Power network state recognition method and system based on data feature analysis

Abstract

The application provides a power network state recognition method and system based on data feature analysis, which continuously improves the power network state recognition accuracy through a data-driven optimization mechanism, provides full-process intelligent decision support for power grid operation and maintenance, and effectively promotes the mode transformation of the power network from passive operation and maintenance to active prevention and control. The method comprises: obtaining a multi-dimensional power data set of a target power network, performing dynamic feature extraction processing on the multi-dimensional power data set to generate a power network operation feature set, inputting the power network operation feature set into a debugged power state recognition model, jointly analyzing voltage fluctuation features, current phase offset features and equipment health degree features through the power state recognition model, generating a power network state classification result, generating a network operation and maintenance strategy according to the power network state classification result, and sending the network operation and maintenance strategy to a power dispatch terminal to activate corresponding equipment maintenance operations or load adjustment operations.

Description

Technical Field

[0001] This application belongs to the field of power data analysis technology, specifically relating to a power network status identification method and system based on data feature analysis. Background Technology

[0002] As a complex energy transmission system, the accurate identification of the operating status of modern power grids is a core technology for ensuring power supply reliability. Traditional condition monitoring mainly relies on basic operating parameters such as voltage and current collected by SCADA systems, combined with threshold alarm mechanisms to achieve anomaly detection.

[0003] With the large-scale grid connection of new energy sources and the development of power Internet of Things (IoT) technology, the existing monitoring system has gradually exposed problems such as single data dimensions and insufficient analytical granularity. In recent years, the industry has begun to explore processing power data through machine learning algorithms, but most of these are limited to feature analysis on a single time scale, making it difficult to effectively capture the complex state characteristics of the power grid in dynamic operation.

[0004] This reveals significant application bottlenecks in existing solutions: on the one hand, static feature analysis systems struggle to adapt to the dynamic operational needs of modern power grids, failing to provide effective early warnings in the initial stages of system state evolution; on the other hand, discretized state judgment mechanisms lack the ability to collaboratively analyze multi-source data, resulting in insufficient alignment between generated operation and maintenance strategies and actual operating conditions. When dealing with complex power grid faults, existing technologies often require manual intervention for feature correlation analysis, significantly impacting the timeliness of emergency response. These technological shortcomings severely restrict the proactive defense capabilities of smart grids, making the development of a more intelligent power network state identification system a pressing technical challenge. Summary of the Invention

[0005] This application provides a power network status identification method and system based on data feature analysis, which continuously improves the accuracy of power network status identification through a data-driven optimization mechanism, provides intelligent decision support for the entire process of power grid operation and maintenance, and effectively promotes the transformation of power network from passive operation and maintenance to proactive prevention and control.

[0006] In a first aspect, embodiments of this application provide a power network status identification method based on data feature analysis, applied to a power network status identification system. The method includes: acquiring a multi-dimensional power data set of a target power network, the multi-dimensional power data set including real-time operating data, historical load data, and equipment status monitoring data, wherein the equipment status monitoring data is collected by sensors deployed in power network nodes; performing dynamic feature extraction processing on the multi-dimensional power data set to generate a power network operating feature set, the power network operating feature set including voltage fluctuation features, current phase shift features, and equipment health features; inputting the power network operating feature set into a debugged power status identification model, and performing joint analysis of the voltage fluctuation features, current phase shift features, and equipment health features through the power status identification model to generate a power network status classification result, the classification result including normal status, abnormal warning status, and fault alarm status; generating a network operation and maintenance strategy based on the power network status classification result, and sending the network operation and maintenance strategy to a power dispatch terminal to activate corresponding equipment maintenance operations or load adjustment operations.

[0007] Secondly, embodiments of this application provide a power network status identification system, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above-described method.

[0008] Thirdly, embodiments of this application provide a computer-readable storage medium including a computer program, which, when run on a power grid status identification system, causes the power grid status identification system to perform the steps of the above-described method.

[0009] In this application, dynamic perception and precise decision-making of the power network status were achieved through collaborative analysis and intelligent feature modeling that integrates multi-dimensional power data. First, by collecting heterogeneous data sources covering real-time operation, historical load, and equipment status, a dynamic feature system reflecting the overall operation of the power grid was constructed, effectively overcoming the limitations of traditional single-data-dimensional analysis. Second, dynamic feature extraction technology was used to collaboratively mine voltage fluctuation features, current phase shift features, and equipment health features, not only capturing instantaneous changes in the power grid's operational status but also establishing a composite criterion system through the correlation mapping of multi-dimensional features. Then, a power status identification model built based on machine learning algorithms, by jointly analyzing the coupling relationship between equipment health and power grid operating parameters, achieved penetrating identification from basic parameter anomalies to system-level state evolution, significantly improving the predictive judgment capability for potential faults. Finally, the intelligent mapping mechanism between classification results and operation and maintenance strategies enables the system to automatically trigger differentiated handling plans according to different status levels, ensuring both rapid response to sudden faults and preventative intervention in early abnormal states. This design can continuously improve the accuracy of power network status identification through data-driven optimization mechanisms, providing intelligent decision support for the entire process of power grid operation and maintenance, and effectively promoting the transformation of power network from passive operation and maintenance to proactive prevention and control. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating a power network status identification method based on data feature analysis, provided in an embodiment of this application.

[0011] Figure 2 This is a schematic diagram of the structure of a power network status identification system provided in an embodiment of this application. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0013] See Figure 1 This is a power network status identification method based on data feature analysis provided in the embodiments of this application. This method can be applied to a power network status identification system. The specific process is as follows: steps S110-S140.

[0014] Step S110: Obtain a multi-dimensional power data set of the target power network. The multi-dimensional power data set includes real-time operation data, historical load data, and equipment status monitoring data. The equipment status monitoring data is collected by sensors deployed in the power network nodes.

[0015] In this embodiment, the target power network is a large-scale power supply network covering multiple areas. For real-time operational data, the power network status identification system (hereinafter referred to as the system) continuously collects parameters such as voltage and current on each power line. For example, on one of the main transmission lines A, the voltage value is recorded once per second; for instance, at a selected moment, the recorded voltage value is 225V, and the recorded current value is 50A.

[0016] Historical load data is the electricity load information of various power network nodes over a period of time. This data is organized and stored hourly. For example, in the past week, node B's load was 1000kW at 8:00 AM on Monday and 1100kW at 9:00 AM, forming a time series data.

[0017] Equipment condition monitoring data is collected through various sensors deployed at different nodes in the power network. Taking transformer equipment as an example, vibration sensors can be deployed on its outer surface, a distributed temperature sensor array can be embedded between the internal windings and heat sinks, and an insulation monitoring device can be installed in the grounding circuit. The vibration sensors collect mechanical vibration waveform data of the equipment in real time, the distributed temperature sensor array collects temperature sampling data from multiple parts of the equipment, and the insulation monitoring device collects leakage current waveform data in real time. Through these sensors, comprehensive information on the operating status of the equipment can be obtained.

[0018] Step S120: Perform dynamic feature extraction processing on the multi-dimensional power data set to generate a power network operation feature set, which includes voltage fluctuation features, current phase offset features, and equipment health features.

[0019] In detail, the process of dynamically extracting features from the multi-dimensional power data set to generate a power network operation feature set includes: performing waveform decomposition processing on the real-time operation data to obtain the fundamental component and harmonic components, and generating voltage fluctuation features based on the amplitude stability index of the fundamental component and the amplitude change rate of the harmonic components; performing time series alignment processing on the historical load data to extract the load change gradient of each power network node within a preset time window, and generating current phase offset features by calculating the differences in load change gradients between adjacent time windows; performing abnormal signal detection on the equipment status monitoring data to obtain equipment vibration amplitude, temperature change rate, and insulation resistance value, and performing weighted fusion processing on the ratio of the vibration amplitude to a preset safety threshold, the time integral value of the temperature change rate, and the decay rate of the insulation resistance value to generate the equipment health feature; and determining the power network operation feature set based on the voltage fluctuation feature, the current phase offset feature, and the equipment health feature.

[0020] When performing waveform decomposition on real-time operational data, taking the voltage waveform of transmission line A as an example, existing waveform decomposition algorithms (such as Fourier transform decomposition methods) are used to decompose the voltage waveform into fundamental and harmonic components. For example, after decomposition, the amplitude of the fundamental component is 220V, and the amplitude of a certain harmonic component is 5V.

[0021] Then, voltage fluctuation characteristics are generated based on the amplitude stability index of the fundamental component and the amplitude change rate of the harmonic components. The fundamental component amplitude stability index is determined by calculating the range of change in the fundamental amplitude over a period of time. For example, if the fundamental amplitude fluctuates between 218V and 222V within 10 minutes, its fluctuation range is calculated to be 4V, and then converted into a stability index according to the corresponding rules. The harmonic component amplitude change rate is calculated by calculating the proportion of change in harmonic amplitude at adjacent time points. For example, if the harmonic amplitude is 4V in the previous minute and 5V in the current minute, the change rate is (5-4) / 4 = 0.25. Finally, these information are combined to generate voltage fluctuation characteristics.

[0022] Historical load data undergoes time-series alignment processing. Taking node B as an example, the preset time window is 1 hour. The load change gradient of each power network node within the preset time window is extracted. For example, if the load of node B increases from 1000kW to 1100kW between 8:00 and 9:00 on a certain day, the load change gradient is (1100-1000) / 1 = 100kW / h. Current phase shift characteristics are generated by calculating the differences in load change gradients between adjacent time windows. For example, if the load change gradient in the previous time window (7:00 to 8:00) is 80kW / h, the difference is 100-80 = 20kW / h. The differences between multiple time windows are calculated in this way to generate current phase shift characteristics.

[0023] Abnormal signal detection is performed on equipment condition monitoring data. Taking transformer equipment as an example, the vibration amplitude, temperature change rate, and insulation resistance value are obtained. Vibration waveform data collected by vibration sensors is processed to obtain the vibration amplitude; for example, the vibration amplitude at a certain moment is 0.5 mm. A distributed temperature sensor array collects temperature sampling data from multiple locations, and after spatial alignment processing, an internal temperature distribution matrix is ​​generated. For example, if the temperature at one monitoring point in the matrix is ​​50℃ at a target time, and the temperature at an adjacent monitoring point is 52℃, the temperature gradient difference is calculated. Leakage current waveform data collected by the insulation monitoring device is high-pass filtered to extract the harmonic component amplitude outside the power frequency cycle; for example, the extracted harmonic component amplitude is 0.1A. The ratio of vibration amplitude to a preset safety threshold (which can be set to 1 mm), the time integral value of the temperature change rate, and the decay rate of the insulation resistance value are weighted and fused to generate equipment health characteristics. For example, the ratio of vibration amplitude to preset safety threshold is 0.5 / 1 = 0.5. The time integral value of temperature change rate is obtained by integrating the temperature change rate over a period of time. The decay rate of insulation resistance value is calculated by comparing insulation resistance values ​​at different time points. For example, these three values ​​can be assigned weights of 0.3, 0.3, and 0.4 respectively, and the weighted fusion is used to generate equipment health characteristics.

