Deep learning-based electrical equipment monitoring method, system, device, and medium

Abstract

The application discloses an electrical equipment monitoring method, system, device and medium based on deep learning, comprising the following steps: S1, obtaining historical data of electrical equipment, and preprocessing the historical data to obtain preprocessed data; S2, constructing a deep learning model based on the preprocessed data; S3, obtaining first running data of the electrical equipment at a first time and second running data of the electrical equipment at a second time; S4, inputting the first running data into the deep learning model to obtain continuous first comparison data graphs; S5, obtaining comparison points corresponding to the second time according to the first comparison data graphs; S6, obtaining a comparison interval of the second running data according to the comparison points; and S7, comparing the second running data with the comparison interval, and outputting an alarm signal if the second running data is not located in the comparison interval, so that the fault of the electrical equipment is accurately prevented.

Description

Technical Field

[0001] This invention belongs to the field of electrical equipment management technology, and in particular relates to a method, system, device and medium for monitoring electrical equipment based on deep learning. Background Technology

[0002] Stable operation of electrical equipment is the core guarantee for industrial production. However, its operating data is affected by many factors such as environment, lifespan, and load. Traditional monitoring methods have obvious limitations: relying solely on existing data thresholds for judgment cannot provide early warning of potential faults, which can easily lead to production line shutdowns; maintenance costs are high, and excessive warnings will increase maintenance workload, while insufficient warnings may cause safety accidents; the aging patterns of equipment in historical data are not fully explored, making it difficult to adapt to the dynamic changes throughout the equipment's life cycle; equipment in the same batch may have common defects, but traditional methods cannot conduct coordinated monitoring to avoid systemic risks.

[0003] Therefore, there is an urgent need to design a method, system, device, and medium for monitoring electrical equipment based on deep learning to solve the problems mentioned above. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, device, and medium for monitoring electrical equipment based on deep learning, so as to achieve accurate prevention of electrical equipment failures.

[0005] To achieve the above objectives, the specific technical solution of the present invention, which provides a deep learning-based method, system, device, and medium for monitoring electrical equipment, is as follows: A deep learning-based method for monitoring electrical equipment includes the following steps: S1. Obtain historical data of electrical equipment and preprocess the historical data to obtain preprocessed data; S2. Construct a deep learning model based on the preprocessed data; S3. Obtain the first operating data of the electrical equipment at a first time and the second operating data of the electrical equipment at a second time; S4. Input the first running data into the deep learning model to obtain a continuous first comparison data graph; S5. Obtain the comparison point corresponding to the second time based on the first comparison data graph; S6. Obtain the comparison interval of the second running data based on the comparison point; S7. Compare the second operating data with the comparison interval. If the second operating data is not within the comparison interval, output an alarm signal.

[0006] Furthermore, preprocessing the historical data to obtain preprocessed data includes the following steps: Historical data is filtered using one or a combination of wavelet transform, smoothing filter, low-pass filter, high-pass filter, and band-pass filter to obtain filtered data. The filtered data is denoised and normalized using time-series correlation and density clustering algorithms to obtain preprocessed data.

[0007] Further, obtaining the comparison interval of the second running data based on the comparison point includes the following steps: Set the upper and lower distances corresponding to the ordinate of the comparison point, obtain historical data at the third time point, and obtain the third running data; Determine whether the second running data is the same as the historical data corresponding to the second time. If they are not the same, increase the upper distance and lower distance by a first ratio. If they are the same, determine whether the difference between the second running data and the third running data is within the data fluctuation range. If it exceeds the data fluctuation range, decrease the upper distance and lower distance by a first ratio. The comparison interval is obtained by subtracting the adjusted lower distance from the ordinate of the comparison point and adding the adjusted upper distance to the ordinate of the comparison point.

[0008] Further, obtaining the comparison interval of the second running data based on the comparison point includes the following steps: Set the upper and lower distances corresponding to the ordinate of the comparison point, obtain historical data at the third time point, and obtain the third running data; Determine whether the first running data is the same as the historical data corresponding to the first time. If they are not the same, increase the upper distance and lower distance by a first ratio. If they are the same, determine whether the difference between the first running data and the third running data is within the data fluctuation range. If it exceeds the data fluctuation range, decrease the upper distance and lower distance by a first ratio. The comparison interval is obtained by subtracting the adjusted lower distance from the ordinate of the comparison point and adding the adjusted upper distance to the ordinate of the comparison point.

[0009] Furthermore, the deep learning-based electrical equipment monitoring method also includes the following steps: S8. After outputting the alarm signal, acquire the electrical equipment that outputs the alarm signal, obtain the electrical equipment of the same batch as the electrical equipment that outputs the alarm signal, and test the electrical equipment of the same batch.

