Cable fault big data early warning system

Abstract

The invention belongs to the technical field of cable fault early warning, and particularly provides a cable fault big data early warning system, which mainly comprises an operation data acquisition module for acquiring operation data of a cable, the operation data including cable length, A-phase and C-phase conductor current, cable skin temperature and environment temperature; the characteristic factor generation module is used for generating an eddy current index and a sheath circulation correction factor according to the operation data of the cable; the insulation characteristic analysis module is used for processing an eddy current index and a sheath circulation correction factor by adopting a cable insulation characteristic analysis model; and the early warning result generation module is used for generating an early warning grade and a hidden danger position according to the partial discharge intensity and the air gap defect index. According to the method, accurate decoupling quantization of the insulation degradation degree and the hidden danger position can be realized, the early defect detection sensitivity and the anti-interference capability are improved, real-time positioning and graded early warning of hidden dangers in the whole life cycle of the cable are supported, and a multi-dimensional decision basis is provided for active protection of a power grid.

Description

Technical Field

[0001] The present invention belongs to the technical field of cable fault early warning, and particularly relates to a big data early warning system for cable faults. Background Art

[0002] During the operation of a cable, the electromagnetic-thermal coupling effect induces distortion of the conductor current distribution and microscopic defects in the insulation layer, forming irreversible deterioration accumulation. The accumulation of partial discharge energy and the distortion of the air gap electric field accelerate the insulation aging, significantly shortening the cable life and triggering hidden faults.

[0003] Traditional monitoring methods only rely on a single temperature or current threshold, which makes it difficult to effectively distinguish environmental interference from early cable defects in practical applications. In addition, for short cables, the thermal lag phenomenon is more obvious, while long cables are easily affected by noise. The superposition of the two is extremely likely to lead to misjudgment of the monitoring results. Existing cable early warning methods are difficult to adapt to parameter differences such as cable length and cross-sectional area, and cannot well adapt to these parameter changes. They often easily ignore the non-linear correlation between current phase fluctuations and temperature conduction, resulting in difficulty in accurately quantifying the insulation deterioration degree of the cable and determining the location of potential hazards, ultimately causing late alarms and vague positioning. Summary of the Invention

[0004] The present invention effectively solves the problems in the prior art that cable early warning methods are difficult to adapt to parameter differences such as cable length and cross-sectional area, cannot well adapt to these parameter changes, often easily ignore the non-linear correlation between current phase fluctuations and temperature conduction, resulting in difficulty in accurately quantifying the insulation deterioration degree of the cable and determining the location of potential hazards, realizes the accurate decoupling and quantification of the insulation deterioration degree and the location of potential hazards, improves the sensitivity of early defect detection and anti-interference ability, supports the real-time positioning and hierarchical early warning of potential hazards during the entire life cycle of the cable, and provides multi-dimensional decision-making basis for the active protection of the power grid.

[0005] To achieve the above object, the present invention adopts the following technical solutions: The present invention provides a big data early warning system for cable faults, including: an operation data acquisition module: acquiring the operation data of the cable, where the operation data includes cable length, conductor currents of phase A and phase C, cable surface temperature, and ambient temperature.

[0006] A characteristic factor generation module: generating an eddy current index and a sheath circulating current correction factor according to the operation data of the cable.

[0007] Insulation feature analysis module: It uses a cable insulation feature analysis model to process the eddy current index and the sheath circulating current correction factor. The cable insulation feature analysis model is based on the Wide&Deep architecture. On the Wide side, the eddy current index is input and connected to the cable length to generate linear features. On the Deep side, a fully connected network is used for processing, and the number of neurons is determined according to the cable length, and the partial discharge intensity and the air gap defect index are output.

[0008] Early warning result generation module: It generates the early warning level and the potential hazard location according to the partial discharge intensity and the air gap defect index.

[0009] Further, an eddy current index is generated according to the operating data of the cable, including: performing skin effect compensation on the conductor currents of phase A and phase C to respectively generate the compensated first current value and the second current value; taking a specified power frequency period as a window to calculate the sample entropy value of the phase difference between the first current value and the second current value; calculating the conductor eddy current index according to the first current value and the second current value.

