Power equipment state monitoring method and system based on Internet of Things

Abstract

The invention relates to the field of equipment monitoring, in particular to a power equipment state monitoring method and system based on the Internet of Things. The method comprises the following steps: acquiring operation data of power equipment, and generating an original data set containing temperature, vibration and electrical parameters; performing three-stage collaborative verification processing on the data to obtain a characteristic spectrum marked with an abnormal frequency band; generating a data acquisition strategy matrix including sensor priority adjustment and edge-cloud resource allocation according to a dynamic mapping relation between the characteristic spectrum and the equipment health degree index; performing multi-modal data fusion based on the strategy matrix to generate an equipment state feature vector; and updating the health degree evaluation model according to the feature vector, and feeding back and adjusting a sampling strategy. The problems that in the prior art, data processing is not standard, the intelligent degree is low, and the equipment health state cannot be accurately evaluated are effectively solved, and the intelligent level and reliability of power equipment monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of equipment monitoring, and more specifically to a method and system for monitoring the condition of power equipment based on the Internet of Things. Background Technology

[0002] With the rapid development of smart grids, the application of IoT technology in power equipment condition monitoring is becoming increasingly widespread. Traditional power equipment condition monitoring mainly relies on manual inspections and simple sensor monitoring, which is inefficient and makes it difficult to achieve real-time and comprehensive monitoring of equipment status. In recent years, although IoT technology has been gradually applied to the field of power equipment monitoring, existing technologies still have many shortcomings. For example, the deployment of monitoring terminals is not standardized, and data transmission protocols and acquisition channels are not unified, resulting in poor data integration and incomplete functions at various levels of master station platforms. In addition, existing systems have a low level of intelligence in data processing and analysis, failing to fully utilize big data and advanced algorithms to deeply mine massive amounts of monitoring data, and are unable to effectively assess the health status of equipment and provide accurate maintenance decisions. For example, patent CN201911308178.1 proposes an IoT-based intelligent power equipment full life cycle comprehensive monitoring and early warning system. Although it achieves data acquisition and transmission, there is still room for improvement in data integration and intelligent analysis. Therefore, there is an urgent need for an IoT-based power equipment condition monitoring method and system. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for monitoring the status of power equipment based on the Internet of Things, so as to solve the problems of non-standard data processing, low level of intelligence, and inability to accurately assess the health status of power equipment in the prior art.

[0004] To achieve the above objectives, the following technical solution is adopted.

[0005] IoT-based power equipment condition monitoring methods include,

[0006] S1. Collect power equipment operation data and generate a raw dataset containing temperature, vibration, and electrical parameters;

[0007] S2. Perform a three-level collaborative verification process on the original dataset to obtain a feature spectrum containing abnormal frequency band markers;

[0008] S3. Based on the dynamic mapping relationship between the feature spectrum and the device health index, a data acquisition strategy matrix is ​​generated. The strategy matrix includes: S31. Sensor priority adjustment instructions; S32. Edge-cloud resource allocation ratio;

[0009] S4. Perform multimodal data fusion based on the aforementioned strategy matrix.

[0010] S41. Align the temperature data with the vibration spectrum in time and frequency to generate a synchronized dataset;

[0011] S42. Perform weighted fusion on the synchronization dataset according to the weight coefficients in the policy matrix to generate a device state feature vector;

[0012] S5. Update the device health assessment model based on the feature vector, and feed back the health change trend output by the model to S1 to dynamically adjust the multi-source sensor sampling strategy.

[0013] Optionally, step S2 specifically includes,

[0014] S21. Perform value range verification based on the equipment's rated parameter range and output the first compliant dataset;

[0015] S22. Perform temporal abrupt change detection on the first compliant dataset. When the data change rate exceeds the threshold, activate adaptive filtering compensation to generate a second corrected dataset.

[0016] The adaptive filtering compensation in S22 includes,

[0017] S221. Select either Kalman filtering or wavelet thresholding denoising algorithm based on the temporal abrupt change detection results;

[0018] S222. Perform residual analysis on the compensated data, and trigger the sensor calibration process when the residual variance exceeds the threshold.

[0019] The output of S222 will be used as the input parameter for the multi-scale frequency domain analysis in S23;

[0020] S23. Perform multi-scale frequency domain analysis on the second corrected dataset and output a feature spectrum containing anomalous frequency band markers.

