Distribution network cable state evaluation method, system and terminal

Abstract

The invention relates to the technical field of distribution network cable evaluation, and particularly discloses a distribution network cable state evaluation method and system and a terminal, which adopt semantic segmentation to suppress background information in a current oscillogram, and then extract time domain correlation information among local current waveforms as an input basis of a cable state evaluator. And the state of the cable is intelligently identified from the dynamic change information of the current waveform. In this way, an intelligent scheme is adopted to replace common manual recognition, and a large number of manpower resources are saved.

Description

Technical Field

[0001] The present application relates to the technical field of distribution network cable evaluation, and more specifically, to a distribution network cable status evaluation method, system and terminal. Background Art

[0002] Power cables are cables used to transmit and distribute electrical energy. As an important part of the operation of the power system, their failure will inevitably cause significant losses to power companies and residents.

[0003] Power cable faults are usually divided into three stages: cable defects, incipient faults (early failures), and permanent failures. Specifically, cable defects caused by environmental stress, mechanical stress, etc., become incipient faults as partial discharge increases. Over time, incipient faults occur repeatedly and eventually develop into permanent faults.

[0004] In the prior art, incipient faults are usually detected by manually identifying overcurrent signals. However, interference signals such as transformer excitation inrush current and capacitor switching also show overcurrent characteristics, and the subtle differences between these signals make it difficult to manually identify incipient faults in power cables.

[0005] Therefore, a new solution is expected. Summary of the invention

[0006] The technical problem to be solved by the present application is to provide a distribution network cable status assessment method, system and terminal, which solves the problem of difficulty in manually identifying early faults of power cables in the prior art.

[0007] The technical problem to be solved by this application is achieved by adopting the following technical solutions: In a first aspect, the present application provides a distribution network cable status assessment method, comprising: Obtaining a current waveform diagram of the cable to be detected within a predetermined time, and obtaining a sequence of local current waveform diagrams after data preprocessing; Inputting the sequence of the current waveform local graphs into a semantic segmenter based on a conditional generative adversarial network to obtain a sequence of significant current waveform local graphs; Inputting the sequence of the significant current waveform local images into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector; A cable state evaluator based on a classifier is used to classify the current waveform time domain correlation feature vector to obtain a cable state evaluation result.

[0008] Furthermore, a current waveform diagram of the cable to be detected within a predetermined time is obtained, and a sequence of current waveform partial diagrams is obtained after data preprocessing, including: dividing the current waveform diagram at predetermined time intervals to obtain the sequence of current waveform partial diagrams.

[0009] Further, the semantic segmenter based on the conditional generative adversarial network includes a generator and a discriminator.

[0010] Further, the generator adopts a U-Net structure and is used to perform semantic segmentation on the input local map of the current waveform to obtain the significant local map of the current waveform.

[0011] Further, the generator includes an input layer, a convolutional layer, an attention distribution bridging layer, a skip connection layer, and an output layer.

[0012] Further, the attention distribution bridging layer is used for: processing the input feature map using a point convolutional filter to obtain an attention score vector; processing the attention score vector using an activation function to obtain an attention score feature vector; using each element value in the attention score feature vector as a weight value to weight each feature matrix along the channel dimension of the input feature map to obtain an attention distribution applied feature map; wherein, the attention distribution applied feature map is further used for transposed convolution and upsampling processing.

[0013] Further, the discriminator receives cable condition data as a conditional input, wherein the cable condition data includes at least one of voltage, load rate, and ambient temperature.

[0014] Further, inputting the sequence of the significant local maps of the current waveform into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector includes: inputting the sequence of the significant local maps of the current waveform into a Transformer model or an LSTM model to obtain the current waveform time-domain correlation feature vector.

[0015] In a second aspect, the present application provides a distribution network cable condition assessment system, including: a current waveform acquisition module, configured to acquire a current waveform map of a cable to be detected within a predetermined time, and obtain a sequence of local maps of the current waveform through data preprocessing; a semantic segmentation module, configured to input the sequence of the local maps of the current waveform into a semantic segmenter based on a conditional generative adversarial network to obtain a sequence of significant local maps of the current waveform; a time-domain correlation feature extraction module, configured to input the sequence of the significant local maps of the current waveform into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector; A cable status evaluation module, configured to classify the time-domain correlation feature vectors of the current waveforms by using a cable status evaluator based on a classifier, so as to obtain a cable status evaluation result.