[0024] Next, based on the voltage fluctuation characteristics, current phase offset characteristics, and equipment health characteristics generated above, the set of power network operation characteristics is determined.

[0025] In another optional embodiment, the dynamic feature extraction process in step S120 further includes:

[0026] Step S121: Perform three-level wavelet packet decomposition on the mechanical vibration waveform data of the equipment, extract the energy distribution spectrum of the third-level wavelet node, generate mechanical vibration energy entropy based on the energy ratio of high-frequency subband to low-frequency subband in the energy distribution spectrum, match the mechanical vibration energy entropy with the energy entropy threshold range of bearing wear and component loosening in the historical fault case library, and output the mechanical component health index.

[0027] When processing the mechanical vibration waveform data of equipment, we continue to take transformer equipment as an example and perform three-level wavelet packet decomposition on the vibration waveform data collected by its vibration sensor. Wavelet packet decomposition is a method of decomposing a signal at different frequency scales, which allows for more detailed analysis of signal characteristics. After decomposition, the energy distribution spectrum of the third-level wavelet nodes is obtained. For example, at a selected time (which can be selected according to actual needs), the energy of the high-frequency subband is 10, and the energy of the low-frequency subband is 20. The energy ratio of the high-frequency subband to the low-frequency subband is calculated to be 10 / 20 = 0.5. Based on this energy ratio, the mechanical vibration energy entropy is generated. The mechanical vibration energy entropy is calculated by comprehensively calculating the energy ratio and other information using existing formulas. For example, the calculated mechanical vibration energy entropy is 0.3. Then, this mechanical vibration energy entropy is matched with the energy entropy threshold ranges of bearing wear and component loosening in the historical fault case library. The energy entropy threshold range of bearing wear in the historical fault case library can be set to 0.2 to 0.4, and the energy entropy threshold range of component loosening can be set to 0.4 to 0.6. By comparison, it is determined that the mechanical vibration energy entropy of 0.3 falls within the energy entropy threshold range of bearing wear. Thus, the mechanical component health index is output. For example, based on the matching result, the mechanical component health index is set to 0.8 (indicating that the mechanical component is in good health, but has a certain tendency to wear).

[0028] Step S122: Perform three-dimensional interpolation reconstruction on the internal temperature distribution matrix of the device to generate a three-dimensional temperature field distribution model inside the device. Extract the area ratio of high-temperature regions exceeding the preset safe temperature in the three-dimensional temperature field distribution model. Calculate the device heat dissipation efficiency attenuation coefficient based on the temporal correlation between the area ratio of the high-temperature regions and the speed of the device's cooling fan. Combine the heat sink surface temperature uniformity index to generate a thermal stability index.

[0029] Taking a transformer as an example, the internal temperature distribution matrix of the equipment is reconstructed using three-dimensional interpolation. This is achieved by spatially aligning the temperature sampling data from multiple locations collected by a distributed temperature sensor array. Three-dimensional interpolation reconstruction is a method of interpolating two-dimensional or multi-dimensional data in space to generate a more continuous and accurate three-dimensional model. This method generates a three-dimensional temperature field distribution model inside the equipment. For example, at a selected moment, the model shows a region with a high temperature inside the equipment. The proportion of the high-temperature region exceeding a preset safe temperature (60℃) in the three-dimensional temperature field distribution model is extracted. Calculations show that the high-temperature region accounts for 10% of the total internal area of ​​the equipment. The temporal correlation between the proportion of the high-temperature region and the speed of the cooling fan is then analyzed. For example, over a period of time, as the proportion of the high-temperature region increases, the cooling fan speed also increases accordingly, but the rate of increase gradually slows down. By analyzing and calculating this correlation, the equipment's heat dissipation efficiency attenuation coefficient (e.g., 0.2) is calculated. Simultaneously, the surface temperature uniformity index of the radiator is considered. For example, by calculating the temperature difference at different locations on the radiator surface, a value representing temperature uniformity is obtained (e.g., 0.8, where a higher value indicates more uniform temperature). The heat dissipation efficiency attenuation coefficient and the radiator surface temperature uniformity index are comprehensively considered to generate a thermal stability index. For example, through weighted calculation (with weights set at 0.6 and 0.4 respectively), the thermal stability index = 0.6 × 0.2 + 0.4 × 0.8 = 0.44.

[0030] Step S123: Input the equipment insulation degradation trend index into the pre-trained insulation life prediction model, output the equipment insulation remaining life estimate, and generate the insulation aging correction coefficient based on the difference between the insulation remaining life estimate and the equipment operating years.

[0031] When processing data related to equipment insulation, the leakage current waveform data collected by the insulation monitoring device is high-pass filtered to extract the harmonic component amplitude. Based on its cumulative growth rate and abrupt change frequency over a continuous monitoring period, an equipment insulation degradation trend index is generated. For example, if the harmonic component amplitude increases from 0.05A to 0.1A over 10 consecutive monitoring periods, with a cumulative growth rate of (0.1-0.05) / 0.05 / 10 = 0.1 and an abrupt change frequency of 2, the existing algorithm generates an equipment insulation degradation trend index of 0.3. This equipment insulation degradation trend index is then input into a pre-trained insulation life prediction model. This model, trained on a large amount of historical data, can predict the estimated remaining insulation life of the equipment based on the input insulation degradation trend index. For example, the insulation life prediction model outputs an estimated remaining insulation life of 5 years. Given that the equipment has been in operation for 3 years, the difference between the estimated remaining insulation life and the equipment's operating years is calculated to be 5-3 = 2 years. An insulation aging correction factor is generated based on this difference, for example, by using an existing linear transformation rule to convert the difference into an insulation aging correction factor of 0.6.

[0032] Step S124: The health index of the mechanical components, the thermal stability index and the insulation aging correction coefficient are linearly superimposed according to preset weights to generate a comprehensive health score of the equipment, and the comprehensive health score of the equipment is mapped to a preset health level range to generate the health characteristics of the equipment.

[0033] The previously calculated mechanical component health index (set to 0.8), thermal stability index (set to 0.44), and insulation aging correction coefficient (set to 0.6) are linearly superimposed according to preset weights. With preset weights of 0.4, 0.3, and 0.3, the overall equipment health score is calculated as: 0.4 × 0.8 + 0.3 × 0.44 + 0.3 × 0.6 = 0.632. The preset health level ranges are: 0-0.4 for poor, 0.4-0.6 for average, 0.6-0.8 for good, and 0.8-1 for excellent. By comparing the overall equipment health score of 0.632 with this health level range, it is determined that it falls within the good range of 0.6-0.8. This generates the equipment health characteristic, which can be represented as a numerical value or label corresponding to the good level.

[0034] Step S130: Input the set of power network operation features into the power status identification model that has been debugged, and perform joint analysis on the voltage fluctuation features, the current phase offset features and the equipment health features through the power status identification model to generate power network status classification results. The classification results include normal status, abnormal warning status and fault alarm status.

[0035] In this embodiment, the aforementioned generated set of power network operation features is input into the power condition identification model after commissioning. The power condition identification model is an intelligent model that has undergone extensive training and optimization, capable of comprehensively analyzing and judging various input features. For example, the power condition identification model receives voltage fluctuation features, which include multi-dimensional information such as the voltage fluctuation range and stability over a period of time; current phase offset features include information such as the differences in load change gradients across different time windows; and equipment health features include comprehensive assessment information on equipment vibration, temperature, insulation, etc. The power condition identification model performs joint analysis of these features, calculating and judging them through internally embedded algorithms and rules.

[0036] After analysis and calculation by the power status identification model, when voltage fluctuation characteristics are within the normal range, current phase shift characteristics are also within the normal fluctuation range, and equipment health characteristics indicate that the equipment is operating well, the power status identification model determines that the power network is in a normal state. When voltage fluctuation characteristics show a certain degree of abnormality, such as the fluctuation range exceeding the set value, current phase shift characteristics also showing significant differences, and equipment health characteristics indicating potential problems in some equipment, the power status identification model determines that the power network is in an abnormal warning state. When voltage fluctuation characteristics show severe abnormalities, current phase shift characteristics show abrupt changes, and equipment health characteristics indicate serious potential faults in the equipment, the power status identification model determines that the power network is in a fault alarm state. Finally, the power network status classification results are generated, providing a basis for subsequent operation and maintenance decisions.

[0037] In one possible application scenario, in step S130, regarding voltage fluctuation characteristics, the voltage fluctuation range during normal operation is set to be within ±5% of the rated voltage, and the fluctuation frequency is no more than 10 times per minute. For example, if the rated voltage of the power network is 220V, and during the monitoring period, the actual monitored voltage fluctuation range is between 210V and 230V (i.e., a fluctuation range of ±4.5%), with 8 fluctuations per minute. Based on these values, the voltage fluctuation characteristics are within the normal range.

[0038] Regarding the current phase shift characteristics, the preset normal current phase difference is within ±10° and remains relatively stable over a long period of time. During the same monitoring period, the measured phase difference between the currents of each line was within ±8°, and the change in phase difference over the past hour was also minimal, remaining within ±2°, indicating that the current phase shift characteristics are also within normal limits.

[0039] Furthermore, the equipment health characteristics are generated by integrating multiple equipment parameters. Taking one key piece of equipment as an example, the normal threshold for equipment vibration amplitude is set to no more than 0.5 mm, the temperature change rate to no more than 5°C per hour, and the insulation resistance value to be stable above 10 MΩ. In this monitoring, the equipment vibration amplitude was 0.3 mm, the temperature rose from 40°C to 43°C in the past hour (i.e., the temperature change rate was 3°C / h), and the insulation resistance value was 12 MΩ. After comprehensively calculating these parameters (for example, through weighted summation, with vibration amplitude weighted at 0.3, temperature change rate weighted at 0.3, and insulation resistance value weighted at 0.4), the comprehensive score of the equipment health characteristics is 0.8 (this score is within the range indicating good equipment health).

[0040] It is understandable that after the power status identification model receives the above features, it performs joint analysis on these features based on the historical data it has learned internally and the established judgment rules. Since the voltage fluctuation features, current phase offset features, and equipment health features are all within the normal range, it can be determined that the power network is in a normal state.

