A high-voltage cable insulation state online evaluation method, system, terminal device and storage medium

Abstract

The application discloses a kind of high-voltage cable insulation state online evaluation method, system, terminal equipment and storage medium, belong to cable monitoring field.The method is applicable to the control center of high-voltage cable insulation state online evaluation system, after low-frequency signal injection source module injects test signal to the cable to be evaluated, dielectric loss data collected by dielectric loss signal monitoring module is acquired in real time;Hidden state sequence is obtained according to dielectric loss data dynamic modeling;According to hidden state sequence, dynamic change rate, trend characteristics and key point characteristics are obtained;According to hidden state sequence and time convolution network model, local feature, long-term dependence feature and multi-scale feature are obtained;The feature matrix of dynamic change rate, trend characteristics, key point characteristics, local feature, long-term dependence feature and multi-scale feature is integrated and input into the preset cable insulation state evaluation model for evaluation, and insulation state evaluation result is obtained.By implementing the application, the problem of low accuracy of cable insulation state evaluation in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of cable monitoring technology, and in particular to an online assessment method, system, terminal equipment, and storage medium for the insulation condition of high-voltage cables. Background Technology

[0002] In power systems, high-voltage cables are critical equipment for power transmission, and their operational reliability directly affects the safety and stability of the entire power system. However, during long-term operation, high-voltage cables are inevitably affected by various factors, leading to a gradual deterioration of their insulation. These factors include, but are not limited to, electrical aging, thermal aging, mechanical damage, and environmental corrosion. Deterioration of insulation causes an abnormal increase in the cable's dielectric loss factor, a crucial indicator of cable insulation performance. An abnormally high dielectric loss factor signifies a decline in insulation performance, potentially leading to faults such as localized overheating and insulation breakdown, and posing a serious threat to the normal operation of the power system.

[0003] Current methods for monitoring dielectric loss have significant limitations. On the one hand, traditional offline testing methods require removing the cable from the operating system, which not only affects the normal power supply of the power system but also fails to reflect changes in the insulation status of the cable during operation in a timely manner. On the other hand, existing online monitoring methods are affected by industrial noise interference, which reduces the accuracy of monitoring data and makes it difficult to accurately assess the insulation status of the cable. Summary of the Invention

[0004] This invention provides a method, system, terminal equipment, and storage medium for online assessment of the insulation status of high-voltage cables. It effectively solves the problem that existing online monitoring methods are affected by industrial noise interference, which reduces the accuracy of monitoring data and makes it difficult to accurately assess the insulation status of cables.

[0005] One embodiment of the present invention provides an online assessment method for the insulation condition of high-voltage cables, applicable to the control center of an online assessment system for the insulation condition of high-voltage cables; the online assessment system for the insulation condition of high-voltage cables further includes: a low-frequency signal injection source module and a dielectric loss signal monitoring module;

[0006] The online assessment method for the insulation condition of high-voltage cables includes:

[0007] After the low-frequency signal injection source module injects a test signal of a preset frequency into the cable to be evaluated, the dielectric loss data of the cable to be evaluated collected by the dielectric loss signal monitoring module is acquired in real time.

[0008] Dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence; and sequence feature extraction is performed based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features.

[0009] Convolutional feature extraction is performed based on the hidden state sequence and the preset temporal convolutional network model to obtain local features for characterizing short-term changes in dielectric loss data, long-term dependency features for characterizing long-term trends in dielectric loss data, and multi-scale features.

[0010] The feature matrix is ​​obtained by integrating dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features.

[0011] The feature matrix is ​​input into a preset cable insulation condition assessment model to perform insulation condition assessment, and the insulation condition assessment result of the cable to be assessed is obtained.

[0012] Furthermore, dynamic modeling is performed based on the aforementioned dielectric loss data to obtain a hidden state sequence, including:

[0013] Based on the dielectric loss data, determine the time series characteristics of the dielectric loss data;

[0014] Based on a preset neural differential equation, the time series features of the media loss data are mapped onto the preset neural differential equation to obtain a neural differential equation model.

[0015] The initial hidden state is determined based on the neural differential equation model, and the initial hidden state is initialized to a zero vector.

[0016] The hidden state sequence is obtained by numerically solving the zero vector and the neural differential equation model.

[0017] Furthermore, the pre-defined temporal convolutional network model includes causal convolutional layers, dilated convolutional layers, and multiple convolutional layers;

[0018] Convolutional feature extraction is performed based on the hidden state sequence and a preset temporal convolutional network model to obtain local features characterizing short-term changes in dielectric loss data, long-term dependency features characterizing long-term trends in dielectric loss data, and multi-scale features, including:

[0019] Based on the hidden state sequence, the change pattern within a preset short window is extracted in the causal convolutional layer of the temporal convolutional network model to obtain local features that characterize the short-term changes of the medium loss data.

[0020] Based on the hidden state sequence, the long-term trend of the settlement data is captured by the dilated convolutional layer in the temporal convolutional network model, and long-term dependency features are obtained to characterize the long-term trend of the medium loss data.

[0021] Based on the hidden state sequence, features at different time scales are extracted in the multiple convolutional layers of the temporal convolutional network model to obtain multi-scale features.