[0016] In a third aspect, the present application provides a terminal, including: A processor, configured to be coupled with a memory, and read and execute instructions stored in the memory; When the processor runs, it executes the instructions, so that the processor is configured to execute the distribution network cable status evaluation method.

[0017] The present application includes at least one of the following beneficial technical effects: Preprocessing the overall current waveform diagram into a sequence of local diagrams, so that the subsequent semantic segmenter based on a conditional generative adversarial network can pay more attention to the details in the waveform diagram; Reducing the attention to irrelevant background waveform information through semantic segmentation; Extracting the time-domain correlation information between the local current waveforms as the input information of the cable status evaluator based on a classifier, so that the cable status evaluator can identify the cable status from the dynamic change information of the current waveforms. Description of the Drawings

[0018] Figure 1 It is a schematic flowchart of the distribution network cable status evaluation method provided by the present application.

[0019] Figure 2 It is a schematic structural diagram of the generator provided by the present application.

[0020] Figure 3 It is a schematic flowchart of the operation of the attention allocation bridging layer provided by the present application.

[0021] Figure 4 It is a schematic structural diagram of the distribution network cable status evaluation system provided by the present application.

[0022] Figure 5 It is a schematic structural diagram of the terminal provided by the present application.

[0023] In the figure: 100, distribution network cable status evaluation system; 110, current waveform acquisition module; 120, semantic segmentation module; 130, time-domain correlation feature extraction module; 140, cable status evaluation module; 121, generator; 1211, input layer; 1212, convolutional layer; 1213, attention allocation bridging layer; 1214, skip connection layer; 1215, output layer; 10, processor; 20, memory; 30, instructions. Detailed Embodiments

[0024] To clearly understand the technical means, creative features, achieved objectives and effects of this application, the following further elaborates this application in conjunction with specific illustrations.

[0025] In the operation monitoring of distribution network cables, current is one of the important parameters for evaluating cable status. When the current exceeds a preset threshold, it usually means that there may be abnormalities in the cable (and / or cable connection parts), such as: Overload: The current exceeds the rated value, which may cause the cable temperature to rise and accelerate insulation aging.

[0026] Short circuit: The current suddenly increases, which may be due to internal or external short circuit faults in the cable.

[0027] Ground fault: Abnormal current may indicate a decrease in the cable's insulation performance to the ground.

[0028] That is to say, in the process of traditional distribution network cable status evaluation and analysis, usually when the detected current reaches the preset threshold (for example, 1.5 times the rated current), data collection and subsequent status evaluation and analysis are triggered.

[0029] This method actively shields the signals below the threshold. The implicit assumption is that the signals below the threshold are all harmless normal fluctuations or noises and are not worthy of attention; only those exceeding the threshold are possible fault signals.

[0030] However, this assumption completely ignores the pollution and masking effect of environmental noise on the valuable weak fault features below the threshold.

[0031] Specifically, the overcurrent signals generated by initial cable faults (such as weak partial discharges) may have amplitudes far lower than the fixed threshold set to prevent permanent faults (such as short circuits). These valuable early warning signals are extremely weak in themselves, and when superimposed with environmental noise, they are easily submerged in the "normal" background fluctuations and completely ignored by the system, resulting in missed judgments.

[0032] To avoid such missed judgments, the most direct way is to adjust the preset threshold so that these missed judgment situations can be detected.

[0033] However, when strong environmental noise (such as the impact current at the moment of large motor startup) causes the instantaneous current value to exceed the fixed threshold, the system will wrongly trigger the status evaluation process. Analysts or subsequent algorithms need to spend time and resources to distinguish these "false overcurrents" caused by noise, resulting in misjudgments. This increases the ineffective workload and may even lead to unnecessary equipment inspections or outages.

[0034] For example, when encountering a severe thunderstorm, if lightning strikes the ground near a substation or the lightning protection wire, an extremely strong transient electromagnetic field will be generated. This electromagnetic field induces high-frequency and high-amplitude surge currents and voltages (traveling waves) on the nearby cable shielding layer, armor layer, and even phase conductors. The frequency range of these pulse noises is extremely wide and the energy is huge.