[0041] In another application scenario, regarding voltage fluctuation characteristics, the voltage fluctuation range increased to ±10% of the rated voltage, reaching 200V-240V, and the fluctuation frequency increased to 15 times per minute. In terms of current phase offset characteristics, the phase difference widened to ±15°, and the phase difference changed frequently over the past half hour, with a fluctuation range of ±5°. Regarding equipment health characteristics, the equipment vibration amplitude increased to 0.7mm, the temperature change rate reached 7°C per hour, and the insulation resistance value decreased to 8MΩ. The comprehensive calculated equipment health characteristic score was 0.6 (indicating certain potential problems with the equipment). In this situation, after analyzing these characteristics, the power condition identification model determined that although a single characteristic had not reached an extremely serious level, multiple characteristics simultaneously showed abnormalities of varying degrees. Based on this, according to the pre-set rules within the model, a comprehensive judgment could be made that the power network was in an abnormal warning state, prompting maintenance personnel to pay attention to these potential problems and promptly investigate and handle them to prevent further deterioration.

[0042] In application scenarios exhibiting severe anomalies, voltage fluctuations manifest as a sudden voltage drop, instantly decreasing from 220V to 150V, and remaining unstable for several minutes with extremely high fluctuation frequencies. In terms of current phase shift characteristics, the phase difference abruptly increases to ±30° and cannot be restored to the normal range. Regarding equipment health characteristics, equipment vibration amplitude increases sharply to 1.2mm, temperature surges from 40℃ to 70℃ in a short period (the temperature change rate far exceeds the normal range), and insulation resistance drops sharply to 3MΩ. The overall calculated equipment health characteristic score is extremely low, only 0.2. Based on this, upon receiving the aforementioned severe anomaly characteristic values, the power condition identification model, according to its judgment mechanism, will clearly determine that the power network is in a fault alarm state.

[0043] For example, the power state identification model can jointly analyze the voltage fluctuation features, current phase shift features, and equipment health features to generate power network state classification results, which can be implemented based on machine learning. For instance, a decision tree algorithm can use features such as voltage fluctuations, current phase shifts, and equipment health as decision criteria, setting thresholds for different features as nodes, and classifying states based on the data's trajectory at each node, such as determining whether to enter an abnormal classification branch based on a voltage fluctuation amplitude threshold. Another example is a support vector machine, which can use kernel functions to map the above features to a high-dimensional space and find the optimal hyperplane to distinguish different state categories.

[0044] For example, the power network status classification result can be generated by jointly analyzing the voltage fluctuation characteristics, current phase shift characteristics, and equipment health characteristics through the power status identification model. This can also be achieved using deep learning technology. For instance, by constructing a network structure with multiple hidden layers, deep analysis of the input multi-dimensional features can be performed to learn the feature patterns corresponding to normal, abnormal warning, and fault alarm states from a large amount of historical data, thereby achieving accurate classification.

[0045] The two types of technologies mentioned above can achieve accurate classification of power network status through effective analysis and processing of features.

[0046] Step S140: Generate a network operation and maintenance strategy based on the power network status classification result, and send the network operation and maintenance strategy to the power dispatch terminal to activate the corresponding equipment maintenance operation or load adjustment operation.

[0047] Optionally, if the classification result is an abnormal warning state, then the following are extracted: the number of fluctuations exceeding a first threshold in the voltage fluctuation characteristics, the monitoring duration for a phase difference exceeding a preset angle in the current phase offset characteristics, and the rate of decrease in insulation resistance value in the equipment health characteristics. For example, the first threshold is set to 1.2 times the normal voltage fluctuation range, and statistically, the number of fluctuations exceeding this first threshold in the voltage fluctuation characteristics is 5. The preset angle is set to 15°, and the monitoring duration for a phase difference exceeding 15° in the current phase offset characteristics is 30 minutes. The rate of decrease in insulation resistance value in the equipment health characteristics is calculated by comparing insulation resistance values ​​at different time points, for example, a rate of decrease of 0.05 MΩ / h.

[0048] Based on the mapping relationship between the number of fluctuations and equipment overload risk, the correlation between monitoring duration and abnormal line impedance, and the correspondence between the rate of decrease and equipment aging, a composite maintenance instruction is generated, which includes partial load transfer, equipment infrared detection, and increased line inspection frequency. For example, according to the mapping relationship, 5 fluctuations correspond to a certain overload risk for the equipment, requiring partial load transfer; a monitoring duration of 30 minutes indicates that the line impedance may be abnormal, requiring increased line inspection frequency; and an insulation resistance decrease rate of 0.05 MΩ / h indicates that the equipment is showing signs of aging, requiring infrared detection. It is understandable that after generating the composite maintenance instruction, it is sent to the power dispatch terminal.

[0049] Optionally, if the classification result is a fault alarm state, an emergency response strategy is generated based on the identified equipment nodes whose vibration amplitude exceeds the second threshold in the equipment health characteristics and the line segments with abrupt phase difference changes in the current phase offset characteristics. This strategy includes faulty equipment isolation, backup power switching, and maintenance personnel dispatch. For example, the second threshold is set to 1.5 times the normal vibration amplitude, and the equipment node whose vibration amplitude exceeds this second threshold in the equipment health characteristics is node C. The line segment with abrupt phase difference changes in the current phase offset characteristics is line D. Based on this information, an emergency response strategy is generated, which involves isolating the faulty equipment node C, switching to the backup power supply to ensure power supply, and dispatching maintenance personnel to equipment node C for repairs. Similarly, after generating the emergency response strategy, it is sent to the power dispatch terminal to activate the corresponding operations.

[0050] In an optional embodiment, the execution process of the composite operation and maintenance instruction includes:

[0051] Step S141: Based on the transferable load capacity of the target power network node and the load carrying capacity of adjacent power network nodes in the local load transfer instruction, generate a load allocation scheme that includes the load reduction amount of the target power network node, the load receiving priority of adjacent power network nodes, and the maximum current threshold of the transmission line.

[0052] When executing composite operation and maintenance instructions, taking a local load transfer instruction as an example, the transferable load capacity of the target power network node E is 200kW, the load carrying capacity margin of the adjacent power network node F is 150kW, and the load carrying capacity margin of the adjacent power network node G is 100kW. A load allocation scheme is generated through calculation and analysis. First, the load reduction amount for the target power network node E is determined. Considering the carrying capacity of adjacent nodes, the load reduction amount can be determined to be 150kW. Then, the load receiving priority of adjacent power network nodes is determined. Based on the load carrying capacity margin and other factors (such as line distance, transmission loss, etc.), the load receiving priority of node F is determined to be higher than that of node G. Simultaneously, based on the parameters and safety requirements of the transmission line, the maximum current threshold of the transmission line is determined and set to 200A. Thus, a load allocation scheme including the load reduction amount, load receiving priority, and maximum current threshold can be generated.

[0053] Step S142: Based on the location identifiers of the target power network node and adjacent power network nodes in the load allocation scheme, start the circuit breaker control unit deployed on the tie line between the target power network node and the adjacent power network node, and simultaneously activate the distributed power output regulation module associated with the target power network node to perform the dynamic matching operation of load reduction amount and receiving priority defined in the load allocation scheme.

[0054] Based on the generated load allocation scheme, the location identifiers of the target power network node E and its adjacent power network nodes F and G are known. Using these location identifiers, the circuit breaker control unit deployed on the interconnecting lines between the target power network node E and its adjacent power network nodes F and G is activated. For example, via a remote control signal, the control circuit of the circuit breaker connecting nodes E and F is opened, enabling it to operate according to instructions. Simultaneously, the distributed power generation output regulation module associated with the target power network node E is activated. This distributed power generation can be a solar power system, which adjusts the output power of its solar panels to provide necessary power support during load transfer. Then, a dynamic matching operation between the load reduction amount and receiving priority defined in the load allocation scheme is executed. That is, according to the load receiving priority, 100kW of load is first transferred from node E to node F, and then 50kW of load is transferred to node G. During the transfer process, parameters such as current and voltage of the lines are monitored in real time to ensure that they do not exceed the maximum current carrying threshold of the transmission lines.

[0055] Step S143: Based on the location information of the target device in the infrared detection command, trigger the thermal imaging sensor group deployed on the wiring terminals, insulating sleeves and heat sink surfaces of the target device to collect the surface temperature distribution data of the target device during the steady-state operation phase after load transfer.

[0056] For the infrared detection command, taking transformer H as the target device, its location information is known. Based on this location information, the thermal imaging sensor group deployed on the terminals, insulating bushings, and radiator surface of transformer H is triggered. After the load transfer is completed and a period of stable operation has passed, the thermal imaging sensor group begins to work. For example, the thermal imaging sensor at the terminal collects temperature data once per second, and the sensors on the insulating bushings and radiator surface also collect data at a set frequency. These sensors can capture temperature information at different locations on the device surface, forming a multi-dimensional temperature distribution data set. For example, the temperature at point A of the terminal is 55℃, and at point B it is 58℃; the temperature of a part of the insulating bushing is 52℃; the temperature at the center of the radiator surface is 48℃, and the temperature at the edge is 45℃, etc. The above data constitute the temperature distribution data of the device surface.

[0057] Step S144: Compare the three-phase joint temperature difference gradient, radiator axial temperature decay rate, and insulating sleeve radial hot spot distribution characteristics in the surface temperature distribution data of the equipment with the reference temperature parameters of the corresponding components in the historical temperature distribution map pixel by pixel to generate infrared detection analysis results including the coordinates of potential overheating areas, temperature difference alarm levels, and heat dissipation anomaly types.

[0058] Optionally, the collected surface temperature distribution data of the equipment is analyzed in detail. First, the temperature gradient of the three-phase connectors is calculated. For example, if the three-phase connectors are U-phase, V-phase, and W-phase, with a temperature of 60℃ for phase U-phase, 65℃ for phase V-phase, and 70℃ for phase W-phase, the temperature gradient is obtained by calculating the ratio of the temperature difference between adjacent phase connectors to their distance: (65-60) / preset distance = first target temperature gradient value (where the distance is set to one unit length), (70-65) / preset distance = second target temperature gradient value. The radiator axial temperature decay rate is analyzed along the radiator's axial direction, examining the temperature change with distance. For example, from the radiator's center to its edge, the temperature is measured at preset distances, and the temperature decay rate is calculated. The radial hot spot distribution characteristics of the insulating sleeve determine the areas with higher temperatures and their distribution along the radial direction of the insulating sleeve.

[0059] Then, these features are compared pixel-by-pixel with the reference temperature parameters of the corresponding components in the historical temperature distribution map. The historical temperature distribution map is a reference standard formed by collecting a large amount of data under normal equipment operation. For example, the normal temperature difference gradient range of the three-phase joint in the historical map is within a preset value range. If the temperature difference gradient of the currently collected data exceeds this range, it is marked as abnormal. A similar comparison is made for the axial temperature decay rate of the radiator and the radial hot spot distribution characteristics of the insulating sleeve.