[0010] Further, a sheath circulating current correction factor is generated according to the operating data of the cable, including: setting a cross-sectional threshold; obtaining the conductor cross-sectional area of the cable; when the conductor cross-sectional area is greater than the cross-sectional threshold, generating the sheath circulating current correction factor according to the conductor cross-sectional area and the sample entropy value.

[0011] Further, generating the sheath circulating current correction factor according to the conductor cross-sectional area and the sample entropy value includes: performing adaptive wavelet packet decomposition on the first current value and the second current value, and extracting the high-frequency band energy ratios of a specified number; setting an entropy threshold, if the sample entropy value is greater than the entropy threshold, activating the high-frequency energy strengthening mark; setting a basic correction factor, if the high-frequency strengthening mark is activated, using the product of the strengthening coefficient and the basic correction factor as the enhanced correction factor; setting a first period window and a second period window, if the high-frequency energy strengthening mark is activated, performing time-domain smoothing on the basic correction factor or the enhanced correction factor using the mean value of the first period window, otherwise performing time-domain smoothing on the basic correction factor or the enhanced correction factor using the mean value of the second period window; wherein, the mean value of the second period window is greater than the mean value of the first period window.

[0012] Further, generating the early warning level and the potential hazard location according to the partial discharge intensity and the air gap defect index includes: performing condition coding on the partial discharge intensity and the air gap defect index to generate coded data; generating the true temperature rise and the temperature rise acceleration characteristics according to the cable surface temperature; using a fault decision model to process the true temperature rise, the temperature rise acceleration characteristics and the coded data. The fault decision model is based on the gradient boosting tree architecture, and determines the splitting features according to the coded data to construct a decision tree, and outputs the early warning level and the potential hazard location.

[0013] Further, perform condition coding on the partial discharge intensity and the air gap defect index to generate coded data, including: obtaining the operation years of the cable, setting a first trigger threshold according to the operation years, and generating a first coding result when the partial discharge intensity is greater than the first trigger threshold; setting a second trigger threshold and a third trigger threshold, and generating a second coding result when the air gap defect index is greater than the second trigger threshold and the sample entropy value is greater than the third trigger threshold.

[0014] Further, generate the true temperature rise according to the cable surface temperature, including: obtaining the cable insulation layer thickness; performing thermal resistance compensation on the cable surface temperature according to the cable surface temperature, the ambient temperature, and the cable insulation layer thickness to generate the true temperature rise.

[0015] Further, generate the temperature rise acceleration feature, including: performing piecewise polynomial fitting on the true temperature rise to generate a baseline temperature rise curve; marking the difference between the true temperature rise and the baseline temperature rise curve as a dynamic residual sequence; extracting the extreme point distribution density of the dynamic residual sequence; setting a density threshold; when the extreme point distribution density is greater than the density threshold: performing empirical mode decomposition on the dynamic residual sequence, and extracting the energy proportion of the first specified order intrinsic mode components; calculating the permutation entropy value of the dynamic residual sequence, setting an entropy threshold, and activating an abnormal transient mark when the permutation entropy value is greater than the entropy threshold; fusing the energy proportion and the permutation entropy value according to the abnormal transient mark to generate the temperature rise acceleration feature.

[0016] Further, fusing the energy proportion and the permutation entropy value according to the abnormal transient mark to generate the temperature rise acceleration feature, including: when the abnormal transient mark is activated, performing a product operation on the energy proportion of the specified order intrinsic mode components and the permutation entropy value to generate the temperature rise acceleration feature, otherwise taking the cumulative sum of the energy proportion as the temperature rise acceleration feature.

[0017] Further, determine the splitting feature according to the coded data to construct a decision tree, including: when the first coding result is generated, constructing a decision tree with the temperature rise acceleration feature as the preferred splitting feature; when the second coding result is generated, using the product of the true temperature rise and the temperature rise acceleration feature as the splitting basis.