[0035] The current waveforms collected by the sensor will be severely contaminated by these sharp and fast pulse noises. Dense spikes and oscillations appear on the waveforms. That is, being deceived by strong environmental noises, a large number of invalid analyses are triggered, wasting resources.

[0036] To alleviate this dilemma, this application sets up a precondition "checkpoint" before obtaining current data and status assessment.

[0037] Specifically, by introducing external environmental data, such as temperature, weather, and equipment aging degree, and comprehensively analyzing the external environmental data and current data, it is judged whether cable status assessment is needed at this time to avoid wasting a large amount of resources.

[0038] Based on the above technical concept, the implementation steps are as follows: First, when the current data exceeds a fixed threshold, obtain external environmental data, including temperature, weather, and equipment aging degree; Next, calculate the credible assessment probability under the current environment; Finally, determine whether cable status assessment is needed according to the credible assessment probability to avoid "ineffective overtime".

[0039] Among them, the temperature data uses the cable surface temperature or the ambient real-time temperature; the weather data uses the weather data released in real time; the equipment aging degree depends on the service life of the cable equipment, the deviation of the cable equipment's last calibration, and / or the number of partial discharge history times.

[0040] Example scenario: When the 10kV cable triggers an overcurrent alarm, it is necessary to combine environmental data to judge whether to start the status assessment process.

[0041] Step 1. Data input: ① Cable surface temperature: The current cable surface temperature measured by the distributed temperature sensing (DTS) system is 62°C, and the cable surface temperature one hour ago was 58°C.

[0042] ② Weather data: Lightning warning: None; Humidity: 80%; Wind speed: Level 3.

[0043] ③ Equipment aging degree: Obtain the service life of 8 years through the operation and maintenance database; the deviation of the last calibration is +1.2%; the number of partial discharge history times is 2 times / year.

[0044] Step 2: Calculate the credible evaluation probability in the current environment: Adopt a weighted logistic regression model, with the input features and weights as follows: ① Temperature feature (weight 30%): Temperature rise rate = (Current temperature - Temperature 1 hour ago) / 1 = (62°C - 58°C) / 1 = 4°C / h; Temperature overrun ratio = (62°C - 70°C) / 70°C = -0.11 (not overrun); Among them, 70°C means that the normal range of the cable surface temperature is set to ≤70°C.

[0045] ② Weather feature (weight 25%): Humidity penalty factor = max[0, (80% - 60%) / 20%] = 1.0; Lightning risk = 0 (no lightning); Among them, the general formula for the humidity penalty factor is expressed as: Humidity penalty factor = max(0, Current humidity - Humidity threshold / Humidity tolerance range); The reference humidity threshold is 60%, which represents the upper safety humidity limit for cable monitoring.

[0046] Exceeding this humidity threshold may cause: The cable joints get damp, resulting in a decrease in insulation performance and an increase in sensor measurement errors (such as the accuracy of capacitive humidity sensors decreases under high temperature and high humidity).

[0047] Of course, this humidity threshold can be adjusted according to the cable type (for example, set to 65% for oil-immersed cables and 60% for XLPE cables). It should not be limited by this application in the specific implementation process.

[0048] In addition, the humidity tolerance range refers to the buffer interval that allows the humidity to exceed the threshold, which is used to quantify the risk level. For example: If the current humidity ≤ 60% → Penalty factor = 0 (no risk); If the current humidity = 80% → (80 - 60) / 20 = 1.0 (linear penalty); If the current humidity ≥ 80% → Penalty factor ≥ 1.0 (exponential growth option).

[0049] Here, the role of the humidity penalty factor is to quantify the negative impact of humidity on the evaluation credibility into a value in [0, +∞), which is used for subsequent probability calculation. For example: 0: The humidity is within the safe range, and the credibility is not deducted; 1.0: When the humidity exceeds the threshold by 20%, it fully occupies the tolerance range, significantly reducing the evaluation credibility. >1.0: After exceeding the tolerance range, the penalty accelerates (which can be achieved by replacing it with an exponential function).

[0050] Using the max function can ensure that the penalty factor does not become negative when the humidity is below the threshold, avoiding reverse compensation.

[0051] In addition, lightning risk = {3.0 if lightning distance ≤ 1km; 1.5 if 1km < lightning distance < 5km; 0.2 if 5km ≤ lightning distance < 15km; 0 if lightning distance ≥ 15km}.