[0060] Infrared detection analysis results are generated through comparison. If the temperature of the target area is determined to be significantly higher than the historical baseline temperature, the area is identified as a potential overheating area, and its coordinates are recorded. The temperature difference alarm level is determined based on the magnitude of the temperature difference; for example, a temperature difference within a certain range is a low-level alarm, while a larger range is a high-level alarm. The type of heat dissipation anomaly is determined based on the abnormal temperature distribution of the radiator, such as whether there is local blockage leading to uneven heat dissipation. For example, the analysis results show a potential overheating area at the location coordinates (x, y) of the insulating sleeve, with a high temperature difference alarm level and a heat dissipation anomaly type of local obstruction of the heat dissipation channel.

[0061] Step S145: Based on the coordinates of the potential overheated area and the temperature difference alarm level in the infrared detection analysis results, extract the topological connection relationship of the line segment where the target device is located and the energized state of adjacent intervals, and calculate the safe approach path and detection dwell time of the inspection robot to the potential overheated area in the line inspection frequency increase command.

[0062] Based on the infrared detection analysis results, the coordinates of potential overheated areas and the temperature difference alarm level are known. Taking the target equipment transformer H as an example, the topological connection relationship of its line segment is extracted, including the connection method of this line with other lines, node distribution, and other information. Detailed information can be obtained through the power network topology database, such as which other equipment and line nodes the line containing transformer H is connected to. Simultaneously, the energization status of adjacent bays is determined. Through real-time monitoring system or related electrical equipment status information, it can be determined whether adjacent bays are energized and their voltage levels.

[0063] Based on the above information, a safe approach path for the inspection robot to potentially overheated areas is calculated. Considering safety factors, the inspection robot cannot directly approach energized areas; a safe path needs to be planned based on the topological connections and energized status. For example, by avoiding energized gaps, it moves along designated line branches and nodes to reach the vicinity of the potential overheated area. Simultaneously, the detection dwell time is calculated based on factors such as the importance of the potential overheated area, the temperature difference alarm level, and the equipment's operating status. For potential overheated areas with high temperature difference alarm levels, the inspection robot may need to stay for a longer period for detailed inspection; for example, a dwell time of 10 minutes is calculated to ensure comprehensive acquisition of information such as temperature changes in the area.

[0064] Step S146: Based on the safe approach path and detection dwell time, update the inspection cycle parameters of the inspection robot, and re-plan the travel speed, image acquisition frequency and obstacle avoidance distance threshold in the inspection path planning parameters, so that the inspection robot performs multi-angle thermal imaging data supplementation and insulation defect location operations on the potential overheated area within the adjusted inspection cycle.

[0065] Optionally, based on the calculated safe approach path and detection dwell time, the parameters of the inspection robot are adjusted. First, the inspection cycle parameter is updated. For example, if the original inspection cycle was once a day, it is shortened to once every half day due to the need for more frequent inspections of potentially overheated areas. Then, the travel speed in the inspection path planning parameters is re-planned. Considering the need to reach the potentially overheated area on time on the safe approach path and complete the inspection, the travel speed is appropriately adjusted. For example, if the original travel speed was 2 kilometers per hour, it is adjusted to 1.5 kilometers per hour to ensure sufficient time for inspection operations.

[0066] Optionally, the image acquisition frequency can also be adjusted as needed. To obtain more detailed information on potential overheating areas, the image acquisition frequency can be increased, for example, from once every 5 minutes to once every 2 minutes. Simultaneously, the obstacle avoidance distance threshold can be adjusted based on the surrounding environment and equipment conditions of the potential overheating area. If there are obstacles around the potential overheating area, the obstacle avoidance distance threshold can be appropriately increased to ensure that the inspection robot does not collide with obstacles during operation; for example, the obstacle avoidance distance threshold can be adjusted from 0.5 meters to 1 meter. Through these parameter adjustments, the inspection robot can perform multi-angle thermal imaging data acquisition and insulation defect location operations on potential overheating areas within the adjusted inspection cycle. For example, after reaching a potential overheating area, the inspection robot can acquire thermal imaging data from different angles to further analyze whether insulation defects exist. By comparing thermal imaging data from different angles, the location and extent of insulation defects can be more accurately located.

[0067] In an optional embodiment, the power state identification model is obtained through the following steps:

[0068] Step S210: Obtain historical power dataset, which includes multi-dimensional power data samples from multiple historical time points and corresponding status labels.

[0069] When debugging the power status identification model, the first step is to acquire a historical power dataset, which covers the operational data of the power network over a relatively long period. For example, multi-dimensional power data samples were collected at different times each week over the past year. These samples include real-time operational data for each power network node, such as specific voltage and current values; historical load data, recording the power load of each node at different time periods; and equipment status monitoring data, including data on equipment vibration, temperature, insulation, etc. Each data sample also has a corresponding status label, for example, "Normal" for a normal state, "Warning" for an abnormal warning state, and "Fault" for a fault alarm state. Taking Tuesday at 10:00 AM as an example, the real-time operational data in one data sample shows that the voltage of one line is 222V and the current is 48A; historical load data shows that the load change at this node has been stable over the past hour; and equipment status monitoring data shows that the equipment vibration amplitude is within the normal range, the temperature is normal, and the insulation resistance value is stable. The status label for this data sample is "Normal." By collecting a large number of these historical data samples and their corresponding labels, a historical power dataset is constructed.

[0070] Step S220: Perform feature extraction processing on the multi-dimensional power data samples at each historical time point to generate historical voltage fluctuation features, historical current phase shift features, and historical equipment health features, and combine the historical voltage fluctuation features, the historical current phase shift features, and the historical equipment health features into a set of historical feature vectors.

[0071] For real-time operational data, taking a historical point in time as an example, the waveform decomposition method is also used to decompose the voltage waveform into fundamental and harmonic components. For example, after decomposition, the amplitude of the fundamental component is 220V, and the amplitude of a certain harmonic component is 3V. The stability index of the fundamental component amplitude is calculated by statistically analyzing the fluctuation range of the fundamental amplitude over a period of time. For example, if the fundamental amplitude fluctuates between 219V and 221V in the past half hour, the fluctuation range is calculated as 2V, and this is converted into a stability index according to certain rules. The rate of change of the harmonic component amplitude is calculated by comparing the harmonic amplitude at adjacent time points. For example, if the harmonic amplitude was 2V at the previous moment and is 3V at the current moment, the rate of change is (3-2) / 2 = 0.5. These information are combined to generate historical voltage fluctuation characteristics.

[0072] Historical load data is processed, taking a power network node as an example, with a preset time window of half an hour. The load change gradient of this node within the preset time window is calculated. For example, if the load increases from 800kW to 850kW within half an hour, the load change gradient is (850-800) / 0.5 = 100kW / h. The difference in load change gradient between adjacent time windows is calculated. For example, if the load change gradient in the previous half hour was 90kW / h, the difference is 100-90 = 10kW / h. Historical current phase offset characteristics are generated through the above calculations.

[0073] The equipment status monitoring data is processed to obtain the equipment vibration amplitude, temperature change rate, and insulation resistance value. For example, at a historical time point, the equipment vibration amplitude is 0.3 mm, the temperature change rate is calculated by comparing the temperature at different time points, such as the temperature rising from 45℃ to 48℃ in the past hour, the temperature change rate is (48-45) / 1 = 3℃ / h, and the insulation resistance value is 5MΩ. The ratio of vibration amplitude to a preset safety threshold (set to 0.8 mm), the time integral value of the temperature change rate (e.g., calculated as 2℃ after integration), and the decay rate of the insulation resistance value (e.g., calculated as 0.02MΩ / h by comparing the insulation resistance values ​​at different time points) are weighted and fused to generate historical equipment health characteristics. The weights are set to 0.3, 0.3, and 0.4, and the weighted fused historical equipment health characteristics are 0.3×(0.3 / 0.8)+0.3×2+0.4×0.02=0.6525.

[0074] The generated historical voltage fluctuation features, historical current phase shift features, and historical equipment health features are combined into a set of historical feature vectors. For example, a historical feature vector might be [historical voltage fluctuation feature value, historical current phase shift feature value, historical equipment health feature value]. In this way, each data sample at a historical time point corresponds to a historical feature vector, and numerous historical feature vectors constitute a set of historical feature vectors.

[0075] Step S230: Obtain an initial state classification model, which includes a feature weight allocation layer and a state decision layer. The feature weight allocation layer is used to dynamically adjust the weight coefficients according to the contribution of different features in the historical feature vector set.

[0076] In this embodiment, an initial state classification model is obtained, which consists of a feature weight allocation layer and a state decision layer. The function of the feature weight allocation layer is to dynamically adjust the weight coefficients based on the contribution of different features in the historical feature vector set. For example, in historical data, voltage fluctuation features play an important role in determining whether the power network is in a fault alarm state, while current phase offset features contribute significantly in determining abnormal warning states, and equipment health features have a certain impact on the overall state assessment. The feature weight allocation layer dynamically adjusts the weight coefficients of each feature based on these factors.

[0077] In the initial stage, an initial weight may be assigned to each feature, for example, historical voltage fluctuation feature weight is 0.3, historical current phase offset feature weight is 0.3, and historical device health feature weight is 0.4. As the model is trained and the data is continuously analyzed, the weight allocation layer will dynamically adjust these weight coefficients according to the performance of different features in actual classification to improve the classification accuracy of the model.

[0078] Step S240: Input the set of historical feature vectors into the initial state classification model for iterative debugging, calculate the difference between the output of the state decision layer and the state label through a preset loss function, and update the parameters of the feature weight allocation layer in reverse until the difference is lower than a preset threshold.

[0079] The set of historical feature vectors is input into the initial state classification model for iterative debugging. Each time a historical feature vector is input, the state decision layer calculates and analyzes the feature vector based on the weight coefficients assigned by the feature weight allocation layer, outputting a classification result. For example, for a historical feature vector, the state decision layer calculates and outputs a classification probability distribution based on the current weight coefficients, such as a normal state probability of 0.6, an abnormal warning state probability of 0.3, and a fault alarm state probability of 0.1.