[0022] Furthermore, the training of the cable insulation condition assessment model includes:

[0023] Obtain historical dielectric loss data;

[0024] Insulation status is labeled on historical dielectric loss data to obtain a labeled training dataset;

[0025] Dynamic modeling is performed based on the labeled training dataset to obtain the historical hidden state sequence; and sequence feature extraction is performed based on the historical hidden state sequence to obtain the historical dynamic change rate, historical trend features, and historical key point features.

[0026] Convolutional feature extraction is performed based on the historical hidden state sequence to obtain historical local features, historical long-term dependency features, and historical multi-scale features.

[0027] The historical feature matrix is ​​obtained by integrating historical dynamic change rate, historical trend characteristics, historical key point characteristics, historical local characteristics, historical long-term dependence characteristics, and historical multi-scale characteristics.

[0028] The historical feature matrix is ​​input into the teacher model to be trained for pre-training to obtain the predicted insulation state of the teacher model;

[0029] The historical feature matrix and the attention weights corresponding to the teacher model during each pre-training are used as the inputs for each training of the student model to be trained, and the student model to be trained is trained to obtain the predicted insulation state of the student model.

[0030] Based on the predicted insulation state of the teacher model and the labeled training dataset, calculate the first loss function value of the preset distillation loss function; and based on the predicted insulation state of the student model and the labeled training dataset, calculate the second loss function value of the preset student model loss function.

[0031] The total loss function value is calculated based on the first loss function value, the second loss function value, and the preset loss function weights.

[0032] When the total loss function converges, the trained student model is obtained; and the trained student model is used as the trained cable insulation state evaluation model.

[0033] The teacher model to be trained and the student model to be trained are convolutional neural network models with the same structure.

[0034] Furthermore, it also includes: filtering and screening the dielectric loss data to obtain filtered dielectric loss data;

[0035] The filtered dielectric loss data are normalized to obtain the final dielectric loss data.

[0036] Furthermore, the online insulation condition assessment system for high-voltage cables also includes a dielectric loss monitoring synchronization module; the dielectric loss monitoring synchronization module includes an optical fiber and a photoelectric converter; the optical fiber is laid on the surface of the cable to be assessed;

[0037] The optical fiber is used to transmit the optical signal injected by the control center to the photoelectric converter;

[0038] The photoelectric converter is used to convert optical signals into electrical signals and transmit the electrical signals to the dielectric loss signal monitoring module, so that the dielectric loss signal monitoring module can synchronously collect dielectric loss data of the cable to be evaluated according to the time difference of the electrical signals.

[0039] Furthermore, it also includes: continuing to monitor the dielectric loss data of the cable to be evaluated when the insulation condition assessment result is good;

[0040] If the insulation condition assessment indicates a minor fault, increase the frequency of optical signal injection.

[0041] If the insulation condition assessment result is a moderate fault, local repair measures shall be taken for the cable to be assessed.

[0042] If the insulation condition assessment result is a severe fault, disconnect the power supply circuit of the cable being assessed and replace the cable.

[0043] As an improvement to the above solution, another embodiment of the present invention provides an online evaluation system for the insulation status of high-voltage cables, including: a control center, a low-frequency signal injection source module, and a dielectric loss signal monitoring module;

[0044] The low-frequency signal injection source module is used to receive low-frequency signal injection instructions from the control center, and after receiving the low-frequency signal injection instructions from the control center, inject a test signal of a preset frequency into the cable to be evaluated.

[0045] The dielectric loss signal monitoring module is used to collect dielectric loss data of the cable to be evaluated after injecting a test signal of a preset frequency into the cable to be evaluated.

[0046] The control center is used to send low-frequency signal injection commands to the low-frequency signal injection source module and to acquire dielectric loss data of the cable to be evaluated in real time.

[0047] Dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence; and sequence feature extraction is performed based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features.

[0048] Convolutional feature extraction is performed based on the hidden state sequence and the preset temporal convolutional network model to obtain local features for characterizing short-term changes in dielectric loss data, long-term dependency features for characterizing long-term trends in dielectric loss data, and multi-scale features.

[0049] The feature matrix is ​​obtained by integrating dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features.

[0050] The feature matrix is ​​input into a preset cable insulation condition assessment model to perform insulation condition assessment, and the insulation condition assessment result of the cable to be assessed is obtained.

[0051] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an online evaluation method for the insulation status of a high-voltage cable as described in the above embodiments.

[0052] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the online evaluation method for the insulation status of a high-voltage cable as described in the above embodiment.