[0052] When the lightning distance ≤ 1km, a very high risk (3.0) is directly assigned to suppress the positive contribution of other features; when the lightning distance ≥ 5km, it is considered safe (0.2); when the lightning distance ≥ 15km, it is considered to have no impact at all (0).

[0053] ③ Aging characteristics (weight 45%): Service life score = min(8 / 15, 1) = 0.53; Calibration deviation score = 1 - (1.2% / 2%) = 0.4; Partial discharge frequency score = 1 - (2 / 5) = 0.6; Aging comprehensive score = (0.53 + 0.4 + 0.6) / 3 = 0.51.

[0054] Then use a normalization function, such as the Sigmoid function, for calculation.

[0055] Among them, the input vector = [4.0, -0.11, 1.0, 0, 0.51]; The weight vector = [0.10, 0.20, 0.15, 0.10, 0.45] (after normalization); The result of the linear combination is z = 4.0×0.10 + (-0.11)×0.20 + 1.0×0.15 + 0×0.10 + 0.51×0.45 = 0.4 - 0.022 + 0.15 + 0 + 0.2295 = 0.7575; The final probability is P = 1 / (1 + e -0.7575 ) = 0.68 (i.e., 68% credibility).

[0056] Step 3. Determine whether cable status evaluation is required based on the credible evaluation probability: Assuming that the current confidence threshold is 0.65, the current credible assessment probability is 0.68>0.65, and the cable status assessment is performed.

[0057] It is worth mentioning that in the actual application process, a lightning correction mechanism is also introduced.

[0058] That is, in actual engineering, a secondary penalty is imposed on lightning risk: the final probability P final =P×lightning veto coefficient.

[0059] Wherein, lightning rejection coefficient = {0.25 if lightning distance ≤ 1km; 0.6 if 1km<lightning distance ≤ 3km; 1.0 Other}.

[0060] In this way, consideration of external environmental factors is added to the status assessment process to automate the "self-review" of preconditions, thereby saving a lot of invalid analysis, avoiding "ineffective overtime", and increasing the operating efficiency of the system.

[0061] In power systems, the detection of early cable faults usually relies on manual analysis of overcurrent signals. Although a large number of invalid overcurrent signals caused by weather conditions, temperature changes, and equipment aging are screened out by the introduction of environmental conditions, the remaining current signals entering the analysis and evaluation process still cannot avoid similar overcurrent signals caused by many normal operations or grid disturbances.

[0062] In other words, the workload of manual identification is still huge and complex.

[0063] In this regard, Figure 1 As shown, the present application provides a distribution network cable status assessment method, and its specific steps include: S1, obtaining a current waveform diagram of the cable to be detected within a predetermined time, and obtaining a sequence of current waveform local diagrams through data preprocessing; S2, inputting the sequence of current waveform local maps into a semantic segmenter based on a conditional generative adversarial network to obtain a sequence of significant current waveform local maps; S3, inputting the sequence of significant current waveform local images into the waveform time-domain correlation feature extraction model to obtain the current waveform time-domain correlation feature vector; S4. Use a classifier-based cable state evaluator to classify the time-domain correlation feature vector of the current waveform to obtain a cable state evaluation result.

[0064] In the above method, the overall current waveform diagram is preprocessed into a sequence of local diagrams, enabling the subsequent semantic segmenter based on the conditional generative adversarial network to pay more attention to the details in the waveform diagram; semantic segmentation is used to reduce the attention to irrelevant background waveform information; then, the time-domain correlation information between individual local current waveforms is extracted as the input information for the cable status evaluator based on the classifier, enabling the cable status evaluator to identify the cable status from the dynamic change information of the current waveform.

[0065] Specifically, in an embodiment of the present application, a Rogowski coil is used to collect the current signal of the cable to be detected, and the current waveform diagram is obtained through reduction processing by a digital integrator.

[0066] Among them, the non-contact Rogowski coil has the advantage of high-frequency response for accurately capturing nanosecond-level transient pulses. In addition, the coreless design of the Rogowski coil avoids core saturation and can still maintain linear output in large current impact scenarios such as motor startup. In practical applications, it also has high installation flexibility, can be wound around the laid cable, is suitable for the renovation of old lines, and saves deployment costs.

[0067] In another embodiment of the present application, a Hall sensor is used to collect the current waveform diagram to cover low-frequency application scenarios.