[0021] Optionally, the strategy matrix generation method of S3 includes,

[0022] S301. Analyze the energy proportion of anomalous frequency bands in the characteristic spectrum;

[0023] S302. Calculate the equipment health degradation rate based on the energy ratio;

[0024] S303. Based on the dynamic relationship between attenuation rate and network quality index, determine data acquisition density and communication protocol priority; wherein the output parameters of S303 directly control the weighted fusion coefficient in S4.

[0025] Optionally, the model update method of S5 includes,

[0026] S51. Extract a dynamic feature subset from the feature vector that matches the current operating condition;

[0027] S52. Compare the dynamic feature subset with the historical health status benchmark database;

[0028] S53. When the similarity is lower than the set threshold, the incremental training process is triggered to optimize the model decision boundary;

[0029] The updated model parameters generated in S53 will be used to correct the health decay rate calculation algorithm in S3.

[0030] Optional, also includes,

[0031] S6. The abnormal propagation suppression process includes the following sub-steps:

[0032] S61. Monitor the intermediate data quality indicators for each stage from S2 to S5;

[0033] S62. When a data verification failure is detected, freeze the current processing pipeline and activate the redundant channel;

[0034] S63. Add an isolation flag to the abnormal data stream and reinitialize the input buffer of the affected module.

[0035] Optionally, the re-initialization method of S63 includes,

[0036] S631. Backtrack the feature vector copy of the most recent valid state;

[0037] S632. Reconstruct the intermediate state of the processing pipeline based on the backtracking data;

[0038] S633. Resume operation after verifying the consistency between the reconstructed state and the current sensor readings.

[0039] Optionally, the multimodal data fusion in S4 further includes,

[0040] S43. Establish the physical constraint equations relating temperature, vibration, and electrical parameters;

[0041] S44. Perform constraint satisfaction verification on the fusion results. When physical association constraints are violated...

[0042] S441. Adjust the weight coefficients in the strategy matrix;

[0043] S442. Re-execute time-frequency alignment and weighted fusion;

[0044] S443. Record constraint conflict events to the model update training set.

[0045] Optionally, the feedback adjustment method includes,

[0046] S501. Analyze the feature contribution distribution of the health assessment model output;

[0047] S502. Generate sensor sensitivity adjustment instructions based on contribution distribution;

[0048] S503. Dynamically reconstructs the topology of the sensor array to improve the sampling density of key data channels.

[0049] An IoT-based power equipment condition monitoring system includes,

[0050] A multi-source sensor array is used to collect operating data of power equipment and generate a raw dataset containing temperature, vibration, and electrical parameters.

[0051] The value range verification module is used to receive the original dataset and output the first compliant dataset;

[0052] The time-domain processing module is used to perform mutation detection and adaptive filtering on the first compliant dataset to generate the second corrected dataset;

[0053] The frequency domain analysis module is used to perform multi-scale decomposition on the second corrected dataset and output the feature spectrum;

[0054] The dynamic decision center is used to analyze the energy distribution of abnormal frequency bands in the feature spectrum and generate a strategy matrix that includes sensor priority, communication protocol, and computing resource allocation.

[0055] A multimodal fusion unit is used to receive and execute policy matrix instructions.

[0056] Time-frequency alignment unit is used to synchronize the timestamps of data from multiple source sensors;

[0057] The weighted fusion unit is used to generate device state feature vectors based on policy weights;

[0058] A self-learning model library is used to store and update health assessment models and convert model outputs into feedback control signals.

[0059] The closed-loop control interface converts feedback control signals into real-time adjustment commands for a multi-source sensor array.

[0060] Optionally, it also includes an anomaly isolation subsystem, which contains:

[0061] Data quality probes monitor the intermediate processing results of each module;

[0062] A redundant channel switch activates a backup processing path when an abnormal data flow is detected.

[0063] The state reconstruction unit restores the processing pipeline based on historical valid data: the anomaly isolation subsystem establishes a bidirectional connection with the self-learning model library and imports the features of abnormal events into the model training set.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] This application presents an IoT-based power equipment condition monitoring method and system. By collecting power equipment operating data and performing a three-level collaborative verification process, a feature spectrum containing abnormal frequency band markers is generated. This is followed by dynamically generating a data acquisition strategy matrix and performing multimodal data fusion. Finally, the health assessment model is updated based on the equipment status feature vector. This effectively solves the problems of non-standard data processing, low intelligence, and inaccurate equipment health status assessment in existing technologies. First, the three-level collaborative verification process accurately identifies abnormal frequency bands in the data, improving data accuracy and reliability. Second, the dynamic generation of the data acquisition strategy matrix optimizes sensor priority adjustment and resource allocation, enhancing the flexibility and efficiency of the monitoring system. Furthermore, multimodal data fusion technology further enhances data correlation and usability, making equipment status assessment more comprehensive and accurate. Finally, the closed-loop feedback mechanism dynamically adjusts the sampling strategy, further improving the intelligence level of the monitoring system, effectively reducing operation and maintenance costs and improving equipment reliability.