[0053] By implementing this invention, at least the following beneficial effects are achieved:

[0054] This invention provides a method, system, terminal equipment, and storage medium for online assessment of the insulation status of high-voltage cables. The method utilizes a low-frequency signal injection source module and a dielectric loss signal monitoring module to collect dielectric loss data in real time without interrupting the cable being assessed, avoiding power outages caused by testing and ensuring the normal power supply of the power system. Simultaneously, the control center acquires the dielectric loss data of the cable being assessed in real time and then performs online insulation status assessment, reflecting changes in the cable's insulation status during operation and improving the timeliness of the assessment. By extracting dynamic change rate, trend features, and key point features, short-term noise can be filtered from the sequence evolution pattern. A preset temporal convolutional network model extracts local features, long-term dependency features, and multi-scale features, distinguishing noise disturbances from actual insulation changes at different time scales. The dynamic change rate, trend features, key point features, local features, long-term dependency features, and multi-scale features are integrated to obtain a feature matrix, further reducing noise interference and improving the accuracy of the feature matrix relative to the dielectric loss data. This results in a more accurate final output of the insulation status assessment result for the cable being assessed. Attached Figure Description

[0055] Figure 1This is a flowchart illustrating an online evaluation method for the insulation status of high-voltage cables according to an embodiment of the present invention.

[0056] Figure 2 This is a schematic diagram of an online monitoring platform for dielectric loss of high-voltage cables provided in an embodiment of the present invention;

[0057] Figure 3 This is a flowchart of a cable dielectric loss data feature extraction method based on neural differential equations provided in an embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram of the training of a cable insulation state assessment model based on knowledge distillation provided in an embodiment of the present invention;

[0059] Figure 5 This is a schematic diagram of the structure of an online evaluation system for the insulation status of high-voltage cables provided in an embodiment of the present invention. Detailed Implementation

[0060] 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, and 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.

[0061] See Figure 1 To address the problem that existing online monitoring methods are susceptible to industrial noise interference, leading to reduced accuracy of monitoring data and making it difficult to accurately assess the insulation status of cables, an embodiment of the present invention provides a flowchart of an online assessment method for the insulation status of high-voltage cables. This method is applicable to the control center of an online assessment system for the insulation status of high-voltage cables. The online assessment system for the insulation status of high-voltage cables further includes a low-frequency signal injection source module and a dielectric loss signal monitoring module.

[0062] Specifically, the low-frequency signal injection source module is connected to the cable under evaluation; the cable under evaluation is connected to the dielectric loss signal monitoring module; and the control center is connected to both the low-frequency signal injection source module and the dielectric loss signal monitoring module. Through the collaboration between the control center, the low-frequency signal injection source module, and the dielectric loss signal monitoring module, online evaluation of the insulation condition of the cable under evaluation is achieved. The dielectric loss signal monitoring module consists of a dielectric loss current monitoring component and a dielectric loss voltage monitoring component, containing two sensors and a data acquisition card, such as... Figure 2As shown, it includes a magnetically coupled dielectric loss current transformer and a dielectric loss voltage transformer. The magnetically coupled dielectric loss current transformer is fitted onto the three-phase cable (the cable to be evaluated) to measure the zero-sequence ultra-low frequency current on the cable conductor; the dielectric loss voltage transformer is connected to the output terminal of the injection source in the low-frequency signal injection source module to measure the output voltage of the injection source; the data acquisition card is connected to the control center to acquire the two signals and perform digital processing.

[0063] like Figure 2 As shown, the low-frequency signal injection source module injects a 0.1Hz characteristic frequency test signal into the open delta side of the neutral point voltage transformer of the cable to be evaluated. Figure 2 The test signal is transmitted to ground via the impedance to ground of the cable under evaluation (0.1Hz test voltage). By monitoring the voltage and current of this circuit, online monitoring of the dielectric loss data of the cable under evaluation can be achieved.

[0064] Preferably, the online insulation condition assessment system for high-voltage cables further includes a dielectric loss monitoring synchronization module; the dielectric loss monitoring synchronization module includes an optical fiber and a photoelectric converter; the optical fiber is laid on the surface of the cable to be assessed;

[0065] The optical fiber is used to transmit the optical signal injected by the control center to the photoelectric converter;

[0066] The photoelectric converter is used to convert optical signals into electrical signals and transmit the electrical signals to the dielectric loss signal monitoring module, so that the dielectric loss signal monitoring module can synchronously collect dielectric loss data of the cable to be evaluated according to the time difference of the electrical signals.

[0067] In a preferred embodiment of the present invention, the dielectric loss monitoring synchronization module is a module that synchronizes the timing of low-frequency signal injection and dielectric loss data acquisition, solving the synchronization error problem caused by electromagnetic interference under high-voltage conditions. During online dielectric loss monitoring, the control center controls the injection of optical signals into the optical fiber every 20 seconds. The optical signals are then converted by a photoelectric converter (…). Figure 2 The high-speed photoelectric conversion module in the optical fiber converts the signal into an electrical signal, which is then received by the data acquisition card. The control center uses the time difference of the electrical signal acquired by the data acquisition card to perform synchronous monitoring of dielectric loss current and voltage. The optical fiber transmission of the optical signal is resistant to high-voltage electromagnetic interference (avoiding 50Hz power frequency and harmonic interference around the cable), solving the interference problem of traditional electrical signal synchronization; the photoelectric conversion response speed is fast, and the time difference is fixed and calibrable, ensuring the timing synchronization of dielectric loss data acquisition and low-frequency signal injection; it is compatible with cables of different lengths and laying environments (such as direct burial and tunnel laying), and has strong versatility.

[0068] The online assessment method for the insulation condition of high-voltage cables includes:

[0069] S1. After the low-frequency signal injection source module injects a test signal of a preset frequency into the cable to be evaluated, the dielectric loss data of the cable to be evaluated collected by the dielectric loss signal monitoring module is obtained in real time.