[0010] Furthermore, the testing of the same batch of electrical equipment includes the following steps: S81. After setting a first preset number of times and reducing the upper and lower distances of the same batch of electrical equipment by a second ratio, the same batch of electrical equipment jumps to step S1 and runs steps S1 to S7. If an alarm signal is output again, step S81 is repeated until the number of jumps exceeds the first preset number of times, and all the same batch of electrical equipment outputs an alarm signal.

[0011] Furthermore, a second preset number of times is set. When the number of alarms of the electrical equipment exceeds the third preset number of times within a unit time, the operating data corresponding to the second time in the first comparison data graph of the electrical equipment that outputs the alarm signal is configured as the fourth operating data, the average value of the fourth operating data and the second operating data is configured as the fifth operating data, and the fourth operating data in the first comparison data graph of the electrical equipment that outputs the alarm signal is replaced with the fifth operating data to generate a second comparison data graph.

[0012] A deep learning-based electrical equipment monitoring system, employing the aforementioned deep learning-based electrical equipment monitoring method, includes: a preprocessing module, a deep learning module, an acquisition module, an input module, and a judgment module; The preprocessing module is used to preprocess historical data to obtain preprocessed data; The deep learning module is used to construct a deep learning model based on the preprocessed data; The acquisition module is used to acquire first operating data of the electrical equipment at a first time and second operating data of the electrical equipment at a second time. The input module is used to input the first running data into the deep learning model to obtain a continuous first comparison data image; The judgment module is used to obtain the comparison point corresponding to the second time and the comparison interval of the second running data according to the first comparison data graph, compare the second running data with the comparison interval, and output an alarm signal if the second running data is not located within the comparison interval.

[0013] A device comprising at least one processor and at least one memory; The at least one memory is used to store computer programs; The at least one processor is used to execute the computer program to implement the above-described method.

[0014] A computer-readable storage medium for storing executable instructions that, when executed by a processor, cause the processor to implement the method described above.

[0015] The deep learning-based electrical equipment monitoring method, system, device, and medium of the present invention have the following advantages: 1. Strong predictive power: It predicts the data trends of electrical equipment through deep learning models and combines dynamic intervals to detect potential faults in electrical equipment in advance; 2. Prevention closed loop: A complete closed loop is formed from alarm to maintenance and effect verification to reduce the failure recurrence rate; 3. Batch risk management: Linked monitoring and prevention of equipment in the same batch to avoid systemic risks; 4. Maintenance cost optimization: Dynamically adjust the maintenance cycle of electrical equipment based on data to reduce over-maintenance and downtime losses; 5. High versatility: Applicable to a variety of electrical equipment, and can adapt to the fault characteristics of different equipment through customized models. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the deep learning-based electrical equipment monitoring method of the present invention. Figure 2 This is a heatmap of the confusion matrix between predicted and actual values ​​under different operating conditions of the present invention. Figure 3 This is an accuracy diagnostic graph of the learning curve of the deep learning-based electrical equipment monitoring method of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the claims, any of the claimed embodiments can be used in any combination.

[0019] The following reference Figures 1 to 3 This invention describes a deep learning-based method, system, device, and medium for monitoring electrical equipment.

[0020] A deep learning-based method for monitoring electrical equipment, such as Figure 1 The steps shown are as follows: S1. Obtain historical data of electrical equipment and preprocess the historical data to obtain preprocessed data; S2. Construct a deep learning model based on the preprocessed data; S3. Obtain the first operating data of the electrical equipment at a first time and the second operating data of the electrical equipment at a second time; S4. Input the first running data into the deep learning model to obtain a continuous first comparison data graph; S5. Obtain the comparison point corresponding to the second time based on the first comparison data graph; S6. Obtain the comparison interval of the second running data based on the comparison point; S7. Compare the second operating data with the comparison interval. If the second operating data is not within the comparison interval, output an alarm signal.