[0066] This application optimizes the performance of the monitoring system. The refinement of the three-level collaborative verification process ensures the accuracy and consistency of data across different dimensions, effectively avoiding potential omissions by single verification methods. By analyzing the energy proportion of abnormal frequency bands in the feature spectrum and dynamically adjusting the data acquisition density and communication protocol priority, the system's adaptability is further enhanced. The introduced model update mechanism, through comparison of dynamic feature subsets with a historical health status benchmark database, achieves continuous model optimization, improving the accuracy of health assessment. This application further improves the system's anomaly handling and data fusion mechanisms, ensuring the stability and reliability of the monitoring system. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of an embodiment of the power equipment condition monitoring method based on the Internet of Things according to the present invention.

[0068] Figure 2 This is a schematic diagram of a module of an embodiment of an Internet of Things-based power equipment condition monitoring system according to the present invention. Detailed Implementation

[0069] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0070] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0071] Example 1

[0072] like Figure 1 As shown, in the implementation of the IoT-based power equipment condition monitoring method, it is first necessary to build a multi-source data acquisition network.

[0073] Collect power equipment operation data (step S1)

[0074] In this embodiment, a multi-source sensor array needs to be deployed on the power equipment first. These sensors include, but are not limited to, temperature sensors, vibration sensors, current transformers, and voltage transformers. The sensor layout should be optimized according to the specific type of power equipment and its operating environment. For example, in a high-voltage transformer, temperature sensors should be installed on the transformer windings and tank surface, vibration sensors should be installed on the transformer casing, and current and voltage transformers should be installed at the transformer's input and output terminals.

[0075] The data collected by the sensors includes parameters such as temperature, vibration acceleration, current, and voltage. This data is transmitted to a data processing center via IoT communication modules (such as LoRa, NB-IoT, or 5G). To ensure data integrity and accuracy, the sensor's sampling frequency should be adjusted according to the device's operating characteristics. For example, for high-frequency vibration signals, the sampling frequency must meet the requirements of the Nyquist sampling theorem and is typically set to at least twice the highest frequency of the signal.

[0076] Three-level collaborative verification process (step S2)

[0077] The collected raw dataset needs to undergo three levels of collaborative validation to ensure the accuracy and reliability of the data. The specific steps are as follows:

[0078] Value range verification: The raw data is verified based on the rated parameter range of the equipment to filter out data that conforms to the normal operating range of the equipment, forming the first compliant dataset. For example, the rated voltage of a transformer is 220kV and the rated current is 1000A. Any data points that exceed these ranges will be marked as abnormal and removed.

[0079] Temporal abrupt change detection and adaptive filtering compensation: Temporal abrupt change detection is performed on the first compliant dataset. When the rate of change of the data exceeds a preset threshold, adaptive filtering compensation is activated. For example, when the rate of change of vibration data exceeds 10%, the system will initiate filtering compensation. A suitable filtering algorithm, such as Kalman filtering or wavelet thresholding denoising, is selected based on the temporal abrupt change detection results. Kalman filtering is suitable for handling noise problems in linear systems, while wavelet thresholding denoising is more suitable for handling nonlinear and non-stationary signals.

[0080] Multi-scale frequency domain analysis: This involves performing multi-scale frequency domain analysis on the filtered and compensated data. Using multi-scale analysis methods such as wavelet transform, the data is decomposed into components of different frequency bands, and anomalous frequency bands are identified. For example, in the vibration signal of a transformer, abnormal energy distribution in certain frequency bands may indicate potential equipment failure. By labeling these anomalous frequency bands, a feature spectrum containing these anomalous frequency band labels is formed.

[0081] Generate the data acquisition strategy matrix (step S3)

[0082] A data acquisition strategy matrix is ​​generated based on the dynamic mapping relationship between the feature spectrum and the device health index. The specific steps are as follows:

[0083] Analyzing the energy proportion of abnormal frequency bands in the characteristic spectrum: This involves analyzing the energy distribution of abnormal frequency bands in the characteristic spectrum to calculate the rate of decline in equipment health. For example, if the energy proportion of a certain abnormal frequency band continues to increase, it indicates that the equipment's health status may be deteriorating.