[0070] Specifically, dielectric loss data reflects the dielectric loss characteristics of high-voltage cable insulation, including dielectric loss angle data. The control center sends instructions to the low-frequency signal injection source module to inject a test signal of a preset frequency, such as a test signal with a frequency of 0.1-1Hz, which is lower than the main frequency of industrial noise (50Hz) and harmonics, thus reducing noise interference. At the same time, the dielectric loss signal monitoring module collects the dielectric loss data of the cable in real time and transmits it to the control center, such as sampling the dielectric loss angle data every 10ms.

[0071] S2. Dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence; and sequence feature extraction is performed based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features;

[0072] Specifically, the hidden state sequence is a state sequence generated during the dynamic modeling process that cannot be directly observed but reflects the inherent laws of dielectric loss data. The control center performs dynamic modeling on the collected dielectric loss data to generate the hidden state sequence. The dynamic change rate is calculated by the difference between the hidden states at adjacent time points, reflecting the speed of fluctuation in the dielectric loss data; the trend feature indicates whether the dielectric loss data is rising, falling, or stable by fitting a linear trend through a sliding window; the key point feature indicates the abrupt change points in the hidden state sequence.

[0073] Preferably, dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence, including:

[0074] Based on the dielectric loss data, determine the time series characteristics of the dielectric loss data;

[0075] Based on a preset neural differential equation, the time series features of the media loss data are mapped onto the preset neural differential equation to obtain a neural differential equation model.

[0076] The initial hidden state is determined based on the neural differential equation model, and the initial hidden state is initialized to a zero vector.

[0077] The hidden state sequence is obtained by numerically solving the zero vector and the neural differential equation model.

[0078] Specifically, considering the dynamic characteristics of cable dielectric loss data, a modeling method based on neural differential equations is constructed. Neural differential equations map the time-series characteristics of dielectric loss data to a continuous dynamic system, capturing the nonlinear variation patterns of the data and obtaining a neural differential equation model. The mathematical expression of the neural differential equation model is as follows:

[0079]

[0080] in, The hidden state sequence represents the characteristic representation of dielectric loss data at time t; For neural network parameters; The dynamic function defined for the neural network describes the changing patterns of the hidden states. The media loss data x(t) is input into the neural differential equation model, and the dynamic changes of the media loss data, i.e., the hidden state sequence h(t), are obtained through optimization of the initial hidden state h(0) and the dynamic function f. The initial hidden state h(0) is first initialized to a zero vector or a random vector. A multilayer perceptron is used as the dynamic function. In this embodiment, the input dimension d of the multilayer perceptron is... h =64; two hidden layers, each with 64 neurons; using the ReLU activation function; output dimension d out =64. Then, the Runge-Kutta method is used to solve the neural differential equation, obtaining the hidden state sequence h(t). Its formula is:

[0081]

[0082] Among them, the integral variable Iterate through any intermediate time between 0 and t to characterize the continuous evolution of the hidden state over time.

[0083] Specifically, sequence features are extracted based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features:

[0084] Dynamic rate of change: This reflects the instantaneous changing trend of dielectric loss data.

[0085] Trend characteristics: Where T=10 is the size of the time window. Trend characteristics are analyzed... The changing trend of dielectric loss data is used to extract long-term changing patterns.

[0086] Key features: through monitoring Inflection points are used to extract abnormal changes in dielectric loss data. The moment corresponding to the abrupt change in the dielectric loss angle of the cable is used in this embodiment to characterize the nonlinear abrupt change in the dielectric loss data. ,in The threshold represents the lower limit of the rate of change of the inflection point in dielectric loss data. Inflection point The larger the value, the stronger the abrupt change in the cable dielectric loss angle. When the inflection point value exceeds the threshold, ... An output value of 1 indicates that a media-induced mutational event has occurred; conversely, an output value of 0 indicates that no media-induced mutational event has occurred.

[0087] Schematic, it also includes: filtering the dielectric loss data to obtain filtered dielectric loss data;

[0088] The filtered dielectric loss data are normalized to obtain the final dielectric loss data.

[0089] Specifically, the filtered cable dielectric loss data is screened to remove outliers; the screened dielectric loss data is then normalized to ensure that the data is distributed within the interval [0, 1]. Normalization is performed using the following formula:

[0090] . This indicates the final dielectric loss data; This represents the minimum value of the dielectric loss data; This indicates the maximum value of the dielectric loss data; This indicates dielectric loss data.

[0091] Initialization with the initial hidden state zero vector ensures a unified modeling starting point and improves the consistency of evaluations among different cables; the numerical solution process can filter high-frequency noise, such as instantaneous fluctuations caused by industrial interference, to obtain a more robust hidden state sequence.

[0092] S3. Based on the hidden state sequence and the preset temporal convolutional network model, perform convolutional feature extraction to obtain local features for characterizing short-term changes in dielectric loss data, long-term dependency features for characterizing long-term trends in dielectric loss data, and multi-scale features.