[0080] The difference between the output of the state decision layer and the state label is calculated using a preset loss function. Existing loss functions such as cross-entropy loss can be used. Taking cross-entropy loss as an example, the difference between the classification probability distribution output by the state decision layer and the state label is calculated. Based on the calculated difference, the parameters of the feature weight allocation layer are updated in reverse. Optimization algorithms such as gradient descent are used to adjust the weight coefficients of each feature in the feature weight allocation layer according to the magnitude and direction of the difference. For example, if it is determined that the voltage fluctuation feature has a low weight, leading to inaccurate classification, its weight coefficient is increased. This process is repeated continuously, inputting all vectors from the historical feature vector set into the model for calculation and adjustment until the difference is lower than a preset threshold. The preset threshold is a pre-set value, such as 0.1. When the difference is lower than this threshold, the model is considered to have reached good performance, and the debugging process ends.

[0081] In an optional embodiment, the method further includes:

[0082] Step S310: After the power dispatch terminal performs equipment maintenance operations, it continuously collects the operating status data of the equipment under maintenance.

[0083] After equipment maintenance is performed at the power dispatch terminal, the operational status of the maintained equipment is continuously monitored. For example, for a transformer that has undergone maintenance, various sensors deployed around it continuously collect operational status data. These sensors include vibration sensors, which collect real-time mechanical vibration waveform data to monitor vibration during operation; temperature sensors, which monitor temperature changes in different parts of the equipment to ensure that the temperature remains within the normal range during operation; and current sensors, which collect current data to understand the equipment's load. For example, on the first day after maintenance, data is collected hourly. Vibration waveform data collected by the vibration sensors shows that the vibration amplitude of the equipment is 0.4 mm at a selected moment; the temperature sensor detects a temperature of 52℃ at a key internal part of the equipment; and the current sensor detects a current value of 45A. Continuously collecting this data provides support for subsequent evaluation of the maintenance effectiveness and verification of the model.

[0084] Step S320: Input the operating status data into the power status identification model, and verify the accuracy of the classification result through the power status identification model to obtain the verification result.

[0085] The collected operational status data of the maintenance equipment is input into the power condition identification model. This model processes and analyzes this data. First, following the previous feature extraction methods, voltage fluctuation features, current phase shift features, and equipment health features are extracted from the operational status data. For example, current data is analyzed to calculate current fluctuations and generate current phase shift features; temperature and vibration data are processed to generate equipment health features. Then, the model classifies the data based on these features and outputs a classification result. This classification result is compared with a pre-defined expected status label. For example, if the expected status label is "normal" and the model outputs a classification result of "normal," the verification result is correct; if the model outputs a classification result of "abnormal warning" or "fault alarm," the verification result is incorrect. Through this method, the accuracy of the power condition identification model's classification results is verified, and the verification results are obtained to evaluate the model's reliability in practical applications.

[0086] Step S330: If the verification results of a preset number of consecutive times are inconsistent with the expected state label, the model parameter calibration process is activated to recalculate the weight allocation ratio of the voltage fluctuation feature, the current phase offset feature and the device health feature.

[0087] In this embodiment, a preset number of iterations can be set, such as five consecutive verification results that are inconsistent with the expected state label, to activate the model parameter calibration process. In this process, the weight allocation ratios of voltage fluctuation features, current phase shift features, and equipment health features are recalculated. First, the model is analyzed under what circumstances it misclassifies; for example, when some parameters in the equipment health features change, the model often misclassifies. Then, based on historical data and current verification results, the importance of each feature in the classification is reassessed. The weight allocation ratios are adjusted using optimization algorithms (such as genetic algorithms and simulated annealing algorithms). For example, if the original weights for voltage fluctuation features (0.3), current phase shift features (0.3), and equipment health features (0.4) were 0.4, after recalculation and adjustment, they might become 0.2, 0.4, and 0.4 respectively. Through this method, the model parameters are optimized, improving the model's classification accuracy and making it better suited to the actual operation of the power network.

[0088] In a standalone embodiment, after sending the network operation and maintenance strategy to the power dispatch terminal to activate the corresponding equipment maintenance operation or load adjustment operation, the method further includes:

[0089] Step S410: Real-time acquisition of voltage waveform data, current amplitude sequence, and equipment vibration signal of power network nodes after the execution of equipment maintenance operation or load adjustment operation; dynamic baseline comparison of the voltage waveform data to obtain voltage recovery stability index; phase synchronization analysis of the current amplitude sequence to obtain load balance index; and frequency domain energy spectrum decomposition of the equipment vibration signal to obtain mechanical condition improvement coefficient.

[0090] After the network operation and maintenance strategy is sent to the power dispatch terminal and the corresponding operation is activated, relevant data from power network nodes are collected in real time. For real-time voltage waveform data, for example, within 24 hours after equipment maintenance is completed, the real-time voltage waveform is collected every 10 minutes. The collected voltage waveform is compared with a pre-set dynamic baseline. The dynamic baseline is determined based on the voltage waveform characteristics during normal operation of the power network, such as the amplitude range and frequency of the normal voltage waveform. By comparing the currently collected voltage waveform with the dynamic baseline, indicators such as the voltage fluctuation range and the degree of deviation from the baseline are calculated to obtain the voltage recovery stability index. For example, if the amplitude of the currently collected voltage waveform fluctuates within ±5% of the dynamic baseline amplitude and the frequency is stable, then the voltage recovery stability can be considered good, and the corresponding voltage recovery stability index can be set to a higher value, such as 0.8; if the voltage waveform amplitude fluctuation range exceeds ±10%, or the frequency deviates significantly, the voltage recovery stability index will be reduced accordingly, such as set to 0.4.

[0091] For the current amplitude sequence, data is continuously collected over a period of time after the operation is executed. For example, a set of current amplitude data is collected every 5 minutes to form a current amplitude sequence. Phase synchronization analysis is then performed on this sequence. Phase synchronization analysis determines whether the load distribution in the power network is uniform by comparing the current phase relationships of different lines or devices. For example, by calculating the phase difference between the currents of each line, if the phase difference of most line currents is within a small range (e.g., ±10°), it indicates that the load distribution is relatively uniform and the load balance index is high, set at 0.7; if some line currents have large phase differences, exceeding ±20°, it indicates that the load is unbalanced and the load balance index is low, possibly 0.3.

[0092] For equipment vibration signals, vibration signal data is collected after operation. For example, vibration sensors installed on key parts of the equipment are used to collect vibration signals at a sampling frequency of 100 times per second. The collected vibration signals are then subjected to frequency domain energy spectrum decomposition. Using methods such as Fourier transform, the time-domain vibration signal is converted to the frequency domain to analyze the energy distribution of different frequency components. If the proportion of energy within the normal operating frequency range is high, and there are few abnormal energy components at high and low frequencies, it indicates that the mechanical condition of the equipment has improved well, and the mechanical condition improvement coefficient is high, for example, 0.6. Conversely, if large energy peaks appear in the abnormal frequency range, it indicates that the equipment may have potential mechanical failures, and the mechanical condition improvement coefficient is low, for example, 0.2.

[0093] Step S420: Input the voltage recovery stability index, load balance index and mechanical condition improvement coefficient into the preset operation and maintenance effect evaluation model, calculate the execution effect score of the equipment maintenance operation or load adjustment operation through the operation and maintenance effect evaluation model, and compare the execution effect score with the preset operation and maintenance target threshold.

[0094] The preset operation and maintenance effect evaluation model is a model that comprehensively evaluates the operation and maintenance effect based on multiple indicators. The voltage recovery stability index, load balance index, and mechanical condition improvement coefficient obtained earlier are input into this model, and the model performs weighted calculations on these indicators according to the pre-set weights.

[0095] For example, the voltage recovery stability index has a weight of 0.4, the load balancing index has a weight of 0.3, and the mechanical condition improvement coefficient has a weight of 0.3. The performance score = 0.4 × voltage recovery stability index + 0.3 × load balancing index + 0.3 × mechanical condition improvement coefficient. As another example, if the voltage recovery stability index is 0.6, the load balancing index is 0.5, and the mechanical condition improvement coefficient is 0.4, then the performance score = 0.4 × 0.6 + 0.3 × 0.5 + 0.3 × 0.4 = 0.51.

[0096] The preset operation and maintenance target threshold is a desired score value set according to the operation requirements and standards of the power network, for example, set to 0.6. The difference between the performance score and the preset operation and maintenance target threshold is calculated as 0.6 - 0.51 = 0.09. This difference provides a clear understanding of the gap between the performance of the operation and maintenance work and the expected target.

[0097] Step S430: If the difference comparison result exceeds the preset tolerance range, the load allocation ratio parameter is adjusted according to the deviation direction of the voltage recovery stability index, the equipment maintenance cycle parameter is adjusted according to the low-frequency energy ratio of the mechanical condition improvement coefficient, a dynamically updated network operation and maintenance strategy is generated and reissued to the power dispatch terminal.

[0098] The preset tolerance range is an allowable range of difference fluctuations, for example, set to ±0.05. When the difference comparison result exceeds the preset tolerance range, i.e., 0.09 > 0.05, the relevant parameters need to be adjusted.

[0099] If the voltage recovery stability index is lower than expected, it indicates that the voltage recovery is not ideal, which may be due to unreasonable load distribution. Based on the direction of deviation of the voltage recovery stability index, such as a persistently low voltage amplitude, it may be that some lines are overloaded, requiring adjustment of the load distribution ratio parameters. The original load distribution ratio for lines A, B, and C was set at 3:3:4. Analysis determined that line C was overloaded, affecting voltage recovery. Therefore, the load distribution ratio was adjusted to 3.5:3.5:3, reducing the load on line C.

[0100] For the mechanical condition improvement coefficient, analyze its low-frequency energy proportion. If the low-frequency energy proportion is high, it indicates that the equipment may have some chronic mechanical problems, and the equipment maintenance cycle needs to be appropriately shortened. For example, if the original equipment maintenance cycle is one year, based on the low-frequency energy proportion, the maintenance cycle can be adjusted to six months to promptly detect and address potential mechanical faults in the equipment.

[0101] Based on the above adjustments, a dynamically updated network operation and maintenance (O&M) strategy can be generated. The adjusted load allocation ratio parameters, equipment maintenance cycle parameters, and other information are integrated into the new O&M strategy and then redistributed to the power dispatch terminal. Upon receiving the new O&M strategy, the power dispatch terminal adjusts and controls the operation of the power network according to the new parameters and requirements to improve O&M efficiency and make the power network operation more stable and reliable.

[0102] In a standalone embodiment, after sending the network operation and maintenance strategy to the power dispatch terminal to activate the corresponding equipment maintenance operation or load adjustment operation, the method further includes:

[0103] Step S510: Obtain multi-channel high-frequency sampling data of the target power network node corresponding to the fault alarm state, perform transient signal separation processing on the multi-channel high-frequency sampling data, and extract voltage drop waveform segments, current impact pulse sequences and equipment vibration mutation signals at the moment the fault occurs.