[0018] Advantages of the present invention: The present invention constructs an insulation feature analysis model based on the Wide&Deep architecture. On the Wide side, the eddy current index and the cable length are fused, and on the Deep side, the full connection network processes the sheath circulating current correction factor to dynamically adapt to the cable parameters, effectively solving the problem that the cable warning methods in the prior art are difficult to adapt to the parameter differences such as the cable length and cross-sectional area, cannot well adapt to these parameter changes, and often easily ignore the non-linear correlation between the current phase fluctuation and the temperature conduction, resulting in difficulty in accurately quantifying the insulation deterioration degree of the cable and determining the hidden danger location. The present invention realizes the accurate decoupling and quantification of the insulation deterioration degree and the hidden danger location, improves the sensitivity of early defect detection and the anti-interference ability, supports the real-time location and hierarchical warning of hidden dangers during the whole life cycle of the cable, and provides a multi-dimensional decision-making basis for the active protection of the power grid.

[0019] Other features and advantages of the present invention will be described in the following description, and in part, will be obvious from the description, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structure pointed out in the description and the drawings. Brief Description of the Drawings

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 It shows a schematic diagram of a cable fault big data warning system of the present invention. Detailed Embodiments

[0022] In order to solve the problems proposed in the background technology, the present invention constructs an insulation feature analysis model based on the Wide&Deep architecture. On the Wide side, the eddy current index and the cable length are fused, and on the Deep side, the full connection network processes the sheath circulating current correction factor to dynamically adapt to the cable parameters, realizes the accurate decoupling and quantification of the insulation deterioration degree and the hidden danger location, improves the sensitivity of early defect detection and the anti-interference ability, supports the real-time location and hierarchical warning of hidden dangers during the whole life cycle of the cable, and provides a multi-dimensional decision-making basis for the active protection of the power grid.

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the protection scope of the present invention.

[0024] In some embodiments, as Figure 1 shown, the present invention provides a big data early warning system for cable faults, including: an operation data acquisition module, a characteristic factor generation module, an insulation characteristic analysis module, and an early warning result generation module.

[0025] The method for the operation of the big data early warning system for cable faults includes: the method for the operation of the operation data acquisition module, the method for the operation of the characteristic factor generation module, the method for the operation of the insulation characteristic analysis module, and the method for the operation of the early warning result generation module.

[0026] The method for the operation of the operation data acquisition module is: S100. Acquire the operation data of the cable, where the operation data includes the cable length, the conductor currents of phase A and phase C, and the cable surface temperature and the ambient temperature.

[0027] The method for the operation of the characteristic factor generation module is: S200. Generate an eddy current index and a sheath circulating current correction factor according to the operation data of the cable.

[0028] The method for the operation of the insulation characteristic analysis module is: S300. Process the eddy current index and the sheath circulating current correction factor using a cable insulation characteristic analysis model. The cable insulation characteristic analysis model is based on the Wide&Deep architecture. On the Wide side, the eddy current index is input and connected to the cable length to generate linear features. On the Deep side, a fully connected network is used for processing, and the number of neurons is determined according to the cable length, and the partial discharge intensity and the air gap defect index are output.

[0029] The method for the operation of the early warning result generation module is: S400. Generate an early warning level and a potential hazard location according to the partial discharge intensity and the air gap defect index.

[0030] In some embodiments, generating an eddy current index according to the operation data of the cable includes: Sa210. Perform skin effect compensation on the conductor currents of phase A and phase C, and generate a compensated first current value and a compensated second current value respectively.

[0031] Skin effect exists in the cable conductor under high-frequency current, resulting in an underestimation of the actual effective values of the currents of phase A and phase C. A frequency-domain compensation algorithm can be used to correct the currents of phase A and phase C.

[0032] Original phase A current and the current of phase C , and the sampling frequency can be 10 kHz.

[0033] Calculate the skin depth based on the conductor material and cross-sectional shape . For example, the formula for calculating the skin depth of a copper conductor at 50 Hz power frequency is: Among them, represents the resistivity of the conductor, represents the angular frequency , represents the permeability.

[0034] Perform frequency-domain compensation on the currents of phase A and phase C conductors to generate the compensated first current value and second current value. The compensation formula can be: ; among them, and represent the compensated first current value and second current value respectively, and d represents the conductor diameter of the cable.