[0084] Determine the data acquisition density and communication protocol priority based on the dynamic relationship between the health decay rate and the network quality index. The network quality index can comprehensively consider factors such as communication bandwidth, latency, and packet loss rate. When the health decay rate is high and network conditions permit, increase the data acquisition density and prioritize the use of high-speed, stable communication protocols to transmit critical data.

[0085] Generate a strategy matrix: The strategy matrix contains sensor priority adjustment instructions and edge-cloud resource allocation ratios. By dynamically adjusting the sensor sampling frequency and data transmission path, the system's resource utilization efficiency is optimized. For example, when the proportion of energy in the abnormal temperature frequency band of a certain photovoltaic panel increases, the system can increase the sampling frequency of the photovoltaic panel's temperature sensor and prioritize transmitting relevant data to the cloud for analysis, while simultaneously adjusting the resource allocation of edge computing nodes.

[0086] Multimodal data fusion (step S4)

[0087] Based on the generated data acquisition strategy matrix, multimodal data fusion is performed. The specific steps are as follows:

[0088] Time-frequency alignment: Temperature data and vibration spectra are aligned in time and frequency to generate a synchronized dataset. This process uses techniques such as timestamp calibration to ensure the consistency of different modal data in the time dimension. For example, for temperature and vibration data acquired simultaneously, interpolation or time delay compensation methods are used to align the sampling points of the two in time.

[0089] Weighted fusion: The synchronized dataset is weighted and fused according to the weight coefficients in the policy matrix to generate a device state feature vector. The weight coefficients can be adjusted based on the degree of influence of different modal data on the device health status. For example, in some cases, vibration signals may be more indicative of device fault diagnosis, so vibration data can be given higher weight.

[0090] Physical correlation constraint verification: Physical correlation constraint equations are introduced to establish the physical correlation between temperature, vibration, and electrical parameters. During data fusion, constraint satisfaction verification is performed on the fusion results. If the fusion results are found to violate the physical correlation constraints, such as the relationship between temperature and current not conforming to the thermal characteristic model of the equipment, the weight coefficients in the strategy matrix are adjusted, time-frequency alignment and weighted fusion are re-executed, and constraint conflict events are recorded in the model update training set to provide data support for subsequent model optimization.

[0091] Updates and feedback on the equipment health assessment model (Step S5)

[0092] The device health assessment model is updated based on the generated device status feature vector. The specific steps are as follows:

[0093] Extracting dynamic feature subsets: Extracting dynamic feature subsets that match the current operating conditions from the feature vector. For example, selecting appropriate feature subsets for analysis based on the equipment's operating status under different seasons or load conditions.

[0094] Similarity Comparison and Incremental Training: A dynamic subset of features is compared with a historical health status benchmark database. This database stores feature vector samples of the device under different health states. The device's health status is assessed by calculating the similarity between the current feature subset and historical samples. When the similarity falls below a set threshold, an incremental training process is triggered to optimize the model's decision boundary. Incremental training introduces new feature data to update the model's parameters, enabling it to better adapt to changes in the device's operating state.

[0095] Feedback to adjust sampling strategy: The updated model parameters will be used to correct the health decay rate calculation algorithm, further improving the model's accuracy and reliability. Finally, the health change trend output by the model is fed back to the data acquisition stage, dynamically adjusting the sampling strategy of the multi-source sensors. For example, if the model predicts that the equipment's health status may deteriorate, the sampling frequency of relevant parameters will be increased to provide early warning and take maintenance measures.

[0096] Example 2

[0097] like Figure 2 As shown, the hardware architecture of the power equipment condition monitoring system consists of two parts: edge computing nodes and a cloud platform. The edge nodes use an NVIDIA Jetson AGX Xavier as the main controller, equipped with a 16-channel synchronous data acquisition card (16-bit resolution, 1MS / s sampling rate), and a built-in Xilinx Kintex-7 FPGA for accelerating wavelet transform of vibration signals. The multi-source sensor array includes: 12 PT100 temperature sensors (range -50℃ to 200℃), 8 IEPE vibration sensors (frequency range 0.5Hz-10kHz), and 4 sets of ultra-high frequency partial discharge detection units (bandwidth 300MHz-1.5GHz). Sensor data is transmitted via an isolated RS-485 bus, using Manchester encoding to ensure anti-interference capability, and the cable shield is grounded at both ends to suppress common-mode noise.