[0021] Specifically, the system acquires the first operating data of the electrical equipment at a first time point (t1) (e.g., the high-voltage winding temperature of a dry-type transformer at time t1 is 88.3℃) and the second operating data at a second time point (t2, t2 > t1) (e.g., the actual temperature at time t2 is 91.4℃). Using historical data from the entire lifecycle of the electrical equipment (including parameter records and maintenance logs for normal operation, pre-fault, and fault states) as the training set, a deep learning model (e.g., an improved LSTM network) is constructed. The deep learning model predicts the trend of the equipment's operating data between t1 and t2 by learning the temporal characteristics of the data. The first operating data is input into the trained model to generate a continuous first comparison data graph (the horizontal axis represents time, and the vertical axis represents the predicted operating data value). The vertical axis point corresponding to the second time point t2 is defined as the "comparison point" (e.g., the model predicts the temperature at time t2 to be 90℃). A comparison interval is generated based on the comparison point; if the second operating data exceeds the interval range, an alarm signal is output. By analyzing the trend of the first comparative data chart, if it is found that the second operating data is within the range but continues to deviate from the comparison point (e.g., the predicted temperature is 90℃, but the actual value rises from 88℃ to 91℃), maintenance personnel can be prompted in advance to check the equipment's heat dissipation system to avoid the fault from escalating.

[0022] Specifically, deep learning models and historical data are used to determine the data variation patterns of the corresponding electrical equipment, and this is combined with the first operating data to generate a first comparative data graph. Then, by comparing the relatively continuous first comparative data graph with the second operating data, the difference between the predicted and actual data can be used to determine whether the electrical equipment should trigger an alarm. This allows staff to perform maintenance on multiple electrical devices based on available data when maintaining them.

[0023] Specifically, this invention integrates information such as electrical equipment parameter information and inspection data based on historical data to achieve integrated supervision of electrical equipment data integration, transmission, centralized display, monitoring and early warning, and fault diagnosis, so as to fully grasp the status information of key power equipment.

[0024] It should be noted that due to various complex factors such as environment and quantity, although the trend of each device's operating data over time is roughly certain and can be simulated by a deep learning model, it requires calibration with the first operating data in real time. That is, when the deep learning model outputs the first comparative data graph, it is generated through calibration and fine-tuning with the first operating data in real time. In contrast, electrical equipment, due to its lifespan and calendar lifespan, requires a large amount of historical data samples to generate the deep learning model used as the predictive model. This deep learning model can be based on existing technologies, such as language models (CHATGPT) based on artificial intelligence technology or other deep learning algorithms; existing technologies will not be discussed further.

[0025] Specifically, the lines or polylines of the first comparative data graph of the deep learning model can be derived from the first running data at the first time, using any similar artificial intelligence, neural network, or prediction algorithm; existing technologies will not be elaborated upon further. That is, using the first running data at the first time, the deep learning model can output a corresponding first comparative data graph and a corresponding line composed of time and running data. This line should be a continuous, curved line, or a relatively dense polyline graph.

[0026] Preferably, the first time is earlier than the second time. The first time can be a time from the previous week, and the second time can be the current time. The alarm signal indicates whether there is a fault in the electrical equipment at this time.

[0027] Specifically, electrical equipment includes switchgear, transformers, and uninterruptible power supplies (UPS), and operating data includes key parameters such as temperature, voltage, current, insulation resistance, and vibration frequency.

[0028] Furthermore, preprocessing the historical data to obtain preprocessed data includes the following steps: Historical data is filtered using one or a combination of wavelet transform, smoothing filter, low-pass filter, high-pass filter, and band-pass filter to obtain filtered data. The filtered data is denoised and normalized using time-series correlation and density clustering algorithms to obtain preprocessed data.

[0029] Specifically, to meet the needs of advanced applications such as monitoring and alarm, fault diagnosis, status assessment, and data mining, the collected data should undergo usefulness and validity screening preprocessing, and data filtering should be performed to remove obviously erroneous data. For different types of data, digital filtering methods such as wavelet transform, smoothing filtering, low-pass filtering, high-pass filtering, and band-pass filtering are selected to filter the data, and interpolation algorithms are used to supplement missing data. Data cleaning methods based on time-series correlation and density clustering algorithms are used to preprocess historical data in the database by denoising and normalizing, identifying outlier data in the historical data, solving for and filling in the missing parts of the outlier data, thus completing the cleaning of outlier data.

[0030] Optionally, a comparative data reference map can be drawn using coordinate points of multiple historical data of the same electrical equipment. The horizontal axis of this comparative data reference map is time, and the vertical axis is the operating data. It is not a line. Instead, a deep learning model is used to draw the above comparative data reference map into a more continuous line. During the drawing process, discontinuous points are deleted to generate a first comparative data map.

[0031] Optionally, historical data of the electrical equipment is acquired, including operating parameters, fault records, and maintenance records of the electrical equipment; the historical data is preprocessed, including data cleaning, data normalization, and data partitioning; a deep learning model is constructed, comprising an input layer, a hidden layer, and an output layer, wherein the input layer receives the preprocessed data, the hidden layer extracts data features, and the output layer outputs prediction results; the deep learning model is trained using the preprocessed data to obtain a trained deep learning model; and the trained deep learning model is used to predict the operating status of the electrical equipment to construct a first comparative data map.