[0035] Sa220. Taking the specified power-frequency period as the window, calculate the sample entropy value of the phase difference between the first current value and the second current value.

[0036] The sample entropy value can quantify the phase fluctuation characteristics of the currents of phase A and phase C.

[0037] Taking 20 ms as the time window, extract the compensated first current value and the second current value of the phase difference sequence .

[0038] Calculate the sample entropy value of the phase difference sequence , which is used to characterize the complexity of the phase difference. The specific method is: Set the embedding dimension m = 2 and the tolerance threshold , among them, represents calculating the standard deviation of the phase difference sequence .

[0039] Statistically analyze the vector matching probabilities of lengths m and m + 1 in the phase difference sequence , and the formula is: , among them, represents the sample entropy value, respectively represent the number of matches within the tolerance threshold r of the vectors of lengths m and m + 1 in the phase difference sequence.

[0040] Sa230. Calculate the conductor eddy current index based on the first current value and the second current value.

[0041] Calculate the first current value and the second current value Ratio of root mean square values: ; where and respectively represent the root mean square of the first current value and the second current value .

[0042] According to the sample entropy value and the current ratio R, generate an eddy current index through linear weighting: ; where and respectively represent the empirical weights of the current ratio R and the sample entropy value , such as 0.7 and 0.3 respectively, and can be determined by regression of historical fault data.

[0043] In some embodiments, generating a sheath circulating current correction factor according to the operating data of the cable includes: Sb210. Set a cross-sectional threshold; obtain the conductor cross-sectional area of the cable.

[0044] Set the cross-sectional threshold according to the cable type (such as cross-linked polyethylene cable) and historical operating data. For example, for medium-voltage cables, the cross-sectional threshold is set to 300 mm². The conductor cross-sectional area can be obtained through cable design parameters or real-time measurement.

[0045] Sb220. When the conductor cross-sectional area is greater than the cross-sectional threshold, generate a sheath circulating current correction factor according to the conductor cross-sectional area and the sample entropy value.

[0046] In some embodiments, generating a sheath circulating current correction factor according to the conductor cross-sectional area and the sample entropy value includes: Sb221. Perform adaptive wavelet packet decomposition on the first current value and the second current value, and extract the energy ratio of a specified number of high-frequency bands.

[0047] The wavelet decomposition can use the db4 wavelet basis function.

[0048] The energy ratio of the high-frequency band (6 - 8 sub-bands) after the 3rd layer of decomposition can be extracted , and the calculation formula is: , where represents the energy value of the k-th sub-band.

[0049] Sb222. Set an entropy threshold. If the sample entropy value is greater than the entropy threshold, activate the high-frequency energy enhancement flag.

[0050] Collect the sample entropy value of the phase difference during normal operation of the device, calculate its mean and standard deviation, and the entropy threshold can be set as the mean plus several times the standard deviation. For example, the entropy threshold can be set .

[0051] Sb223. Set the basic correction factor. If the high-frequency enhancement mark is activated, use the product of the enhancement coefficient and the basic correction factor as the enhanced correction factor.

[0052] Set the basic correction factor The calculation formula is: ; where S represents the conductor cross-sectional area, and the coefficient 0.5 represents the empirical weight.

[0053] If the high-frequency energy enhancement mark is activated, then use the enhancement coefficient to generate the enhanced correction factor , and the enhancement coefficient can be obtained by fitting, theoretical derivation and other methods.

[0054] Sb224. Set the first-period window and the second-period window. If the high-frequency energy enhancement mark is activated, perform time-domain smoothing on the basic correction factor or the enhanced correction factor using the mean value of the first-period window; otherwise, perform time-domain smoothing on the basic correction factor or the enhanced correction factor using the mean value of the second-period window; where the mean value of the second-period window is greater than the mean value of the first-period window.

[0055] Set the first-period window as a short window, such as 10 minutes, and set the second-period window as a long window, such as 60 minutes. When the mark is activated, perform moving average smoothing on the basic correction factor or the enhanced correction factor using the moving average value of the first-period window. When the mark is not activated, perform moving average smoothing using the moving average value of the second-period window.