[0093] Specifically, the temporal convolutional network model is a type of convolutional neural network suitable for processing time-series data. It captures short-term fluctuations and long-term dependencies in data through structures such as causal convolution and dilated convolution.

[0094] Preferably, the preset temporal convolutional network model includes causal convolutional layers, dilated convolutional layers, and multiple convolutional layers;

[0095] Convolutional feature extraction is performed based on the hidden state sequence and a preset temporal convolutional network model to obtain local features characterizing short-term changes in dielectric loss data, long-term dependency features characterizing long-term trends in dielectric loss data, and multi-scale features, including:

[0096] Based on the hidden state sequence, the change pattern within a preset short window is extracted in the causal convolutional layer of the temporal convolutional network model to obtain local features that characterize the short-term changes of the medium loss data.

[0097] Based on the hidden state sequence, the long-term trend of the settlement data is captured by the dilated convolutional layer in the temporal convolutional network model, and long-term dependency features are obtained to characterize the long-term trend of the medium loss data.

[0098] Based on the hidden state sequence, features at different time scales are extracted in the multiple convolutional layers of the temporal convolutional network model to obtain multi-scale features.

[0099] Specifically, a causal convolutional layer refers to a convolutional layer whose convolution operation only depends on current and historical data (without using future data), achieved through left padding to ensure temporal consistency (e.g., a convolution at time t only uses data from times t-2, t-1, and t). A dilated convolutional layer refers to a convolutional layer that inserts "holes" (dilation rate > 1) between kernel elements, expanding the receptive field. For example, with a dilation rate of 2, the receptive field is twice that of a regular convolution, used to capture long-term temporal dependencies. A multi-convolutional layer refers to a parallel convolutional layer containing multiple kernels of different scales (e.g., 1×1, 3×1, 5×1), capable of simultaneously extracting features from different time windows, such as the variation patterns of 10ms, 30ms, and 50ms windows.

[0100] In a preferred embodiment of the present invention, the temporal convolutional network model achieves feature extraction of time series data through dilated convolution and causal convolution, enabling it to capture local features and long-term dependencies in lossy data. Its mathematical expression is:

[0101]

[0102] in, Features extracted for a temporal convolutional network model; d is the hidden state sequence output by the neural differential equation model; d is the expansion factor that controls the convolution receptive field, which is 4 in this embodiment; k is the convolution kernel size, which is 3 in this embodiment. The weights are the convolution kernel weights, initialized with a Gaussian distribution; for The sequence of hidden states at each moment, i.e., the historical hidden states.

[0103] In the causal convolutional layer of a temporal convolutional network model, convolution operations are used to extract change patterns within a short time window, such as abrupt changes and short-term fluctuations in lossy data. The calculation formula is:

[0104]

[0105] in, Let represent the sequence of hidden states at time ti.

[0106] In temporal convolutional network models, dilated convolutional layers capture long-term trends in dielectric loss data, such as periodic variations and slow aging trends. The calculation formula is:

[0107]

[0108] In temporal convolutional network models, multiple convolutional layers extract features at different time scales, forming multi-scale features. The calculation formula is as follows:

[0109]

[0110] Where L represents the number of convolutional layers.

[0111] S4. Integrate the dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features to obtain the feature matrix;

[0112] Specifically, the feature matrix is ​​a matrix formed by integrating multi-dimensional features, which is used as input to the evaluation model for state determination.

[0113] In a preferred embodiment of the present invention, the extracted dynamic change rate, trend features, and key point features are first integrated and represented as a feature matrix. Where T0 is the length of the time series, The dimension is T0×3, meaning there are 3 features at each time point. Then, the local features, long-term dependency features, and multi-scale features extracted by the temporal convolutional network model are integrated and expressed as follows: . Dimension is T0×d z , where d z This refers to the output feature dimension of the temporal convolutional network model, which is 128 in this example. Finally, integrate... , The final feature matrix Z is obtained. ].

[0114] In a preferred embodiment of the present invention, such as Figure 3 As shown, after acquiring dielectric loss data online, the data is preprocessed to remove outliers and normalize, resulting in the final dielectric loss data. Figure 3 The function is to first obtain x(t) by solving the neural differential equations. Then, the hidden state sequence is obtained, and dynamic change features, trend features, and key point features are obtained from the hidden state sequence. At the same time, local features, long-term dependency features, and multi-scale features are obtained by solving the hidden state sequence in the temporal convolutional network model. Finally, the feature matrix Z is constructed.

[0115] S5. Input the feature matrix into the preset cable insulation status assessment model to perform insulation status assessment and obtain the insulation status assessment result of the cable to be assessed.

[0116] Preferably, the training of the cable insulation condition assessment model includes:

[0117] Obtain historical dielectric loss data;

[0118] Insulation status is labeled on historical dielectric loss data to obtain a labeled training dataset;

[0119] Dynamic modeling is performed based on the labeled training dataset to obtain the historical hidden state sequence; and sequence feature extraction is performed based on the historical hidden state sequence to obtain the historical dynamic change rate, historical trend features, and historical key point features.

[0120] Convolutional feature extraction is performed based on the historical hidden state sequence to obtain historical local features, historical long-term dependency features, and historical multi-scale features.