[0068] Here, the Hall sensor has low-frequency stability and is suitable for monitoring slow-changing cable faults.

[0069] Next, data preprocessing is performed on the current waveform diagram to obtain a sequence of local current waveform diagrams, thereby enabling the subsequent semantic segmenter to reduce the attention to irrelevant background information. In an embodiment of the present application, the current waveform diagram is divided at a predetermined time interval to obtain a sequence of local current waveform diagrams.

[0070] In another embodiment of the present application, after obtaining the current waveform diagram, wavelet transform (such as db4 wavelet) is used to remove noise and power frequency interference; then, the current waveform diagram is slid and intercepted at a predetermined time interval (such as 10 ms) to obtain a sequence of local current waveform diagrams.

[0071] Among them, each local current waveform diagram in the sequence of local current waveform diagrams has an overlap rate of 10%-75%. It should be understood that setting an overlap rate of 10%-75% can ensure the complete capture of transient features.

[0072] For example, an overlap rate of 75% can ensure that pulses with a width less than the preset time (such as ≤2.5 ms) appear completely in at least 1 window (such as when the window length is 10 ms) with 100%. An overlap of 50% enables adjacent windows to share data within the preset time (such as 5 ms), facilitating the subsequent model to capture dynamic features such as pulse repetition periods.

[0073] Although dividing the overall current waveform diagram into individual local current waveform diagrams can, to a certain extent, enable subsequent models to place more attention on analyzing the detailed information in the current waveform.

[0074] However, since the information of the initial cable fault manifested in the current waveform diagram may only occupy a very small time domain or amplitude range, directly inputting it into the waveform time-domain correlation feature extraction model is likely to misjudge interference signals (such as the decaying oscillation of inrush current and the transient impact of capacitor switching) as fault features.

[0075] Therefore, in the embodiments of the present application, semantic segmentation is performed on each local current waveform diagram in the sequence of local current waveform diagrams to separate fault-related features and noise information, thereby amplifying the proportion of weak fault features in each local current waveform diagram and highlighting the evolution law of fault features (such as the increasing trend of the frequency of discharge pulses).

[0076] Specifically, a semantic segmenter based on a conditional generative adversarial network is used to process each local current waveform diagram to obtain a sequence of significant local current waveform diagrams.

[0077] Among them, the semantic segmenter based on the conditional generative adversarial network includes a generator and a discriminator. The generator adopts a U-Net structure and is used to perform semantic segmentation on the input local current waveform diagram to obtain a significant local current waveform diagram. Among them, the U-Net structure can retain the multi-scale time-frequency features in the current waveform.

[0078] As Figure 2 shown, the generator 121 includes an input layer 1211, a convolutional layer 1212, an attention distribution bridging layer 1213, a skip connection layer 1214, and an output layer 1215.

[0079] In one embodiment, the first part of the generator includes an input layer and three convolutional layers; the second part of the generator includes three skip connection layers and an output layer; the first part and the second part of the generator are connected through an attention distribution bridging layer.

[0080] Among them, the convolutional layer includes a two-dimensional convolutional component, a batch normalization component, and a max pooling component. The skip connection layer includes an upsampling component, a skip fusion component, a transposed convolutional component, a convolutional component, and a batch normalization component.

[0081] That is, the shallow, middle, and deep features in the local current waveform diagram are extracted by multiple convolutional layers respectively, and the shallow, middle, and deep features are fused by multiple skip connection layers, thereby forming a comprehensive expression of the multi-scale time-frequency features of the current waveform.

[0082] Here, although the skip connections in the original U-Net framework help with the effective transmission of low-resolution information, they often lead to a blurring effect on the obtained image features. That is, when the low-resolution features and high-resolution features are simply concatenated or added in the skip connections, due to the resolution mismatch, direct fusion may cause conflicts, resulting in a blurring effect and making it difficult to distinguish the weak details in the local map of the current waveform in semantic segmentation.

[0083] In the technical solution of this application, by introducing an attention mechanism, that is, adding an attention distribution bridging layer to recalibrate the weights and selectively highlight the important regions in the features.