[0056] The short window can quickly respond to high-frequency abnormal fluctuations, while the long window ensures stability under steady-state conditions.

[0057] By accurately separating high-frequency interference components through adaptive wavelet packet decomposition and dynamically adjusting the correction factor in combination with the sample entropy value, the problem of misjudgment of sheath circulating current caused by changes in conductor cross-sectional area and high-frequency noise can be solved.

[0058] For example, when a large cross-section cable encounters arc discharge, the abnormal characteristics of the sheath circulating current are amplified in time through the enhanced correction factor, improving the accuracy of hidden danger location.

[0059] In some embodiments, S300 is specifically implemented as follows: The input on the Wide side is the eddy current index and the cable length L.

[0060] Perform a linear combination of the eddy current index and the cable length to generate a composite feature . The formula is: ; where and represent weight coefficients, which can be obtained through gradient descent. Represents the bias term.

[0061] The input on the Deep side is the sheath circulating current correction factor, which dynamically adjusts the number of neurons in the fully connected network structure according to the cable length L.

[0062] For example, when occurs, a 3-layer fully connected network is adopted, and the number of neurons is 16, 16, 8; when occurs, it is extended to a 4-layer fully connected network, and the number of neurons is 32, 32, 16, 8.

[0063] The ReLU activation function is used in the hidden layer, and the Sigmoid activation function is used in the output layer.

[0064] The output result of the cable insulation characteristic analysis model is the partial discharge intensity and the air gap defect index . The range of the partial discharge intensity can be 0 - 100 pC. The larger the value, the more serious the insulation deterioration. The range of the air gap defect index can be 0 - 1. The closer the value is to 1, the higher the risk of air gap defects.

[0065] When training the cable insulation characteristic analysis model, the mean squared error function is used for the partial discharge intensity, and the binary cross-entropy is used for the air gap defect index. The total loss is the weighted sum of the partial discharge intensity and the air gap defect index.

[0066] For example: ; where represents the total loss function, represents the mean squared error loss function, represents the binary cross-entropy loss function. 0.7 and 0.3 represent the weights of the mean squared error loss function and the binary cross-entropy loss function respectively, and can be determined by the method of grid search combined with validation set evaluation.

[0067] The Adam optimizer can be selected, with an initial learning rate of 0.001, decaying by 10% every 50 epochs, which can effectively adjust the model parameters and avoid too large step sizes in the later stage; the batch size is set to 64, which can balance the calculation efficiency and memory occupancy and make the training efficient and stable; it is planned to train for 300 epochs, and terminate when the validation set loss does not decrease for 10 consecutive epochs, which can limit the number of learning times and prevent overfitting.

[0068] In some embodiments, the warning level and potential hazard location are generated according to the partial discharge intensity and the air gap defect index, including: S410. Performing condition coding on the partial discharge intensity and the air gap defect index to generate coded data; generating the true temperature rise and temperature rise acceleration characteristics according to the cable skin temperature.

[0069] S420. A fault decision model is used to process the actual temperature rise, temperature rise acceleration characteristics and coded data. The fault decision model is based on the gradient boosting tree, and a decision tree is constructed based on the split characteristics determined according to the coded data to output the warning level and hidden danger location.

[0070] In some embodiments, the partial discharge intensity and the air gap defect index are coded for working conditions to generate coded data, including: S411. Obtain the number of years of operation of the cable, set a first trigger threshold according to the number of years of operation, and generate a first coding result when the local discharge intensity is greater than the first trigger threshold.

[0071] Set the first trigger threshold , for example, when the number of years in operation The first trigger threshold .

[0072] If the partial discharge intensity ,but , triggering the first encoding result , as the first encoding result ,otherwise .

[0073] S412. Set a second trigger threshold and a third trigger threshold, and generate a second encoding result when the air gap defect index is greater than the second trigger threshold and the sample entropy value is greater than the third trigger threshold.

[0074] A second trigger threshold can be set , the third trigger threshold .