[0121] The historical feature matrix is ​​obtained by integrating historical dynamic change rate, historical trend characteristics, historical key point characteristics, historical local characteristics, historical long-term dependence characteristics, and historical multi-scale characteristics.

[0122] The historical feature matrix is ​​input into the teacher model to be trained for pre-training to obtain the predicted insulation state of the teacher model;

[0123] The historical feature matrix and the attention weights corresponding to the teacher model during each pre-training are used as the inputs for each training of the student model to be trained, and the student model to be trained is trained to obtain the predicted insulation state of the student model.

[0124] Based on the predicted insulation state of the teacher model and the labeled training dataset, calculate the first loss function value of the preset distillation loss function; and based on the predicted insulation state of the student model and the labeled training dataset, calculate the second loss function value of the preset student model loss function.

[0125] The total loss function value is calculated based on the first loss function value, the second loss function value, and the preset loss function weights.

[0126] When the total loss function converges, the trained student model is obtained; and the trained student model is used as the trained cable insulation state evaluation model.

[0127] The teacher model to be trained and the student model to be trained are convolutional neural network models with the same structure.

[0128] Specifically, historical dielectric loss data for 100 high-voltage cables (each cable containing 1000 sets of dielectric loss angle sequences, with a sampling interval of 20ms) were collected. These data, combined with maintenance records, were used to label the insulation status (2000 sets of good status, 3000 sets of minor faults, 3000 sets of moderate faults, and 2000 sets of severe faults) to form a training dataset. The training dataset can be divided into training and testing sets in a 7:3 ratio. The teacher model to be trained takes a historical feature matrix as input and outputs a predicted insulation status. It uses a cross-entropy loss function and iteratively trains until the loss converges to obtain a pre-trained teacher model. The student model to be trained takes a historical feature matrix as input plus the attention weights of the teacher model (e.g., the weights of keypoint features in convolutional layers). The parameters are iteratively optimized by mimicking the output and attention distribution of the teacher model.

[0129] In a preferred embodiment of the present invention, the teacher model is an 18-layer CNN model. The reference training parameters given in this embodiment are: initial learning rate of Adam optimizer 0.002; batch size = 16; training epco = 100 and an early stopping mechanism to prevent overfitting. The trained teacher model is saved. A student model is then built, with the same architecture as the teacher model, which is an 18-layer CNN model. A total loss function for the knowledge distillation training method is constructed. Distillation loss function And student model loss function It consists of two parts, the proportion of which is determined by coefficients α and β, and both are calculated using the classification cross-entropy method. Specifically:

[0130]

[0131]

[0132]

[0133] in, This refers to the label output of the teacher model on the training data, i.e., the predicted insulation state of the teacher model. This is the label output of the student model on the training data, i.e., the predicted insulation state of the student model. Let be the predicted probability distribution of the student model for the j-th training data. Let be the true label of the j-th training data point, such as the true category of insulation state. N is the number of training data points. T1 is called the distillation temperature, which measures the strength of the student model's knowledge acquisition from the teacher model. In this embodiment, α is given a reference value of 0.4, β a reference value of 0.6, and T a reference value of 3. The student model is trained according to the constructed total loss function, using the Adam optimizer, an initial learning rate (lr) of 0.02, a training epco of 250, and a batch size of 16. The trained student model is then used as the trained cable insulation state assessment model for online diagnosis of high-voltage cable insulation state. Figure 4 As shown, the feature matrix Z serves as the input to both the teacher model and the student model. Simultaneously, the attention weights of the teacher model are also used as the input to the student model. A comprehensive loss function is constructed to train the student model, resulting in the final insulation state assessment model.

[0134] In a preferred embodiment of the present invention, taking a 110kV high-voltage cable as an example, a low-frequency signal injection source module injects a 0.5Hz test signal; a dielectric loss signal monitoring module collects dielectric loss angle data δ at a 20ms sampling interval and dynamically models and generates a hidden state sequence; the sequence feature extraction yields a dynamic change rate of 0.001° / 20ms, a stable trend (10-minute window fitting slope ≈ 0), and no key points; the temporal convolutional network model extracts: local features reflect fluctuations ≤0.002° within 5 seconds, long-term dependent features reflect no obvious trend within 1 hour, and multi-scale features cover a scale of 10 seconds to 1 minute; the feature matrix is ​​input into the cable insulation state assessment model, and the output state is good, achieving accurate assessment of the insulation state.

[0135] Indicatively, it also includes: continuing to monitor the dielectric loss data of the cable to be evaluated when the insulation condition assessment result is good;

[0136] If the insulation condition assessment indicates a minor fault, increase the frequency of optical signal injection.

[0137] If the insulation condition assessment result is a moderate fault, local repair measures shall be taken for the cable to be assessed.

[0138] If the insulation condition assessment result is a severe fault, disconnect the power supply circuit of the cable being assessed and replace the cable.