[0032] Further, obtaining the comparison interval of the second running data based on the comparison point includes the following steps: Set the upper and lower distances corresponding to the ordinate of the comparison point, obtain historical data at the third time point, and obtain the third running data; Determine whether the second running data is the same as the historical data corresponding to the second time. If they are not the same, increase the upper distance and lower distance by a first ratio. If they are the same, determine whether the difference between the second running data and the third running data is within the data fluctuation range. If it exceeds the data fluctuation range, decrease the upper distance and lower distance by a first ratio. The comparison interval is obtained by subtracting the adjusted lower distance from the ordinate of the comparison point and adding the adjusted upper distance to the ordinate of the comparison point.

[0033] Specifically, by determining whether the historical data of the second time period is the same as or different from the second operational data, when the historical data of the second time period is different from the second operational data, it can be determined whether the point on the first comparison data map at this second time period is a point on the historical data used as a reference or a point predicted by the deep learning model. If it is a predicted point, the upper and lower distances are increased to broaden the coverage of the prediction algorithm and reduce the impact of prediction errors. When the historical data of the second time period is the same as the second operational data, it is necessary to determine whether the third operational data of the third time period is similar to the second operational data. That is, although the horizontal axis is the same, the vertical axis exceeds the data fluctuation range. It is possible that the deep learning algorithm removed the historical data of the third time period when generating the first comparison data map in order to generate a more continuous first comparison data map, which is why the above-mentioned dissimilarity problem occurred. Therefore, this problem point should be given special attention. The upper and lower distances of this second time period should be reduced to increase the alarm probability of this second time period through a smaller comparison interval, thereby making operators pay more attention to this point.

[0034] Specifically, if the number of alarm signals occurring in the second time period exceeds a third preset number, the historical data from the second time period can be removed to prevent the removed historical data from repeatedly affecting the data source of the deep learning model and to ensure the accuracy of the deep learning model. Preferably, the third preset number is 5 times.

[0035] Optionally, the first proportion is 1% to 30%, preferably, the first proportion is 10%.

[0036] Optionally, the data fluctuation range is 5% to 30% of the second running data, preferably 10% of the second running data.

[0037] Specifically, the database pre-stores a 5℃ upper distance and a 5℃ lower distance corresponding to the comparison point (90℃). If there is no record of time t2 in the historical data, the upper and lower distances are increased by a first proportion (10%), expanding the range to 84.5℃~95.5℃; if there is historical data at time t2 (e.g., 88℃), and the difference between the second running data (91.4℃) and the historical data at the third time exceeds 10% (i.e., they are not similar), the upper and lower distances are reduced by 10%, narrowing the range to 85.5℃~94.5℃.

[0038] Specifically, for time points in historical data where faults are likely to occur (such as afternoons during hot seasons), the monitoring sensitivity is improved by narrowing the comparison interval, anomalies (such as abnormal increases in winding temperature) are detected in advance, and preventive checks (such as cleaning heat sinks and tightening terminals) are triggered.

[0039] Further, obtaining the comparison interval of the second running data based on the comparison point includes the following steps: Set the upper and lower distances corresponding to the ordinate of the comparison point, obtain historical data at the third time point, and obtain the third running data; Determine whether the first running data is the same as the historical data corresponding to the first time. If they are not the same, increase the upper distance and lower distance by a first ratio. If they are the same, determine whether the difference between the first running data and the third running data is within the data fluctuation range. If it exceeds the data fluctuation range, decrease the upper distance and lower distance by a first ratio. The comparison interval is obtained by subtracting the adjusted lower distance from the ordinate of the comparison point and adding the adjusted upper distance to the ordinate of the comparison point.

[0040] Specifically, by determining whether the historical data and the first operational data are the same or different, when they are different, it can be determined whether the point on the first comparison data map at that time is a point on the historical data used as a reference or a point predicted by a deep learning model. If it is a predicted point, the upper and lower distances are increased to broaden the coverage of the prediction algorithm and reduce the impact of prediction errors. When the historical data and the first operational data are the same, it is necessary to determine whether the third operational data of the third historical time is similar to the first operational data. That is, although the horizontal axis is the same, the vertical axis exceeds the data fluctuation range. It is possible that the deep learning algorithm removed the historical data of the third time when generating the first comparison data map in order to generate a more continuous first comparison data map, which is why the dissimilarity problem occurred. This problem should be given special attention, and the upper and lower distances of the first time should be reduced to increase the alarm probability of the first time through a smaller comparison interval, thereby making operators pay more attention to this point.