[0084] As Figure 3 shown, the specific implementation method of the attention distribution bridging layer includes: S10. Process the input feature map using a point convolution filter to obtain an attention score vector; S20. Process the attention score vector using an activation function, such as the sigmoid activation function, to obtain an attention score feature vector; S30. Use the respective element values in the attention score feature vector as weight values to weight each feature matrix along the channel dimension of the input feature map to obtain an attention distribution applied feature map; Among them, the attention distribution applied feature map is further used as the input to the skip connection layer for transposed convolution and upsampling processing.

[0085] Considering that the morphology of the cable current signal is affected by the actual working conditions. For example, under high load conditions, the current amplitude increases, which may mask the tiny discharge characteristics of incipient faults. When the ambient temperature rises, the cable insulation resistance decreases and the partial discharge pulse frequency changes.

[0086] In practical applications, if physical constraints are ignored, traditional models are prone to misjudging normal working condition changes as faults. In particular, during the inference process of a semantic segmenter based on a conditional generative adversarial network, cable working condition data is also input as conditional information. In this way, by associating the working condition data, the model can dynamically adjust the discrimination criteria.

[0087] Specifically, the cable working condition data includes at least one of voltage, load rate, and ambient temperature.

[0088] In the specific implementation process, cable condition data is obtained by voltage sensors, load sensors, and temperature sensors. After obtaining at least one of the time series of voltage values, the time series of load rates, or the time series of ambient temperature values using the sensors, a fully connected layer is used to transform them into a voltage feature vector, a load rate feature vector, or an ambient temperature feature vector. Subsequently, the voltage feature vector, the load rate feature vector, or the ambient temperature feature vector is fused with the local current waveform diagram data.

[0089] Optionally, the mean value of the voltage feature vector, the load rate feature vector, or the ambient temperature feature vector is added to or multiplied by each element value in the local current waveform diagram to achieve fusion, and the fused data is used as the input of the generator.

[0090] In addition, during the training process, the discriminator receives the significant local current waveform diagram generated by the generator, and reversely trains the generator based on the loss function value between it and the real significant local current waveform diagram. Among them, the real significant local current waveform diagram is obtained after being independently labeled by experts.

[0091] In an embodiment of the present application, the discriminator uses a binary cross-entropy loss function.

[0092] Since the discharge pulses of incipient faults often exhibit a non-steady time series pattern (such as gradually shortening intervals and slowly rising amplitudes), waveform analysis at a single moment cannot capture this trend.

[0093] In an embodiment of the present application, the sequence of significant local current waveform diagrams is input into a waveform time-domain correlation feature extraction model such as a Transformer model or an LSTM model to capture the dynamic correlation in time series of each significant local current waveform diagram, and a current waveform time-domain correlation feature vector is obtained.

[0094] Then, a cable condition evaluator is constructed using a classifier to classify the current waveform time-domain correlation feature vector, and a cable condition evaluation result is obtained. Among them, the cable condition evaluation result is a probabilistic output, which is used to determine whether there is an incipient fault in the cable to be detected.

[0095] For example, when the cable condition evaluation result is [Normal: 25%, Incipient fault: 75%], it may indicate that there is an incipient fault in the cable to be detected. In actual applications, a predetermined threshold is set to determine whether there is an incipient fault in the cable to be detected, and this threshold is selected by manual experience or obtained through model training.

[0096] As Figure 4 shown, the present application also provides a distribution network cable condition evaluation system 100, including: The current waveform acquisition module 110 is configured to obtain the current waveform diagram of the cable to be detected within a predetermined time, and obtain a sequence of local current waveform diagrams after data preprocessing; The semantic segmentation module 120 is configured to input the sequence of local current waveform diagrams into a semantic segmenter based on a conditional generative adversarial network to obtain a sequence of significant local current waveform diagrams; The time-domain correlation feature extraction module 130 is configured to input the sequence of significant local current waveform diagrams into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector; The cable state evaluation module 140 is configured to classify the current waveform time-domain correlation feature vector using a cable state evaluator based on a classifier to obtain a cable state evaluation result.

[0097] As Figure 5 shown, the present application also provides a terminal, which includes: A processor 10, configured to be coupled with a memory 20, and read and execute instructions 30 stored in the memory; When the processor 10 runs, it executes the instructions 30, so that the processor 10 is configured to execute the distribution network cable state evaluation method.

[0098] Figure 5 Only some components of the terminal are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively.

[0099] The memory 20 may be an internal storage unit of the terminal in some embodiments, such as the hard disk or memory of the terminal.

