalization method to construct the multi-parameter fusion feature vector.

4. The method of claim 1, wherein, In the step S5, the intelligent diagnosis model uses gradient boosting decision tree, random forest, long short-term memory network, support vector machine or one-dimensional convolutional neural network algorithm; and the confidence level of the diagnosis result is evaluated by using a Monte Carlo Dropout method or by calculating the divergence of the prediction results of an ensemble learning model.

5. The method of claim 1, wherein, The decision degradation process in the step S6 comprises: when the confidence level is high, triggering a precise diagnosis early warning; when the confidence level is medium, degrading to an abnormal trend reminder and suggesting manual review; when the confidence level is low or a serious conflict in multi-parameter information is perceived, further degrading to a data quality alarm and triggering a conservative control strategy.

6. The method of claim 1, wherein, Further comprising a sensor health state evaluation step: dynamically calculating a health score of each sensor by analyzing the noise level, output stability and time relationship with maintenance records of the sensor data; using the health score as a data reliability weight to dynamically participate in the feature fusion in the step S4 and the model diagnosis process in the step S5.

7. A multi-parameter intelligent sensing based electrical switchgear condition monitoring and early warning system, characterized in that, The system applies the method of any one of claims 1 to 6, and the system comprises: A multi-parameter intelligent perception layer, including a sensor array based on fault mode and effect analysis and deployed at key nodes of the electrical complete equipment, for collecting raw time series data of electrical parameters, temperature parameters, partial discharge parameters, mechanical vibration parameters and environmental parameters; An edge computing and data acquisition layer, including an edge computing gateway, for receiving and preprocessing the raw time series data, time synchronization based on IEEE 1588 PTP or GPS timing, and simple feature calculation on the edge side; A cloud intelligent diagnosis and early warning platform, for storing full-amount historical data, and running a multi-parameter time series feature fusion module and an intelligent diagnosis and early warning module.

8. The system of claim 7, wherein, The intelligent diagnosis and early warning module includes: A fault diagnosis model trained based on a machine learning algorithm, for receiving the multi-parameter fusion feature vector and outputting a device comprehensive health index and a specific fault mode probability; An uncertainty quantification unit, for evaluating the confidence level of the diagnosis result; A decision degradation unit, for triggering a hierarchical early warning process according to the confidence level.

9. An electronic device comprising: A processor, a memory, a sending device, an input device and an output device, the memory storing a computer program, characterized in that when the computer program is executed by the processor, the method of any one of claims 1 to 6 is implemented.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor, and the method of any one of claims 1 to 6 is implemented.

Images