[0041] Furthermore, the deep learning-based electrical equipment monitoring method also includes the following steps: S8. After outputting the alarm signal, acquire the electrical equipment that outputs the alarm signal, obtain the electrical equipment of the same batch as the electrical equipment that outputs the alarm signal, and test the electrical equipment of the same batch.

[0042] Furthermore, the testing of the same batch of electrical equipment includes the following steps: S81. After setting a first preset number of times and reducing the upper and lower distances of the same batch of electrical equipment by a second ratio, the same batch of electrical equipment jumps to step S1 and runs steps S1 to S7. If an alarm signal is output again, step S81 is repeated until the number of jumps exceeds the first preset number of times, and all the same batch of electrical equipment outputs an alarm signal.

[0043] Furthermore, a second preset number of times is set. When the number of alarms of the electrical equipment exceeds the second preset number of times within a unit time, the operating data corresponding to the second time in the first comparison data graph of the electrical equipment that outputs the alarm signal is configured as the fourth operating data, the average value of the fourth operating data and the second operating data is configured as the fifth operating data, and the fourth operating data in the first comparison data graph of the electrical equipment that outputs the alarm signal is replaced with the fifth operating data to generate a second comparison data graph.

[0044] When the winding temperature of a dry-type transformer (e.g., 91.4℃) exceeds the range (85.5℃~94.5℃), an alarm signal is output to locate the faulty module (e.g., the winding). Preventive measures are then initiated: Immediately stop the machine and check the insulation status of the module (e.g., partial discharge detection); Based on maintenance records of similar faults in historical data, repair solutions (such as replacing insulation materials) are automatically pushed.

[0045] Reduce the upper / lower distance of the transformers in the same batch by a second ratio (e.g., 5%) (e.g., adjust to 4.75°C), and re-execute the model building and monitoring (jump to the step of claim 1); if the number of jumps exceeds the first preset number (e.g., 3 times), all equipment in the same batch will alarm, and batch preventive maintenance will be initiated (e.g., replace aging components in the entire batch).

[0046] If the number of alarms exceeds the second preset number (e.g., 5 times) within 1 hour, the average of the predicted value (90℃) and the actual value (91.4℃) at the second time in the first comparison data chart (90.7℃) is used to replace the original historical data, the model is retrained, and a second comparison data chart is generated. This optimizes the model's adaptability to equipment aging trends and prevents the recurrence of similar faults in the long term.

[0047] Optionally, the second proportion is 1% to 50%, preferably 5%. The second proportion is less than the first proportion, which increases the probability of screening electrical equipment in the same batch.

[0048] A deep learning-based electrical equipment monitoring system, employing the aforementioned deep learning-based electrical equipment monitoring method, includes: a preprocessing module, a deep learning module, an acquisition module, an input module, and a judgment module; The preprocessing module is used to preprocess historical data to obtain preprocessed data; The deep learning module is used to construct a deep learning model based on the preprocessed data; The acquisition module is used to acquire first operating data of the electrical equipment at a first time and second operating data of the electrical equipment at a second time. The input module is used to input the first running data into the deep learning model to obtain a continuous first comparison data image; The judgment module is used to obtain the comparison point corresponding to the second time and the comparison interval of the second running data according to the first comparison data graph, compare the second running data with the comparison interval, and output an alarm signal if the second running data is not located within the comparison interval.

[0049] Furthermore, the deep learning-based electrical equipment monitoring system also includes a prevention module, which is used to automatically generate maintenance work orders based on alarm signals, sort the health status of electrical equipment in the same batch, prioritize the maintenance of electrical equipment with higher risks, and record the effectiveness of preventive measures.

[0050] A device comprising at least one processor and at least one memory; The at least one memory is used to store computer programs; The at least one processor is used to execute the computer program to implement the above-described method.

[0051] The apparatus according to embodiments of the present invention includes a processor that can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) or a program loaded from a storage portion into random access memory (RAM). The processor may, for example, include a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor may also include onboard memory for caching purposes. The processor may include a single processing unit or multiple processing units for performing different actions of the method flow according to embodiments of the present invention.

[0052] The RAM stores various programs and data required for the operation of the electronic device. The processor, ROM, and RAM are interconnected via a bus. The processor executes various operations of the method flow according to embodiments of the present invention by executing programs in the ROM and / or RAM. It should be noted that the programs may also be stored in one or more memories other than ROM and RAM. The processor may also execute various operations of the method flow according to embodiments of the present invention by executing programs stored in said one or more memories.