[0100] The memory 20 may also be an external storage device of the terminal in other embodiments, such as a plug-in hard disk equipped on the terminal, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc.

[0101] Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device.

[0102] The memory 20 is used to store application software installed on the terminal and various types of data, such as program codes installed on the terminal, etc.

[0103] The memory 20 may also be used to temporarily store data that has been output or will be output. In one embodiment, instructions 30 are stored on the memory 20, and the instructions 30 can be executed by the processor 10 to execute the distribution network cable state evaluation method.

[0104] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chips, which are used to run the instructions 30 stored in the memory 20 or process data, such as executing the method for evaluating the status of the distribution network cable, etc.

[0105] The foregoing has shown and described the basic principles, main features, and advantages of the present application. Those skilled in the art should understand that the present application is not limited by the above embodiments. Without departing from the spirit and scope of the present application, the present application will have various changes and improvements, and all these changes and improvements fall within the scope of the claims of the present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the status of distribution network cables, characterized in that, Including: Obtain the current waveform diagram of the cable to be detected within a predetermined time, and obtain a sequence of local current waveform diagrams through data preprocessing; Input the sequence of the local current waveform diagrams into a semantic segmenter based on a conditional generative adversarial network to obtain a sequence of significant local current waveform diagrams; Input the sequence of the significant local current waveform diagrams into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector; Use a cable status evaluator based on a classifier to classify the current waveform time-domain correlation feature vector to obtain a cable status evaluation result.

2. The power distribution cable status evaluation method according to claim 1, wherein Obtain the current waveform diagram of the cable to be detected within a predetermined time, and obtain a sequence of local current waveform diagrams through data preprocessing, including: Divide the current waveform diagram at a predetermined time interval to obtain the sequence of the local current waveform diagrams.

3. The distribution network cable status evaluation method according to claim 1, characterized in that The semantic segmenter based on the conditional generative adversarial network includes a generator and a discriminator.

4. The distribution network cable status evaluation method according to claim 3, characterized in that The generator adopts a U-Net structure and is used to perform semantic segmentation on the input local current waveform diagram to obtain the significant local current waveform diagram.

5. The distribution network cable status evaluation method according to claim 4, characterized in that, The generator includes an input layer, a convolutional layer, an attention allocation bridging layer, a skip connection layer, and an output layer.

6. The distribution network cable status evaluation method according to claim 5, characterized in that, The attention allocation bridging layer is used for: Process the input feature map using a point convolutional filter to obtain an attention score vector; Process the attention score vector using an activation function to obtain an attention score feature vector; Use each element value in the attention score feature vector as a weight value to weight each feature matrix along the channel dimension of the input feature map to obtain an attention allocation applied feature map; Wherein, the attention allocation applied feature map is further used for transposed convolution and upsampling processing.

7. The power distribution cable status evaluation method according to claim 6, characterized in that, The discriminator receives cable working condition data as a conditional input, wherein the cable working condition data includes at least one of voltage, load rate, and ambient temperature.

8. The distribution network cable status evaluation method according to claim 7, wherein Input the sequence of the significant local current waveform diagrams into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector, including: Input the sequence of the significant local current waveform diagrams into a Transformer model or an LSTM model to obtain the current waveform time-domain correlation feature vector.

9. A distribution network cable status evaluation system, characterized in that, Including: A current waveform acquisition module, configured to obtain the current waveform diagram of the cable to be detected within a predetermined time, and obtain a sequence of local current waveform diagrams through data preprocessing; A semantic segmentation module, configured to input the sequence of the local current waveform diagrams into a semantic segmenter based on a conditional generative adversarial network to obtain a sequence of significant local current waveform diagrams; A time-domain correlation feature extraction module, configured to input the sequence of the significant local current waveform diagrams into a waveform time-domain correlation feature extraction model to obtain a current waveform time-domain correlation feature vector; A cable status evaluation module, configured to use a cable status evaluator based on a classifier to classify the current waveform time-domain correlation feature vector to obtain a cable status evaluation result.

10. A terminal, characterized in that, Including: A processor, configured to be coupled with a memory, and read and execute instructions stored in the memory; When the processor runs, it executes the instructions, so that the processor is used to execute the power distribution cable status evaluation method according to any one of claims 1-8.

Images