[0075] When the air gap defect index And the sample entropy , generate the second encoding result ,like ,otherwise .

[0076] In some embodiments, generating the real temperature rise according to the cable skin temperature includes: Sa413. Get the cable insulation thickness.

[0077] The cable insulation thickness H can be obtained through cable design parameters or real-time measurement.

[0078] Sa414. According to the cable skin temperature, ambient temperature and cable insulation thickness, the cable skin temperature is compensated for thermal resistance to generate the real temperature rise.

[0079] Calculate the original temperature difference , ;in, Represents the cable skin temperature represents the ambient temperature.

[0080] According to the insulation layer thickness H, perform thermal resistance compensation on the original temperature difference to generate the true temperature rise : , for every 10 mm increase in the insulation layer thickness, the true temperature rise compensation coefficient doubles to correct the blocking effect of the heat conduction path.

[0081] Through the thermal resistance compensation formula, taking the insulation layer thickness as a key parameter, the true temperature rise is closer to the actual heating state of the conductor.

[0082] In some embodiments, generate the temperature rise acceleration feature, including: Sb413. Perform piecewise polynomial fitting on the true temperature rise to generate a baseline temperature rise curve.

[0083] The piecewise rule can be a 10-minute time window, and quadratic polynomial fitting is used within each window.

[0084] For example, the temperature rise data of a certain cable from 10:00 to 10:10 is: 50°C, 52°C, 55°C, 58°C, 60°C, then the fitting curve is: ; where t represents the time variable.

[0085] Piecewise polynomial fitting can eliminate baseline drift, and the dynamic residual sequence can highlight transient fluctuations.

[0086] Sb414. Mark the difference between the true temperature rise and the baseline temperature rise curve as the dynamic residual sequence; extract the extreme point distribution density of the dynamic residual sequence.

[0087] Calculate the dynamic residual sequence ; respectively represent the true temperature rise changing with time t and the fitting curve

[0088] Extract the extreme point distribution density of the dynamic residual sequence , that is, the number of maximum / minimum values per unit time.

[0089] Exemplarily, if 8 extreme points are detected within 5 minutes, then the extreme point distribution density per minute.

[0090] The dual criteria of extreme point density and permutation entropy can accurately identify abnormal events. For example, per minute and when, it is determined as arc discharge interference.

[0091] Sb415. Set the density threshold; when the distribution density of extreme points is greater than the density threshold: perform empirical mode decomposition on the dynamic residual sequence, and extract the energy proportion of the specified order of the first few intrinsic mode components; calculate the permutation entropy value of the dynamic residual sequence, set the entropy threshold, and activate the abnormal transient mark when the permutation entropy value is greater than the entropy threshold.

[0092] Set the density threshold per minute, when is the case, perform the following operations: Decompose the dynamic residual sequence into multiple intrinsic mode components, and extract the specified order of the first few components, such as the 3rd order.

[0093] If it is the 3rd order, calculate the energy proportion of IMF1 - IMF3 , , .

[0094] Set the embedding dimension m = 3, calculate the permutation entropy value PE of the dynamic residual sequence , set the entropy threshold , if , then activate the abnormal transient mark, such as , otherwise .

[0095] Sb416. Generate the temperature rise acceleration feature by fusing the energy proportion and the permutation entropy value according to the abnormal transient mark.

[0096] In some embodiments, generating the temperature rise acceleration feature by fusing the energy proportion and the permutation entropy value according to the abnormal transient mark includes: When the abnormal transient mark is activated, select the energy proportion of the specified order of the intrinsic mode components and the permutation entropy value for multiplication operation to generate the temperature rise acceleration feature, otherwise use the accumulated sum of the energy proportions as the temperature rise acceleration feature.

[0097] Exemplarily, when , select the product of the energy proportion of the 2nd order intrinsic mode component and the permutation entropy value PE as the temperature rise acceleration feature . When , accumulate the energy proportions of IMF1 - IMF3 as the temperature rise acceleration feature.

[0098] The fusion of the energy proportion and the entropy value can quantify the severity of the temperature rise change.