[0053] According to embodiments of the present invention, the electronic device may further include an input / output (I / O) interface, which is also connected to a bus. The system may also include one or more of the following components connected to the I / O interface: an input section including a keyboard, mouse, etc.; an output section including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section including a hard disk, etc.; and a communication section including a network interface card such as a LAN card, modem, etc. The communication section performs communication processing via a network such as the Internet. A drive is also connected to the I / O interface as needed. Removable media, such as magnetic disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as needed.

[0054] According to embodiments of the present invention, the method flow according to embodiments of the present invention can be implemented as a computer software program. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a processor, it performs the functions defined above in the system of the embodiments of the present invention. According to embodiments of the present invention, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0055] The present invention also provides a computer-readable storage medium for storing executable instructions that, when executed by a processor, cause the processor to implement the above-described method.

[0056] The computer-readable storage medium may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more executable instructions, which, when executed, implement the method according to the embodiments of the present invention.

[0057] According to embodiments of the present invention, the computer-readable storage medium may be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In the present invention, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0058] Example 1 Taking dry-type transformer fault diagnosis as an example, the first step is to distinguish the data differences between normal operation and various faults: (1) Normal operation Normal operation refers to the condition where the temperature of all components of a dry-type transformer is within the design range and the temperature changes smoothly without drastic fluctuations or temperature rises exceeding the limits. This state meets the requirements of rated capacity, rated voltage, and rated frequency, with stable electrical parameters and insulation resistance meeting the specifications.

[0059] Based on the finite element method, the temperature distribution of various parts inside the dry-type transformer during normal operation is shown in the figure, including the winding temperature, core temperature, and air temperature. By comparing the temperature values ​​of different parts, it can be seen that the heat distribution of the dry-type transformer is relatively uniform. The internal temperature changes relatively gradually over a certain period of time. Meanwhile, the temperature of the high-voltage winding is higher than that of the low-voltage winding, and the core also has relatively high heat due to the dense magnetic circuit.

[0060] When a dry-type transformer is in normal operation, its electrical parameters, including rated capacity, rated voltage, and rated power, are stable. At the same time, the insulation condition can be indirectly monitored and evaluated through test methods such as resistance testing. The insulation resistance of a dry-type transformer should meet the requirements during normal operation.

[0061] In the simulation method for normal operation, five load voltage scenarios were set: 1V, 0.95V, 0.98V, 1.02V, and 1.05V. The heat source values ​​were set to the values ​​of core loss and winding loss, and the temperature field under these scenarios was simulated. 36 temperature data points were obtained for each scenario, for a total of 180 temperature data points.

[0062] (2) Overheating of the iron core When the core of a dry-type transformer overheats, it will emit noticeable noise. The most obvious characteristic is that the internal temperature of the transformer rises due to core overheating. Normally, the operating temperature of a transformer should be within the specified range. If the core temperature exceeds the normal range, it will lead to a decline in transformer performance. At the same time, core overheating will generate a large amount of heat and gas. If this heat is not dissipated in time, it will increase the risk of safety accidents such as transformer fires and explosions. When overheating occurs inside the core, it will cause localized temperature increases. In this case, the temperature in a localized area of ​​the core will be significantly higher than the temperature of the surrounding area, and the temperature distribution will be uneven.

[0063] Overheating of the iron core will also cause the winding temperature to rise. The degree of temperature rise depends on the extent and location of the iron core overheating. When the overheating range of the iron core is large, the temperature rise of the winding will also be more significant. Iron core overheating will cause rapid temperature rise and fluctuations because the temperature rises very quickly when the iron core is overheated, resulting in large temperature fluctuations. When the iron core temperature exceeds the design value, the dry-type transformer will be in a dangerous operating state.

[0064] The method of applying a heat source to the core surface is used to simulate the overheating operation of the dry-type transformer core. Simulations were performed on the core sections of the three-phase transformer, obtaining 180 temperature data points.

[0065] (3) Partial discharge Partial discharge is a discharge phenomenon occurring in specific areas of an insulating medium, usually caused by defects in the insulating material, excessively high electric field strength, or other reasons. Partial discharge generates noise of specific frequencies and amplitudes, significantly increasing the noise level of a transformer. It also generates a large amount of heat, causing a rise in the local temperature of the transformer. Furthermore, partial discharge can form discharge channels inside the transformer, affecting its normal operation and increasing maintenance difficulty and costs. Additionally, the number and intensity of partial discharges tend to gradually increase over time.

[0066] During partial discharge, the discharge area experiences a localized temperature rise due to energy release. This temperature rise can be monitored and measured in real time using equipment such as infrared thermal imagers. Simultaneously, the transformer's temperature sensor can also detect temperature changes and transmit the data to a monitoring system for recording and analysis.