[0104] When a power grid is in a fault alarm state, multi-channel high-frequency sampling data is acquired for the corresponding target power grid node. This data is collected simultaneously by multiple sensors deployed around the node, covering information on various aspects such as voltage, current, and equipment vibration. For example, during a fault, the voltage sensor collects voltage data at a high-frequency sampling frequency of 1000 times per second, while the current sensor and vibration sensor also collect data at corresponding high frequencies, forming multi-channel high-frequency sampling data.

[0105] Transient signal separation processing is performed on these multi-channel high-frequency sampled data. Transient signal separation uses existing algorithms and techniques to extract the corresponding signal components at the instant of the fault occurrence from complex mixed signals. For example, for voltage signals, by analyzing the signal's changing trends and characteristics, the moment of voltage drop is identified, and the voltage drop waveform segment is extracted. For instance, at the instant of the fault, the voltage rapidly drops from the normal 220V to 150V, forming a voltage drop waveform in the next few milliseconds, and this voltage drop waveform segment is accurately extracted.

[0106] For current signals, the same analysis and processing are performed to identify the current impulse pulse sequence. For example, at the moment of a fault, the current suddenly increases, forming a series of pulse signals. The amplitude, width, and other features of these pulse signals are extracted using signal processing techniques to form the current impulse pulse sequence.

[0107] For equipment vibration signals, if a sudden change in vibration is detected at the moment of a fault, the vibration change signal is extracted. For example, if the equipment originally had a small and stable vibration amplitude, but the vibration amplitude suddenly increases at the moment of the fault, this sudden vibration signal is separated from the entire vibration data.

[0108] Step S520: Perform traveling wave propagation path tracing analysis on the voltage drop waveform segment to determine the preliminary location range of the fault point; perform polarity feature matching on the current impact pulse sequence to identify the fault type as short circuit or grounding abnormality; and perform resonance frequency analysis on the equipment vibration mutation signal to locate the mechanically damaged components.

[0109] Optionally, traveling wave propagation path tracing analysis is performed on the extracted voltage sag waveform segments. Traveling wave propagation path tracing analysis is a method for determining fault location using the propagation characteristics of voltage traveling waves in a power network. When a fault occurs in a power network, a voltage sag generates traveling waves that propagate along the lines. By analyzing the characteristics of the voltage sag waveform, such as the propagation speed and characteristics of the reflected waves, combined with the power network topology information, a preliminary fault location range can be determined. For example, based on the time and speed information of the traveling wave propagation, it can be determined that the fault point may be located 50-80 kilometers along one of the lines; this range is the preliminary fault location range.

[0110] Optionally, polarity feature matching is performed on the current impulse pulse sequence. Different types of faults, such as short circuits and grounding anomalies, will result in different polarity characteristics in the current impulse pulse sequence. By analyzing the polarity, amplitude variation, and other characteristics of the current impulse pulses, a comparison and matching is made with the known characteristic patterns of short circuits and grounding anomalies. For example, if the polarity of the current impulse pulses exhibits an alternating pattern, and the amplitude variation matches the characteristics of a short circuit fault, then the fault type is identified as a short circuit; if the current impulse pulses exhibit a single polarity and have an amplitude variation pattern different from the normal situation, which matches the characteristics of a grounding anomaly, then the fault type is identified as a grounding anomaly.

[0111] Then, the resonant frequency of the equipment vibration mutation signal is analyzed. The equipment has a corresponding vibration frequency during normal operation. When the target component suffers mechanical damage, the vibration frequency changes, potentially leading to resonance. The resonant frequency components in the signal are identified through spectral analysis of the vibration mutation signal. For example, Fourier transform is used to convert the vibration mutation signal to the frequency domain, analyzing the energy distribution of each frequency component. If a significant energy peak appears at a preset frequency, and this frequency matches the theoretical resonant frequency of the target component, then the component can be identified as mechanically damaged. For instance, if analysis reveals a significant energy peak at 100Hz in the vibration mutation signal, and the theoretical resonant frequency of one of the bearings in the equipment is 100Hz, then the bearing can be preliminarily identified as a mechanically damaged component.

[0112] Step S530: Input the preliminary location range, fault type and mechanical damage component identification information into the preset fault root cause reasoning engine to generate a comprehensive analysis report containing fault point correction coordinates, fault mechanism description and related equipment impact links, and store the comprehensive analysis report in the fault case knowledge base for subsequent state recognition model optimization.

[0113] Optionally, the preliminary location range, fault type, and identification information of mechanically damaged components obtained from the previous analysis can be input into a preset fault root cause reasoning engine. The fault root cause reasoning engine is an intelligent analysis system based on rules and knowledge. It combines information such as power network operation knowledge, equipment characteristics, and historical fault cases to perform in-depth analysis and reasoning on the input information.

[0114] Based on the initial location range, the inference engine further combines detailed geographical information and line parameters of the power network to more accurately correct the location of the fault point, obtaining the corrected coordinates of the fault point. For example, if the initial location range is 50-80 kilometers along one of the lines, the inference engine considers factors such as the line's direction and terrain to determine the corrected coordinates of the fault point as its specific latitude and longitude.

[0115] For each fault type, the inference engine combines the principles of the power system with historical fault data to describe in detail the mechanism of the fault. For example, if the fault type is a short circuit, the inference engine will analyze the cause of the short circuit, such as damage to the line insulation or internal short circuits in the equipment, and will elaborate on the changes in current and voltage in the power network when the short circuit occurs.

[0116] For mechanically damaged components, the inference engine analyzes the impact of the damage on other related equipment, forming a chain of influence on these components. For example, if a bearing is identified as a mechanically damaged component, the inference engine will analyze the impact of the bearing's damage on connected components such as shafts and gears, as well as the potential cascading failures in other equipment.

[0117] Based on the above analysis and reasoning, a comprehensive analysis report is generated, including the fault point correction coordinates, fault mechanism description, and the impact of related equipment on the link. For example, the comprehensive analysis report might include the following: The fault point correction coordinates are (x, y); the fault mechanism is that long-term operation of a section of the line leads to insulation aging, eventually causing a short circuit fault, resulting in a voltage drop and current surge; the impact of related equipment on the link is that bearing damage leads to accelerated shaft wear, which may further affect the connected gear transmission system, posing a risk of causing the entire equipment to shut down.

[0118] Finally, the generated comprehensive analysis report is stored in the fault case knowledge base. The fault case knowledge base is a database that stores historical fault cases and analysis results, providing data support for subsequent optimization of the power network state identification model. When the model encounters similar fault situations during subsequent operation, it can refer to the cases in the knowledge base to improve the model's accuracy and fault diagnosis capabilities, thereby better ensuring the safe and stable operation of the power network.

[0119] In a standalone embodiment, after sending the network operation and maintenance strategy to the power dispatch terminal to activate the corresponding equipment maintenance operation or load adjustment operation, the method further includes:

[0120] Step S610: Continuously capture the strategy execution log data fed back by the power dispatch terminal, extract the maintenance item completion status, spare parts replacement records and maintenance time parameters in the equipment maintenance operation, and extract the power flow distribution change, node voltage recovery time and line loss fluctuation value in the load adjustment operation.

[0121] After the network operation and maintenance policy is sent to the power dispatch terminal and the corresponding operation is activated, the policy execution log data fed back by the power dispatch terminal is continuously captured. This log data records in detail the execution process and results of equipment maintenance operations and load adjustment operations.

[0122] For equipment maintenance operations, the completion status of maintenance items is extracted from the log data. For example, an equipment maintenance operation may include multiple maintenance items, such as checking the electrical connections of the equipment and replacing target components. The log data can determine whether each maintenance item has been completed according to the predetermined plan. For example, the record may show "Check equipment electrical connections: Completed, connections are secure; Replace transformer oil filter: Completed".

[0123] Extract the spare parts replacement record, noting which spare parts were replaced and when. For example, "The circuit breaker contacts of model XXX were replaced at XX time."

[0124] Extract the maintenance time parameters, including the start and end times of the entire maintenance operation, and calculate the maintenance time. For example, if the maintenance operation starts at 9:00 AM and ends at 2:00 PM, the maintenance time is 5 hours.

[0125] For load adjustment operations, the change in power flow distribution is extracted. Power flow distribution refers to the flow of power in the power network. By analyzing the power flow monitoring information in the log data, the changes in power flow distribution before and after the load adjustment are analyzed. For example, if the power flow of one line was 100MW before the adjustment and became 80MW after the adjustment, the change in power flow distribution is 20MW.

[0126] Extract the node voltage recovery time and record the time required for each node voltage to recover from a changed state to a stable state after a load adjustment operation. For example, a node voltage recovers to the normal range after 15 minutes following a load adjustment.

[0127] Line loss fluctuation values ​​are extracted, and the fluctuation of line loss is calculated by comparing the measured data before and after load adjustment. For example, if the line loss was 5MW before load adjustment and became 4MW after adjustment, the line loss fluctuation value is -1MW. These data provide detailed basis for evaluating the implementation effect of operation and maintenance strategies and subsequent parameter adjustments.

[0128] Step S620: Compare the completion status of the maintenance project with the completeness requirements of the preset maintenance procedure to generate a maintenance quality pass rate index, verify the compliance of the power flow distribution change with the power grid safety constraints, and generate a load adjustment safety factor.

[0129] Optionally, compare the completion status of maintenance items in the equipment maintenance operation with the step integrity requirements of the preset maintenance procedures. The preset maintenance procedures specify in detail each step and requirement that should be carried out during equipment maintenance. For example, for the maintenance of one of the transformers, the preset maintenance procedures require multiple steps such as electrical insulation testing and core grounding inspection. Compare the actually completed maintenance items with these requirements. If all the required steps are completed, the maintenance quality of the equipment is qualified; if some steps are not completed, it is unqualified. Generate a maintenance quality qualification rate index by calculating the ratio of the number of qualified equipment to the total number of maintained equipment among all the maintained equipment. For example, if a total of 10 equipment are maintained and all maintenance items of 8 equipment are completed according to the preset maintenance procedures, the maintenance quality qualification rate index = 8 / 10 = 0.8.

[0130] Furthermore, perform compliance verification on the change amount of power flow distribution in the load adjustment operation against the grid security constraint conditions. The grid security constraint conditions include restrictions in multiple aspects such as the maximum current-carrying capacity of the line and the allowable range of node voltages. For example, the maximum current-carrying capacity of one of the lines is 150A, and the current of this line caused by the change in power flow distribution after load adjustment is 120A, which is within the safe range, so this change in power flow distribution complies with the security constraint conditions; if the current exceeds 150A, it does not comply. Check all the lines and nodes involved in the load adjustment to determine whether the change in power flow distribution is within the range of the security constraint conditions. If all the changes are compliant, the load adjustment safety factor is 1; if some are non-compliant, determine a safety factor less than 1, such as 0.8, according to factors such as the degree of non-compliance. These indicators can intuitively reflect the quality and safety of the operation and maintenance operations and provide important references for subsequent decision-making.