[0139] In a preferred embodiment of the present invention, when the insulation condition assessment result is good, the current routine inspection cycle of the high-voltage cable is maintained, such as infrared thermography and partial discharge detection once a quarter, and long-term trend changes in dielectric loss data are continuously monitored; when the insulation condition assessment result is a minor fault, targeted pre-maintenance is triggered: high-frequency inspections are increased, such as monthly dielectric loss retesting and cable partial discharge location, historical data is retrieved to trace the dielectric loss characteristic change curve, and possible weak points in insulation, such as joints and terminals, are investigated; when the insulation condition assessment result is a moderate fault, the emergency operation and maintenance process is initiated: power outage testing is immediately arranged, such as withstand voltage testing and insulation resistance testing, and local repair measures are adopted simultaneously, such as insulation reinforcement of cable accessories, replacement of aging seals, assessment of the risk of fault expansion, and development of load transfer plans; when the insulation condition assessment result is a serious fault, emergency fault handling is performed: the power supply circuit of the cable to be assessed is immediately cut off, the backup power supply channel is activated, cable replacement or overall repair work is carried out, and the root cause of the fault is traced simultaneously, such as analysis of construction records and insulation material aging cycles, and updating of the operation and maintenance knowledge base.

[0140] By implementing this embodiment, through the low-frequency signal injection source module and the dielectric loss signal monitoring module, dielectric loss data can be collected in real time without interrupting the cable under evaluation, avoiding power outages caused by testing and ensuring the normal power supply of the power system. At the same time, the control center obtains the dielectric loss data of the cable under evaluation in real time, and then performs online insulation status evaluation through the control center, which can reflect the changes in the insulation status of the cable during operation in a timely manner, improving the timeliness of the evaluation. By extracting dynamic change rate, trend features, and key point features, short-term noise can be filtered from the sequence evolution law. Local features, long-term dependence features, and multi-scale features are extracted through a preset time convolutional network model, which distinguishes noise disturbances from real insulation changes at different time scales. The dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features are integrated to obtain a feature matrix, which further reduces noise interference, thereby improving the accuracy of the feature matrix relative to the dielectric loss data, and thus making the final output insulation status evaluation result of the cable under evaluation more accurate.

[0141] See Figure 5 This is a schematic diagram of the structure of an online evaluation system for the insulation status of high-voltage cables provided in an embodiment of the present invention, including: a control center, a low-frequency signal injection source module, and a dielectric loss signal monitoring module;

[0142] The low-frequency signal injection source module is used to receive low-frequency signal injection instructions from the control center, and after receiving the low-frequency signal injection instructions from the control center, inject a test signal of a preset frequency into the cable to be evaluated.

[0143] The dielectric loss signal monitoring module is used to collect dielectric loss data of the cable to be evaluated after injecting a test signal of a preset frequency into the cable to be evaluated.

[0144] The control center is used to send low-frequency signal injection commands to the low-frequency signal injection source module and to acquire dielectric loss data of the cable to be evaluated in real time.

[0145] Dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence; and sequence feature extraction is performed based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features.

[0146] Convolutional feature extraction is performed based on the hidden state sequence and the preset temporal convolutional network model to obtain local features for characterizing short-term changes in dielectric loss data, long-term dependency features for characterizing long-term trends in dielectric loss data, and multi-scale features.

[0147] The feature matrix is ​​obtained by integrating dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features.

[0148] The feature matrix is ​​input into a preset cable insulation condition assessment model to perform insulation condition assessment, and the insulation condition assessment result of the cable to be assessed is obtained.

[0149] This invention provides an online evaluation system for the insulation status of high-voltage cables. The system receives low-frequency signal injection commands from a control center via a low-frequency signal injection source module, and injects a test signal of a preset frequency into the cable to be evaluated. A dielectric loss monitoring module collects dielectric loss data of the cable after the preset frequency test signal is injected. The system sends low-frequency signal injection commands to the low-frequency signal injection source module from the control center. It acquires dielectric loss data of the cable to be evaluated in real time and performs dynamic modeling based on the dielectric loss data to obtain a hidden state sequence. The system then... Sequence features are extracted from the state sequence to obtain dynamic change rate, trend features, and key point features. Convolutional features are extracted based on the hidden state sequence and a preset temporal convolutional network model to obtain local features characterizing short-term changes in dielectric loss data, long-term dependency features characterizing long-term trends in dielectric loss data, and multi-scale features. The dynamic change rate, trend features, key point features, local features, long-term dependency features, and multi-scale features are integrated to obtain a feature matrix. The feature matrix is ​​then input into a preset cable insulation condition assessment model to perform insulation condition assessment, obtaining the insulation condition assessment result of the cable to be assessed.

[0150] By using a low-frequency signal injection source module and a dielectric loss signal monitoring module, dielectric loss data can be collected in real time without interrupting the cable under evaluation, avoiding power outages caused by testing and ensuring the normal power supply of the power system. Simultaneously, the control center acquires the dielectric loss data of the cable under evaluation in real time and then performs online insulation status assessment, reflecting changes in the cable's insulation status during operation and improving the timeliness of the assessment. By extracting dynamic change rate, trend features, and key point features, short-term noise can be filtered from the sequence evolution pattern. A preset temporal convolutional network model extracts local features, long-term dependency features, and multi-scale features, distinguishing noise disturbances from actual insulation changes at different time scales. The dynamic change rate, trend features, key point features, local features, long-term dependency features, and multi-scale features are integrated to obtain a feature matrix, further reducing noise interference and improving the accuracy of the feature matrix relative to the dielectric loss data. This results in a more accurate final insulation status assessment result for the cable under evaluation.