[0067] By analyzing the temperature data during partial discharge, the location and intensity of the discharge can be determined, and the insulation condition of the transformer can be assessed. At this time, due to the thermal resistance limitation of the insulation material, the temperature rises rapidly, and local temperature anomalies are significant.

[0068] In the simulation method for partial discharge, a segmented approach was used, dividing the high-voltage winding into four equal parts on a plane with a cutting surface, and applying an additional heat source at one-quarter of the winding. 180 data points were obtained by simulating the heat released during partial discharge in a dry-type transformer.

[0069] (4) Winding short circuit A winding short circuit refers to the phenomenon where a short circuit occurs between windings due to factors such as aging or damage to the insulation material, leading to a decrease in insulation performance and strength. A winding short circuit can cause excessive local current, generating a large amount of heat and causing a rapid increase in the transformer's internal temperature. It can also cause instantaneous changes in the transformer's output voltage and current, severely impacting electrical performance and potentially leading to serious malfunctions. Furthermore, it can cause drastic changes in current and magnetic field, generating noise and vibration, which is usually one of the most noticeable symptoms during operation. Under winding short-circuit conditions, the temperature data characterization and analysis inside a dry-type transformer exhibits the following characteristics: 1) Winding temperature rise: Due to the excessive current caused by a short circuit in the winding, the winding temperature will rise accordingly. The magnitude of the temperature rise is related to the degree of short circuit and the duration of the short circuit.

[0070] 2) Localized temperature anomalies: At the location of a short circuit, the winding temperature will be higher, leading to localized temperature anomalies. This may cause aging of the insulation material and a decrease in insulation performance.

[0071] 3) Uneven temperature distribution along the current direction: At the location of a short circuit, the temperature rises where the current flows, but the temperature distribution along the current direction is uneven. At the two ends of the short circuit, the temperature is lower.

[0072] For the case of winding short circuits, this invention mainly considers short circuits occurring in the high-voltage windings of dry-type transformers, and then performs temperature field distribution simulation. A cut surface is made inside the high- and low-voltage windings of the transformer. It is assumed that an inter-turn short circuit occurs in the winding region of the three-phase windings within the cut surface, and a heat source is applied to the short-circuited winding portion. Winding short circuits are simulated separately on the three-phase high-voltage windings.

[0073] Secondly, the data is preprocessed and analyzed: The feature set consists of three-phase temperature data, and the last column is a label set used to represent the fault type. For example... Figure 2 As shown, the label set uses label encoding: 1 represents normal operation, 2 represents core overheating, 3 represents partial discharge, and 4 represents winding short circuit, to facilitate data processing. 20 sets of data are randomly selected for each operating state as test data, for a total of 80 sets of data as the test set. The remaining 70 sets of data for each operating state, for a total of 280 sets of data, are used as the training set.

[0074] See Figure 2As shown, by analyzing the confusion matrix heatmaps of predicted and true values ​​under different operating conditions, the correlation between each feature variable can be analyzed. The heatmap is used for the confusion matrix, where the color of each cell represents the magnitude of the value at that position, and the color gradient shows the intensity of the data values. The closer the value is to 1, the stronger the positive correlation between the two predicted and true values, proving the accuracy of the prediction.

[0075] In dry-type transformer fault identification based on machine learning multi-classification algorithms, the data normalization method employed is Min-Max Normalization. This normalization method is widely used in data preprocessing in machine learning. For input data, the value range of each feature is first mapped to the interval between 0 and 1. Specifically, each feature value is subtracted from its minimum value, and then divided by the feature's value range (i.e., maximum value minus minimum value). The purpose of normalization is to unify the value ranges of different features to the same interval, thereby avoiding excessive influence of certain features on model training.

[0076] The historical data is preprocessed using max-min normalization, as shown in the following formula: in, For the preprocessed data, For historical data to be preprocessed, The maximum value of the historical data to be preprocessed. This represents the minimum value of the historical data to be preprocessed.

[0077] Mapping the value range of each feature to between 0 and 1 makes the value range of different features uniform, which is more conducive to model training.

[0078] Finally, a deep learning model is developed, and its accuracy is verified: Cross-validation is used to optimize generalization ability, and pipeline techniques are employed to preprocess the data and train the model. For example... Figure 3 As shown in the learning rate curve, the accuracy of both the training and validation sets is above 95%, proving the accuracy of the model.