[0131] Step S630: Adjust the acquisition frequency parameter of the equipment status monitoring data according to the maintenance quality qualification rate index, correct the weight allocation parameter of the historical load data according to the load adjustment safety factor, and synchronously update the adjusted parameters to the feature weight allocation layer of the power state recognition model.

[0132] Optionally, adjust the acquisition frequency parameter of the equipment status monitoring data according to the maintenance quality qualification rate index. If the maintenance quality qualification rate is relatively high, it indicates that the equipment maintenance work is relatively reliable and the risk of equipment failure is relatively low, so the acquisition frequency of the equipment status monitoring data can be appropriately reduced. For example, originally the equipment status monitoring data was collected once per hour, and when the maintenance quality qualification rate reaches 0.9, the acquisition frequency can be adjusted to once every two hours. On the contrary, if the maintenance quality qualification rate is relatively low, it indicates that there may be more potential problems with the equipment, and the acquisition frequency needs to be increased. For example, when the qualification rate is 0.6, the acquisition frequency is increased to once every half hour.

[0133] The weighting parameters of historical load data are adjusted based on the load adjustment safety factor. If the load adjustment safety factor is high, it indicates that the load adjustment operation has a relatively small impact on power grid security, and the weight of historical load data in the power state identification model can be appropriately reduced. For example, if the original weight of historical load data was 0.3, when the load adjustment safety factor is 0.9, the weight is adjusted to 0.25. If the load adjustment safety factor is low, it indicates that historical load data is more important for accurately judging the power network state, and its weight needs to be increased. For example, when the safety factor is 0.7, the weight is increased to 0.35.

[0134] The adjusted equipment condition monitoring data acquisition frequency parameters and historical load data weight allocation parameters are synchronously updated to the feature weight allocation layer of the power condition identification model. Based on these new parameters, the feature weight allocation layer adjusts the degree of attention given to different feature data, enabling the model to more accurately reflect the actual operating conditions of the power network, improving its performance and reliability, and providing stronger support for power network condition identification and operation and maintenance decisions. Through these steps and operations, accurate identification of the power network condition and effective operation and maintenance management can be achieved, ensuring the stable and safe operation of the power network.

[0135] This application's embodiments achieve dynamic perception and precise decision-making regarding power network status through collaborative analysis and intelligent feature modeling that integrates multi-dimensional power data. First, by collecting heterogeneous data sources covering real-time operation, historical load, and equipment status, a dynamic feature system reflecting the overall operation of the power grid is constructed, effectively overcoming the limitations of traditional single-data-dimensional analysis. Second, dynamic feature extraction technology is used to collaboratively mine voltage fluctuation features, current phase shift features, and equipment health features, not only capturing instantaneous changes in the power grid's operational status but also establishing a composite criterion system through the correlation mapping of multi-dimensional features. Then, a power status identification model built based on machine learning algorithms achieves penetrating identification from basic parameter anomalies to system-level state evolution by jointly analyzing the coupling relationship between equipment health and power grid operating parameters, significantly improving the predictive ability to identify potential faults. Finally, the intelligent mapping mechanism between classification results and operation and maintenance strategies enables the system to automatically trigger differentiated handling plans based on different status levels, ensuring both rapid response to sudden faults and preventative intervention in early abnormal states. This design can continuously improve the accuracy of power network status identification through data-driven optimization mechanisms, providing intelligent decision support for the entire process of power grid operation and maintenance, and effectively promoting the transformation of power network from passive operation and maintenance to proactive prevention and control.

[0136] Based on the same inventive concept, embodiments of this application also provide a power network status identification system. See also... Figure 2As shown, this is a schematic diagram of a possible power network status identification system provided in an embodiment of this application. Figure 2 In the power network status identification system 200, a processor 210 and a memory 220 are included. The memory 220 stores a computer program that can be executed by the processor 210. By executing the instructions stored in the memory 220, the processor 210 can perform the steps of the power network status identification method based on data feature analysis described above.

[0137] Based on the same inventive concept, embodiments of this application provide a computer-readable storage medium including a computer program. When the computer program is run on a power grid state identification system, it causes the power grid state identification system to perform the steps of the aforementioned power grid state identification method based on data feature analysis. In some possible implementations, various aspects of the power grid state identification method based on data feature analysis provided in this application can also be implemented as a program product, including a computer program. When the program product is run on a power grid state identification system, the computer program causes the power grid state identification system to perform the steps of the aforementioned power grid state identification method based on data feature analysis. For example, the power grid state identification system can perform actions such as... Figure 1 The steps are shown in the figure.

[0138] In the technical solutions involved in the above embodiments of this application, whether it is performing comparison calculations of multi-dimensional features or constructing composite parameters, if there are problems caused by significant differences in the number of dimensions, units of measurement, and semantic meanings of different features, those skilled in the art, based on their professional knowledge and past practical experience, can fully understand that these differences need to be properly handled so that the calculation results are accurate and comparable, and to avoid situations such as logical confusion and unclear mathematical meaning.

[0139] In detail, when faced with features of different numbers of dimensions, those skilled in the art can employ various strategies to accurately calculate the similarity, matching degree, or feature distance between different features.

[0140] Feature selection is a commonly used method. For a high-dimensional feature set, a subset of features that matches the number of low-dimensional features and is most representative can be selected based on indicators such as feature importance and relevance. Feature selection can be performed using methods such as chi-square test and information gain to filter out the features most valuable to the technical solution, thereby reducing the dimensionality of high-dimensional features to a level comparable to low-dimensional features, and then performing similarity or distance calculations.

[0141] Feature extraction is also an effective method. By constructing a suitable feature extraction model, features of different dimensions can be mapped to a common low-dimensional feature space. Principal Component Analysis (PCA) can not only handle differences in dimensions, but also project high-dimensional features onto a low-dimensional space composed of principal components, making features of different dimensions comparable in this low-dimensional space. In addition, deep learning models such as autoencoders can also be used for feature extraction. They can automatically learn the latent representation of input features, converting features of different dimensions into feature vectors of the same dimension, so as to perform subsequent similarity, matching degree, or feature distance calculations.

[0142] Alternatively, kernel methods can be used. Kernel functions can calculate the similarity between features in a high-dimensional space without explicitly mapping features to that space. For features with different numbers of dimensions, appropriate kernel functions can be selected, such as Gaussian kernels or polynomial kernels, to directly calculate their similarity. This method avoids the direct computational difficulties caused by different feature dimensions and can effectively measure the relationship between features in the original feature space or the implicit high-dimensional space.

[0143] In order to achieve comparability alignment of feature spaces, those skilled in the art can use a variety of existing and common technical means when processing the comparison of multidimensional features.

[0144] Standardization preprocessing is a widely used and effective method. It transforms the original feature data into a standard normal distribution with a mean of 0 and a standard deviation of 1 by applying a specific linear transformation. This approach fundamentally eliminates the influence of different dimensions on the data, allowing all features to be compared on the same scale. For example, in a dataset containing features with different dimensions, after standardization preprocessing, these features can be used to calculate similarity or distance on the same scale, avoiding computational bias caused by differences in dimensions.

[0145] Mapping transformation is also an effective way to solve the problem of dimensional differences. It can map the original features to a completely new space based on the specific properties of the features and actual business needs. In this new space, features with different dimensions can have better comparability. For features with nonlinear relationships, those skilled in the art can use logarithmic transformations, power transformations, etc., to convert them into linear relationships, thus facilitating similarity or distance calculations. For example, when dealing with features with exponential growth trends, logarithmic transformations can convert them into linear relationships, making subsequent calculations more accurate and convenient.

[0146] Spatial projection is also an important technique. It projects a high-dimensional feature space onto a low-dimensional space while preserving as much important information as possible between features. By carefully selecting appropriate projection directions and dimensions, those skilled in the art can effectively reduce the impact of dimensional differences on calculation results while reducing data dimensionality. Common spatial projection methods include Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA). Taking PCA as an example, it projects high-dimensional data into a low-dimensional space composed of principal components by finding the principal component directions, simplifying the data structure and reducing the interference of dimensional differences on feature comparison.

[0147] In the process of constructing composite parameters (such as loss function values), different parameter terms often have different dimensions. Those skilled in the art can use normalization processing or an adaptive weight allocation mechanism based on distribution characteristics.

[0148] Normalization unifies the value range of different parameter terms into a fixed interval, such as [0, 1]. This method eliminates the influence of differences in dimensions, ensuring that each parameter term has the same importance during weighted fusion. Common normalization methods include min-max normalization and Z-score normalization. Taking min-max normalization as an example, it performs a linear transformation on the parameter terms, scaling their value range to the [0, 1] interval, allowing parameter terms with different dimensions to be weighted and fused under the same standard.

[0149] An adaptive weight allocation mechanism based on distribution characteristics dynamically adjusts the weights of different parameter terms according to their distribution characteristics. For parameter terms with large variance, those skilled in the art can appropriately reduce their weights; for parameter terms with small variance, they can appropriately increase their weights. This allows the composite loss function to focus more on parameter terms with smaller variances, thereby improving the model's stability and generalization ability. For example, in a composite loss function containing multiple parameter terms, if the variance of a certain parameter term is large, it indicates that its fluctuations are relatively drastic, which may adversely affect the model's stability. In this case, reducing its weight can mitigate this adverse effect; conversely, for parameter terms with small variance, increasing their weights allows the model to focus more on the information reflected by that parameter term, thereby improving the overall performance of the model.

[0150] The aforementioned general techniques for solving feature matching and loss balance problems are all common knowledge in this field. These techniques have been fully verified and widely used in numerous practical applications, and those skilled in the art can skillfully and flexibly apply these methods to handle similar problems involving differences in dimensions.

[0151] The formulas and calculation processes involved in the embodiments of this application, whether used for multidimensional feature comparison or composite loss function construction, strictly adhere to the principle of dimensional correspondence. Each variable in the formula has a clear and explicit physical meaning, and its operational logic fully conforms to basic mathematical and physical logic. The calculation results are necessarily the reasonable results expected by this application. Those skilled in the art are capable of effectively solving various problems arising from the number of dimensions, dimensional differences, etc., in the multidimensional feature comparison calculation and composite loss function construction in the embodiments, based on specific data conditions and business needs, by comprehensively utilizing the aforementioned general technical means, thus ensuring the accuracy, reliability, and feasibility of the technical solution of this application.