[0098] The value range verification module implements real-time filtering at the FPGA hardware layer, configuring a 32-bit floating-point arithmetic unit to perform threshold comparisons. The time-domain processing module deploys an improved Kalman filter algorithm, with state equation parameters dynamically adjusted according to the device's operating mode: during steady-state operation, the process noise covariance Q is set to 0.01, and the measurement noise covariance R = 0.1; during transient processes, Q is automatically increased to 0.1 to enhance tracking capabilities. The frequency domain analysis module adopts a parallel processing architecture, with eight ARM Cortex-A72 cores processing wavelet packet decomposition in different frequency bands, exchanging intermediate results via shared memory.

[0099] The dynamic decision center runs a fuzzy inference engine. Input variables include the percentage of abnormal frequency band energy (0-100%), network quality index (0-1), and equipment lifespan (0-30 years). Output variables cover data acquisition density (1-10Hz), edge computing load rate (20%-100%), and communication protocol priority. The decision rule base contains 127 expert experiences. For example, when the network quality index is <0.6 and the health decay rate is >5% / h, the local simplified model is forcibly enabled and the sampling rate is reduced to 2Hz.

[0100] The multimodal fusion unit features a dedicated data bus, and the time-frequency alignment unit uses hardware timestamps (accuracy ±1μs). Weighted fusion coefficients are stored in dual-port RAM for fast DMA access. The physical constraint verification module integrates a simplified ANSYS solver, enabling transient simulations of temperature-vibration coupled fields within 50ms. The self-learning model library uses TensorRT to accelerate inference; after quantizing the ResNet-34 model to INT8 accuracy, the inference time is reduced to 8ms, meeting real-time requirements.

[0101] The anomaly isolation subsystem employs a dual-ring redundant architecture. The main processing ring utilizes a PowerPC architecture to ensure high reliability, while the backup ring is based on RISC-V for low-cost fault tolerance. Data quality probes monitor 40 key metrics, including CRC error count, buffer overflow flag, and clock offset. When a fatal error is detected, the switch completes channel switching within 5ms, and the state reconstruction unit loads the most recently valid configuration from the EEPROM, ensuring memory consistency through differential verification. The system maintenance interface supports remote firmware updates, using AES-256 encrypted transmission of upgrade packages, maintaining 50% of basic monitoring functions during the update process.

[0102] A distributed time-series database is deployed on the cloud platform, using TimescaleDB to extend PostgreSQL for processing high-frequency data streams. A 100-node Spark cluster is configured for large-scale feature association analysis. The knowledge graph engine uses Neo4j to build a device association network, with nodes containing 56 attributes such as device ID and health index. Edge relationships define fault propagation paths and impact weights. The visualization platform uses WebGL technology to dynamically render 3D device models, with health status displayed intuitively through color gradients (green < 80 < yellow < 90 < red), supporting multi-dimensional data drill-down analysis.

[0103] Standardized operating procedures are implemented during on-site deployment: Vibration probes are rigidly fixed with epoxy resin during sensor installation to ensure a resonant frequency higher than 10kHz; thermally conductive silicone grease is applied to the contact surface of the temperature sensor, with a contact thermal resistance <0.05℃·m² / W. During system integration testing, channel consistency tests are performed, requiring amplitude deviation of each sensor to be <±3% and latency difference to be <100μs. Long-term operation and maintenance include daily automatic zero-point calibration (achieved by switching the reference resistor via a relay), monthly manual inspection of mechanical connection torque (controlled within 5-7 N·m), and quarterly updates to the digital twin model parameters.

[0104] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A method for monitoring the condition of power equipment based on the Internet of Things, characterized in that, include, S1. Collect power equipment operation data and generate a raw dataset containing temperature, vibration, and electrical parameters; S2. Perform a three-level collaborative verification process on the original dataset to obtain a feature spectrum containing abnormal frequency band markers; S3. Based on the dynamic mapping relationship between the feature spectrum and the device health index, a data acquisition strategy matrix is ​​generated. The strategy matrix includes: S31. Sensor priority adjustment instructions; S32. Edge-cloud resource allocation ratio; S4. Perform multimodal data fusion based on the aforementioned strategy matrix. S41. Align the temperature data with the vibration spectrum in time and frequency to generate a synchronized dataset; S42. Perform weighted fusion on the synchronization dataset according to the weight coefficients in the policy matrix to generate a device state feature vector; S5. Update the device health assessment model based on the feature vector, and feed back the health change trend output by the model to S1 to dynamically adjust the multi-source sensor sampling strategy.

2. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, Step S2 specifically includes, S21. Perform value range verification based on the equipment's rated parameter range and output the first compliant dataset; S22. Perform temporal abrupt change detection on the first compliant dataset. When the data change rate exceeds the threshold, activate adaptive filtering compensation to generate a second corrected dataset. The adaptive filtering compensation in S22 includes, S221. Select either Kalman filtering or wavelet thresholding denoising algorithm based on the temporal abrupt change detection results; S222. Perform residual analysis on the compensated data, and trigger the sensor calibration process when the residual variance exceeds the threshold. The output of S222 will be used as the input parameter for the multi-scale frequency domain analysis in S23; S23. Perform multi-scale frequency domain analysis on the second corrected dataset and output a feature spectrum containing anomalous frequency band markers.

3. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, The strategy matrix generation method of S3 includes, S301. Analyze the energy proportion of anomalous frequency bands in the characteristic spectrum; S302. Calculate the equipment health degradation rate based on the energy ratio; S303. Based on the dynamic relationship between attenuation rate and network quality index, determine data acquisition density and communication protocol priority; wherein the output parameters of S303 directly control the weighted fusion coefficient in S4.

4. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, The model update method of S5 includes, S51. Extract a dynamic feature subset from the feature vector that matches the current operating condition; S52. Compare the dynamic feature subset with the historical health status benchmark database; S53. When the similarity is lower than the set threshold, the incremental training process is triggered to optimize the model decision boundary; The updated model parameters generated in S53 will be used to correct the health decay rate calculation algorithm in S3.

5. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, It also includes, S6. The abnormal propagation suppression process includes the following sub-steps: S61. Monitor the intermediate data quality indicators for each stage from S2 to S5; S62. When a data verification failure is detected, freeze the current processing pipeline and activate the redundant channel; S63. Add an isolation flag to the abnormal data stream and reinitialize the input buffer of the affected module.

6. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, The re-initialization method of S63 includes, S631. Backtrack the feature vector copy of the most recent valid state; S632. Reconstruct the intermediate state of the processing pipeline based on the backtracking data; S633. Resume operation after verifying the consistency between the reconstructed state and the current sensor readings.

7. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, The multimodal data fusion of S4 also includes, S43. Establish the physical constraint equations relating temperature, vibration, and electrical parameters; S44. Perform constraint satisfaction verification on the fusion results. When physical association constraints are violated... S441. Adjust the weight coefficients in the strategy matrix; S442. Re-execute time-frequency alignment and weighted fusion; S443. Record constraint conflict events to the model update training set.

8. The method for monitoring the condition of power equipment based on the Internet of Things according to claim 1, characterized in that, The feedback adjustment methods in step S5 include, S501. Analyze the feature contribution distribution of the health assessment model output; S502. Generate sensor sensitivity adjustment instructions based on contribution distribution; S503. Dynamically reconstructs the topology of the sensor array to improve the sampling density of key data channels.

9. A power equipment condition monitoring system based on the Internet of Things (IoT), comprising the power equipment condition monitoring method based on the IoT as described in any one of claims 1-8, characterized in that, include, A multi-source sensor array is used to collect operating data of power equipment and generate a raw dataset containing temperature, vibration, and electrical parameters. The value range verification module is used to receive the original dataset and output the first compliant dataset; The time-domain processing module is used to perform mutation detection and adaptive filtering on the first compliant dataset to generate the second corrected dataset; The frequency domain analysis module is used to perform multi-scale decomposition on the second corrected dataset and output the feature spectrum; The dynamic decision center is used to analyze the energy distribution of abnormal frequency bands in the feature spectrum and generate a strategy matrix that includes sensor priority, communication protocol, and computing resource allocation. A multimodal fusion unit is used to receive and execute policy matrix instructions. Time-frequency alignment unit is used to synchronize the timestamps of data from multiple source sensors; The weighted fusion unit is used to generate device state feature vectors based on policy weights; A self-learning model library is used to store and update health assessment models and convert model outputs into feedback control signals. The closed-loop control interface converts feedback control signals into real-time adjustment commands for a multi-source sensor array.

10. The IoT-based power equipment condition monitoring system according to claim 9, characterized in that, Its features are, It also includes an anomaly isolation subsystem, which contains... Data quality probes monitor the intermediate processing results of each module; A redundant channel switch activates a backup processing path when an abnormal data flow is detected. The state reconstruction unit restores the processing pipeline based on historical valid data: the anomaly isolation subsystem establishes a bidirectional connection with the self-learning model library and imports the features of abnormal events into the model training set.

Images