[0079] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for monitoring electrical equipment based on deep learning, characterized in that, Includes the following steps: S1. Obtain historical data of electrical equipment and preprocess the historical data to obtain preprocessed data; S2. Construct a deep learning model based on the preprocessed data; S3. Obtain the first operating data of the electrical equipment at a first time and the second operating data of the electrical equipment at a second time; S4. Input the first running data into the deep learning model to obtain a continuous first comparison data graph; S5. Obtain the comparison point corresponding to the second time based on the first comparison data graph; S6. Obtain the comparison interval of the second running data based on the comparison point; S7. Compare the second operating data with the comparison interval. If the second operating data is not within the comparison interval, output an alarm signal. The comparison interval of the second running data is obtained based on the comparison point, including the following steps: Set the upper and lower distances corresponding to the ordinate of the comparison point, obtain historical data at the third time point, and obtain the third running data; Determine whether the second running data is the same as the historical data corresponding to the second time. If they are not the same, increase the upper distance and lower distance by a first ratio. If they are the same, determine whether the difference between the second running data and the third running data is within the data fluctuation range. If it exceeds the data fluctuation range, decrease the upper distance and lower distance by a first ratio. The comparison interval is obtained by subtracting the adjusted lower distance from the ordinate of the comparison point and adding the adjusted upper distance to the ordinate of the comparison point. Furthermore, obtaining the comparison interval of the second running data based on the comparison points includes the following steps: Set the upper and lower distances corresponding to the ordinate of the comparison point, obtain historical data at the third time point, and obtain the third running data; Determine whether the first running data is the same as the historical data corresponding to the first time. If they are not the same, increase the upper distance and lower distance by a first ratio. If they are the same, determine whether the difference between the first running data and the third running data is within the data fluctuation range. If it exceeds the data fluctuation range, decrease the upper distance and lower distance by a first ratio. The comparison interval is obtained by subtracting the adjusted lower distance from the ordinate of the comparison point and adding the adjusted upper distance to the ordinate of the comparison point.

2. The electrical equipment monitoring method based on deep learning according to claim 1, characterized in that, Preprocessing the historical data to obtain preprocessed data includes the following steps: Historical data is filtered using one or a combination of wavelet transform, smoothing filter, low-pass filter, high-pass filter, and band-pass filter to obtain filtered data. The filtered data is denoised and normalized using a time-series correlation and density clustering algorithm to obtain preprocessed data.

3. The electrical equipment monitoring method based on deep learning according to claim 1, characterized in that, It also includes the following steps: S8. After outputting the alarm signal, acquire the electrical equipment that outputs the alarm signal, obtain the electrical equipment of the same batch as the electrical equipment that outputs the alarm signal, and test the electrical equipment of the same batch.

4. The electrical equipment monitoring method based on deep learning according to claim 3, characterized in that, The testing of the same batch of electrical equipment includes the following steps: S81. After setting a first preset number of times and reducing the upper and lower distances of the same batch of electrical equipment by a second ratio, the same batch of electrical equipment jumps to step S1 and runs steps S1 to S7. If an alarm signal is output again, step S81 is repeated until the number of jumps exceeds the first preset number of times, and all the same batch of electrical equipment outputs an alarm signal.

5. The electrical equipment monitoring method based on deep learning according to claim 4, characterized in that, A second preset number of times is set. When the number of alarms of the electrical equipment exceeds the second preset number of times within a unit time, the operating data corresponding to the second time in the first comparison data graph of the electrical equipment that outputs the alarm signal is configured as the fourth operating data, the average value of the fourth operating data and the second operating data is configured as the fifth operating data, and the fourth operating data in the first comparison data graph of the electrical equipment that outputs the alarm signal is replaced with the fifth operating data to generate a second comparison data graph.

6. A deep learning-based electrical equipment monitoring system, employing the deep learning-based electrical equipment monitoring method as described in any one of claims 1 to 5, characterized in that, include: The module consists of a preprocessing module, a deep learning module, an acquisition module, an input module, and a judgment module. The preprocessing module is used to preprocess historical data to obtain preprocessed data; The deep learning module is used to construct a deep learning model based on the preprocessed data; The acquisition module is used to acquire first operating data of the electrical equipment at a first time and second operating data of the electrical equipment at a second time. The input module is used to input the first running data into the deep learning model to obtain a continuous first comparison data image; The judgment module is used to obtain the comparison point corresponding to the second time and the comparison interval of the second running data according to the first comparison data graph, compare the second running data with the comparison interval, and output an alarm signal if the second running data is not located within the comparison interval.

7. A device, characterized in that, Includes at least one processor and at least one memory; The at least one memory is used to store computer programs; The at least one processor is used to execute the computer program to implement the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, Used to store executable instructions, which, when executed by a processor, cause the processor to implement the method of any one of claims 1 to 5.

Images