Claims

1. A power network state identification method based on data feature analysis, characterized in that, The method includes: A multi-dimensional power data set of the target power network is obtained. The multi-dimensional power data set includes real-time operation data, historical load data and equipment status monitoring data. The equipment status monitoring data is collected by sensors deployed in power network nodes. Dynamic feature extraction processing is performed on the multi-dimensional power data set to generate a power network operation feature set, which includes voltage fluctuation features, current phase shift features, and equipment health features. The set of power network operation characteristics is input into the power status identification model after commissioning. The power status identification model is used to jointly analyze the voltage fluctuation characteristics, the current phase offset characteristics and the equipment health characteristics to generate power network status classification results. The classification results include normal status, abnormal warning status and fault alarm status. Based on the power network status classification results, a network operation and maintenance strategy is generated, and the network operation and maintenance strategy is sent to the power dispatch terminal to activate the corresponding equipment maintenance operation or load adjustment operation. The process of generating network operation and maintenance strategies based on the power network status classification results includes: If the classification result is an abnormal warning state, then extract the number of fluctuations exceeding the first threshold in the voltage fluctuation feature, the monitoring duration for the phase difference to continuously exceed the preset angle in the current phase offset feature, and the rate of decrease of the insulation resistance value in the equipment health feature. Based on the mapping relationship between the number of fluctuations and the risk of equipment overload, the correlation between the monitoring duration and the abnormal line impedance, and the corresponding rule between the rate of decrease and equipment aging, a composite operation and maintenance instruction is generated that includes local load transfer, equipment infrared detection, and increased line inspection frequency. If the classification result is a fault alarm state, then based on the identified equipment nodes whose vibration amplitude exceeds the second threshold in the equipment health characteristics and the line segments with abrupt phase difference in the current phase offset characteristics, an emergency response strategy including fault equipment isolation, backup power supply switching and maintenance personnel dispatch is generated.

2. The method as described in claim 1, characterized in that, The process of dynamically extracting features from the multi-dimensional power data set to generate a power network operation feature set includes: The real-time running data is subjected to waveform decomposition processing to obtain the fundamental component and harmonic components, and voltage fluctuation characteristics are generated based on the amplitude stability index of the fundamental component and the amplitude change rate of the harmonic components. The historical load data is subjected to time series alignment processing to extract the load change gradient of each power network node within a preset time window, and the current phase offset feature is generated by calculating the difference in load change gradient between adjacent time windows. Abnormal signal detection is performed on the equipment status monitoring data to obtain the equipment vibration amplitude, temperature change rate and insulation resistance value. The ratio of the vibration amplitude to a preset safety threshold, the time integral value of the temperature change rate and the decay rate of the insulation resistance value are weighted and fused to generate the equipment health characteristics. The set of power network operation characteristics is determined based on the voltage fluctuation characteristics, the current phase offset characteristics, and the equipment health characteristics.

3. The method as described in claim 1, characterized in that, The power status identification model was obtained through the following steps: Obtain a historical power dataset, which includes multi-dimensional power data samples from multiple historical time points and corresponding status labels; Feature extraction processing is performed on the multi-dimensional power data samples at each historical time point to generate historical voltage fluctuation features, historical current phase shift features, and historical equipment health features. The historical voltage fluctuation features, historical current phase shift features, and historical equipment health features are then combined into a set of historical feature vectors. An initial state classification model is obtained, which includes a feature weight allocation layer and a state decision layer. The feature weight allocation layer is used to dynamically adjust the weight coefficients according to the contribution of different features in the historical feature vector set. The historical feature vector set is input into the initial state classification model for iterative debugging. The difference between the output of the state decision layer and the state label is calculated using a preset loss function, and the parameters of the feature weight allocation layer are updated in reverse until the difference is lower than a preset threshold.

4. The method as described in claim 3, characterized in that, The reverse update process of the parameters of the feature weight allocation layer includes: Based on the timestamp distribution of the historical voltage fluctuation characteristics in historical fault cases, the cumulative frequency of voltage deviation from the baseline value at each historical time point is extracted, and a first weighting coefficient of the historical voltage fluctuation characteristics is generated according to the ratio of the cumulative frequency to a preset frequency threshold. The initial weighting coefficients are matched with the duration of the phase shift of the historical current phase shift feature in historical load change events in a time series. Abnormal intervals in which the duration of the phase shift exceeds a preset duration are extracted. The weighting coefficients of the historical current phase shift feature are dynamically adjusted according to the coverage of the abnormal intervals in the total historical monitoring time to obtain the second weighting coefficients. The dynamically adjusted weighting coefficients are coupled with the stability index of the historical equipment health characteristics in the continuous monitoring period. The stability index is obtained by calculating the linear combination of the equipment vibration amplitude range, temperature change rate variance and insulation resistance value attenuation in adjacent monitoring periods. The weighting coefficients of the historical equipment health characteristics are corrected based on the mapping relationship between the stability index and the equipment fault warning level to obtain the third weighting coefficient. The first weight coefficient, the second weight coefficient, and the third weight coefficient are input into the gradient descent algorithm of the feature weight allocation layer. The associated weight parameters of the historical voltage fluctuation feature, the historical current phase offset feature, and the historical equipment health feature are updated synchronously and iteratively until the error rate between the classification probability distribution output by the state decision layer and the state label converges to a preset range.

5. The method as described in claim 1, characterized in that, The execution process of the composite operation and maintenance command includes: Based on the transferable load capacity of the target power network node in the local load transfer instruction and the load carrying capacity of adjacent power network nodes, a load allocation scheme is generated that includes the load reduction amount of the target power network node, the load receiving priority of adjacent power network nodes, and the maximum current threshold of the transmission line. Based on the location identifiers of the target power network node and adjacent power network nodes in the load allocation scheme, the circuit breaker control unit deployed on the tie line between the target power network node and the adjacent power network node is activated, and the distributed power output regulation module associated with the target power network node is activated simultaneously to perform the dynamic matching operation of load reduction amount and receiving priority defined in the load allocation scheme. Based on the location information of the target device in the infrared detection command, the thermal imaging sensor group deployed on the terminal block, insulating sleeve and heat sink surface of the target device is triggered to collect the surface temperature distribution data of the target device during the steady-state operation phase after load transfer. The temperature difference gradient of the three-phase joint, the axial temperature decay rate of the radiator, and the radial hot spot distribution characteristics of the insulating sleeve in the surface temperature distribution data of the equipment are compared pixel by pixel with the reference temperature parameters of the corresponding components in the historical temperature distribution map to generate infrared detection analysis results that include the coordinates of potential overheating areas, temperature difference alarm levels, and heat dissipation anomaly types. Based on the coordinates of the potential overheated area and the temperature difference alarm level in the infrared detection analysis results, the topological connection relationship of the line segment where the target device is located and the energized state of adjacent intervals are extracted, and the safe approach path and detection dwell time of the inspection robot to the potential overheated area in the line inspection frequency increase command are calculated. Based on the safe approach path and detection dwell time, the inspection cycle parameters of the inspection robot are updated, and the travel speed, image acquisition frequency and obstacle avoidance distance threshold in the inspection path planning parameters are replanned, so that the inspection robot can perform multi-angle thermal imaging data supplementation and insulation defect location operations on the potential overheated area within the adjusted inspection cycle.

6. The method as described in claim 1, characterized in that, The process of collecting the equipment status monitoring data includes: Vibration sensors deployed on the surface of the power equipment casing collect mechanical vibration waveform data of the equipment. Fourier transform processing is performed on the mechanical vibration waveform data of the equipment to extract the target vibration frequency component and the corresponding amplitude value that matches the rotation frequency of the equipment bearing. The target vibration frequency component and its corresponding amplitude value are input into a preset vibration feature analysis window, and the equipment vibration intensity index is generated based on the range of amplitude values ​​and the number of fluctuation periods within the vibration feature analysis window. Temperature sampling data from multiple parts of the power equipment is collected by a distributed temperature sensor array embedded between the windings and heat sinks inside the power equipment. The temperature sampling data from multiple parts is spatially aligned to generate an internal temperature distribution matrix of the equipment. The thermal stability coefficient of the equipment is calculated based on the temperature gradient difference between adjacent monitoring points and the duration of the maximum temperature difference in the internal temperature distribution matrix of the equipment. Leakage current waveform data is collected in real time by an insulation monitoring device installed in the equipment grounding circuit. The leakage current waveform data is then subjected to high-pass filtering to extract the amplitude of harmonic components outside the power frequency cycle. Based on the cumulative growth rate and abrupt change frequency of the harmonic component amplitude within the continuous monitoring period, an equipment insulation degradation trend index is generated. The vibration intensity index, thermal stability coefficient, and insulation degradation trend index of the equipment are normalized and fused to generate a set of equipment condition monitoring data.

7. The method as described in claim 6, characterized in that, The dynamic feature extraction process also includes: The mechanical vibration waveform data of the equipment is decomposed into three-level wavelet packets, and the energy distribution spectrum of the third-level wavelet node is extracted. Based on the energy ratio of the high-frequency subband to the low-frequency subband in the energy distribution spectrum, the mechanical vibration energy entropy is generated. The mechanical vibration energy entropy is matched with the energy entropy threshold range of bearing wear and component loosening in the historical fault case library, and the mechanical component health index is output. The internal temperature distribution matrix of the device is reconstructed by three-dimensional interpolation to generate a three-dimensional temperature field distribution model inside the device. The area ratio of high-temperature regions exceeding the preset safe temperature in the three-dimensional temperature field distribution model is extracted. Based on the temporal correlation between the area ratio of high-temperature regions and the speed of the device's cooling fan, the device's heat dissipation efficiency attenuation coefficient is calculated. Combined with the heat sink surface temperature uniformity index, a thermal stability index is generated. The insulation degradation trend index of the equipment is input into the pre-trained insulation life prediction model, the remaining insulation life of the equipment is estimated, and the insulation aging correction coefficient is generated based on the difference between the remaining insulation life estimate and the equipment operating years. The mechanical component health index, thermal stability index, and insulation aging correction coefficient are linearly superimposed according to preset weights to generate a comprehensive equipment health score. The comprehensive equipment health score is then mapped to a preset health level range to generate the equipment health characteristics.

8. The method as described in claim 1, characterized in that, The method further includes: After the power dispatch terminal performs equipment maintenance operations, it continuously collects the operating status data of the equipment under maintenance. The operating status data is input into the power status identification model, and the accuracy of the classification result is verified by the power status identification model to obtain the verification result; If the verification results of a preset number of consecutive times are inconsistent with the expected state label, the model parameter calibration process is activated to recalculate the weight allocation ratio of the voltage fluctuation feature, the current phase offset feature, and the device health feature.

9. A power network status identification system, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of any one of the methods described in claims 1 to 8.

Images