[0099] In some embodiments, determining the split feature according to the encoded data to construct a decision tree includes: When generating the first encoded result, construct a decision tree with the temperature rise acceleration feature as the preferred split feature.

[0100] Exemplarily, if the temperature rise acceleration feature , it is determined as a high-risk branch, and after splitting, it enters the sub-nodes of warning level 3 or 4.

[0101] The first coding result focuses on transient temperature rise fluctuations and adapts to the insulation deterioration mode dominated by partial discharge.

[0102] When generating the second coding result, the product of the actual temperature rise and the temperature rise acceleration feature is used as the splitting basis.

[0103] Exemplarily, if , it is determined as a compound fault branch, and after splitting, it enters the sub-node where the hidden danger location is more than 20% away from the starting end.

[0104] The second coding result combines the steady-state temperature rise and transient fluctuations and adapts to the coupled fault of air gap defects and circulation disturbances.

[0105] The output warning level can be the risk levels of 1-4 manually marked, and the hidden danger location is marked as the percentage from the starting end of the cable. For example, 25% means at a quarter of the total length.

[0106] The training process is carried out in batches. Each batch inputs 64 groups of data including actual temperature rise, temperature rise acceleration, and coding data, and outputs the warning level and the hidden danger location.

[0107] In terms of loss calculation, the classification loss is for the warning level. For example, the error weight of the high-risk level (levels 3-4) is increased by 3 times to let the model give priority to learning serious faults; the regression loss is for the hidden danger location, and the absolute difference between the prediction and the actual location is calculated.

[0108] Through multiple experiments, compare the comprehensive performance indicators of the model on the validation set for warning level prediction and hidden danger location prediction under different weight ratios, and select the ratio with the optimal comprehensive performance. For example, the total loss is optimized after weighted summation with the classification loss accounting for 70% and the regression loss accounting for 30%.

[0109] The dynamic parameter update uses the Adam optimizer, with an initial learning rate of 0.001 and a 10% decay every 50 rounds. Set the early stopping mechanism. If the validation set loss does not decrease for 10 consecutive rounds, stop training to prevent overfitting.

[0110] Through the dynamic splitting rule and joint loss optimization, the fault decision-making model can quickly adapt to different fault modes, accurately output the risk level and the hidden danger location, and guide the maintenance personnel to efficiently conduct inspections.

[0111] In some embodiments, the present invention provides a device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of a method for operating a cable fault big data warning system when executing the computer program.

[0112] In some embodiments, the present invention provides a readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the steps of a method for operating a cable fault big data warning system.

[0113] Among them, any reference to a memory, storage, database or other medium used in the embodiments provided by the present invention may include non-volatile and / or volatile memories. The non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory may include random access memory (RAM) or an external cache memory.

[0114] It should be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0115] Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cable fault big data early warning system, characterized in that: include: Operation data acquisition module: acquires operation data of the cable, the operation data including cable length, A-phase and C-phase conductor currents, cable skin temperature and ambient temperature; Characteristic factor generation module: generates eddy current index and sheath circulation correction factor according to the operation data of the cable; Insulation feature analysis module: The cable insulation feature analysis model is used to process the eddy current index and sheath circulation correction factor. The cable insulation feature analysis model is based on the Wide & Deep architecture. The eddy current index is input on the Wide side and connected to the cable length to generate linear features. A fully connected network is used on the Deep side to process the data. The number of neurons is determined according to the cable length, and the local discharge intensity and air gap defect index are output. Warning result generation module: Generates warning level and hidden danger location based on partial discharge intensity and air gap defect index.

2. The cable fault big data early warning system according to claim 1 is characterized in that: Generate eddy current index based on the cable's operating data, including: Perform skin effect compensation on the A-phase and C-phase conductor currents to generate compensated first current values ​​and second current values, respectively; Taking the specified power frequency period as a window, calculating the sample entropy value of the phase difference between the first current value and the second current value; The conductor eddy current index is calculated according to the first current value and the second current value.

3. The cable fault big data early warning system according to claim 2 is characterized in that: Generate sheath circulation correction factors based on cable operating data, including: Set the cross-section threshold; obtain the conductor cross-sectional area of ​​the cable; When the conductor cross-sectional area is greater than the cross-sectional threshold, a sheath circulating current correction factor is generated according to the conductor cross-sectional area and the sample entropy value.

4. The cable fault big data early warning system according to claim 1 is characterized in that: Generate sheath circulating current correction factor based on conductor cross-sectional area and sample entropy value, including: Performing adaptive wavelet packet decomposition on the first current value and the second current value, and extracting a specified number of high frequency band energy proportions; Set the entropy threshold. If the sample entropy value is greater than the entropy threshold, the high-frequency energy enhancement flag is activated. Set the basic correction factor. If the high-frequency enhancement flag is activated, the product of the enhancement coefficient and the basic correction factor is used as the enhancement correction factor. The first cycle window and the second cycle window are set. If the high-frequency energy enhancement mark is activated, the first cycle window mean is used to perform time domain smoothing on the basic correction factor or the enhanced correction factor. Otherwise, the second cycle window mean is used to perform time domain smoothing on the basic correction factor or the enhanced correction factor; wherein the second cycle window mean is greater than the first cycle window mean.

5. The cable fault big data early warning system according to claim 1 is characterized in that: Generate warning levels and hidden danger locations based on partial discharge intensity and air gap defect index, including: The partial discharge intensity and air gap defect index are coded under working conditions to generate coded data; the real temperature rise and temperature rise acceleration characteristics are generated according to the cable skin temperature; A fault decision model is used to process the real temperature rise, temperature rise acceleration characteristics and coded data. The fault decision model is based on the gradient boosting tree, and constructs a decision tree based on the split features determined according to the coded data to output the warning level and hidden danger location.

6. The cable fault big data early warning system according to claim 4 is characterized in that: The partial discharge intensity and air gap defect index are coded according to the working conditions to generate coded data, including: The operation years of the cable are obtained, a first trigger threshold is set according to the operation years, and a first coding result is generated when the local discharge intensity is greater than the first trigger threshold; A second trigger threshold and a third trigger threshold are set, and when the air gap defect index is greater than the second trigger threshold and the sample entropy value is greater than the third trigger threshold, a second encoding result is generated.

7. The cable fault big data early warning system according to claim 4 is characterized in that: Generates true temperature rise based on cable skin temperature, including: Get the thickness of cable insulation layer; According to the cable skin temperature, ambient temperature and cable insulation thickness, the cable skin temperature is compensated for thermal resistance to generate the real temperature rise.

8. The cable fault big data early warning system according to claim 6, characterized in that: Generates temperature rise acceleration characteristics, including: Perform piecewise polynomial fitting on the actual temperature rise to generate a baseline temperature rise curve; Mark the difference between the actual temperature rise and the baseline temperature rise curve as a dynamic residual sequence; extract the distribution density of the extreme points of the dynamic residual sequence; Set the density threshold; when the extreme point distribution density is greater than the density threshold: Perform empirical mode decomposition on the dynamic residual sequence to extract the energy proportion of the intrinsic mode component of the specified order; calculate the permutation entropy value of the dynamic residual sequence, set the entropy threshold, and activate the abnormal transient mark when the permutation entropy value is greater than the entropy threshold; The temperature rise acceleration characteristics are generated according to the abnormal transient marker fusion energy proportion and the permutation entropy value.

9. The cable fault big data early warning system according to claim 8, characterized in that: The temperature rise acceleration characteristics are generated based on the abnormal transient marker fusion energy ratio and the permutation entropy value, including: When the abnormal transient marker is activated, the energy proportion of the eigenmode component of the specified order is selected and multiplied with the permutation entropy value to generate the temperature rise acceleration feature. Otherwise, the accumulated sum of the energy proportions is used as the temperature rise acceleration feature.

10. The cable fault big data early warning system according to claim 5, characterized in that: Determine the split features based on the encoded data to build a decision tree, including: When generating the first encoding result, a decision tree is constructed with the temperature rise acceleration feature as the priority splitting feature; When generating the second encoding result, the product of the real temperature rise and the temperature rise acceleration characteristic is used as the basis for splitting.

Images