[0151] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0152] Those skilled in the art will understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0153] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an online assessment method for the insulation status of high-voltage cables as described in the above embodiments. The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0154] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0155] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0156] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the online evaluation method for the insulation status of a high-voltage cable as described in the above embodiment.

[0157] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0158] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for online assessment of the insulation condition of high-voltage cables, characterized in that, A control center suitable for online assessment systems of high-voltage cable insulation condition; The online assessment system for the insulation status of high-voltage cables also includes: a low-frequency signal injection source module and a dielectric loss signal monitoring module; The online assessment method for the insulation condition of high-voltage cables includes: After the low-frequency signal injection source module injects a test signal of a preset frequency into the cable to be evaluated, the dielectric loss data of the cable to be evaluated collected by the dielectric loss signal monitoring module is acquired in real time. Dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence; and sequence feature extraction is performed based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features. Convolutional feature extraction is performed based on the hidden state sequence and the preset temporal convolutional network model to obtain local features for characterizing short-term changes in dielectric loss data, long-term dependency features for characterizing long-term trends in dielectric loss data, and multi-scale features. The feature matrix is ​​obtained by integrating dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features. The feature matrix is ​​input into a preset cable insulation condition assessment model to perform insulation condition assessment, and the insulation condition assessment result of the cable to be assessed is obtained.

2. The online assessment method for the insulation condition of high-voltage cables as described in claim 1, characterized in that, Dynamic modeling is performed based on the aforementioned dielectric loss data to obtain a hidden state sequence, including: Based on the dielectric loss data, determine the time series characteristics of the dielectric loss data; Based on a preset neural differential equation, the time series features of the media loss data are mapped onto the preset neural differential equation to obtain a neural differential equation model. The initial hidden state is determined based on the neural differential equation model, and the initial hidden state is initialized to a zero vector. The hidden state sequence is obtained by numerically solving the zero vector and the neural differential equation model.

3. The online assessment method for the insulation condition of high-voltage cables as described in claim 1, characterized in that, The preset temporal convolutional network model includes causal convolutional layers, dilated convolutional layers, and multiple convolutional layers; Convolutional feature extraction is performed based on the hidden state sequence and a preset temporal convolutional network model to obtain local features characterizing short-term changes in dielectric loss data, long-term dependency features characterizing long-term trends in dielectric loss data, and multi-scale features, including: Based on the hidden state sequence, the change pattern within a preset short window is extracted in the causal convolutional layer of the temporal convolutional network model to obtain local features that characterize the short-term changes of the medium loss data. Based on the hidden state sequence, the long-term trend of the settlement data is captured by the dilated convolutional layer in the temporal convolutional network model, and long-term dependency features are obtained to characterize the long-term trend of the medium loss data. Based on the hidden state sequence, features at different time scales are extracted in the multiple convolutional layers of the temporal convolutional network model to obtain multi-scale features.

4. The online assessment method for the insulation condition of high-voltage cables as described in claim 1, characterized in that, The training of the cable insulation condition assessment model includes: Obtain historical dielectric loss data; Insulation status is labeled on historical dielectric loss data to obtain a labeled training dataset; Dynamic modeling is performed based on the labeled training dataset to obtain the historical hidden state sequence; and sequence feature extraction is performed based on the historical hidden state sequence to obtain the historical dynamic change rate, historical trend features, and historical key point features. Convolutional feature extraction is performed based on the historical hidden state sequence to obtain historical local features, historical long-term dependency features, and historical multi-scale features. The historical feature matrix is ​​obtained by integrating historical dynamic change rate, historical trend characteristics, historical key point characteristics, historical local characteristics, historical long-term dependence characteristics, and historical multi-scale characteristics. The historical feature matrix is ​​input into the teacher model to be trained for pre-training to obtain the predicted insulation state of the teacher model; The historical feature matrix and the attention weights corresponding to the teacher model during each pre-training are used as the inputs for each training of the student model to be trained, and the student model to be trained is trained to obtain the predicted insulation state of the student model. Based on the predicted insulation state of the teacher model and the labeled training dataset, calculate the first loss function value of the preset distillation loss function; and based on the predicted insulation state of the student model and the labeled training dataset, calculate the second loss function value of the preset student model loss function. The total loss function value is calculated based on the first loss function value, the second loss function value, and the preset loss function weights. When the total loss function converges, the trained student model is obtained; and the trained student model is used as the trained cable insulation state evaluation model. The teacher model to be trained and the student model to be trained are convolutional neural network models with the same structure.

5. The online assessment method for the insulation condition of high-voltage cables as described in claim 1, characterized in that, Also includes: The dielectric loss data is filtered and selected to obtain filtered dielectric loss data; The filtered dielectric loss data are normalized to obtain the final dielectric loss data.

6. The online assessment method for the insulation condition of high-voltage cables as described in claim 1, characterized in that, The online insulation condition assessment system for high-voltage cables also includes a dielectric loss monitoring synchronization module; the dielectric loss monitoring synchronization module includes an optical fiber and a photoelectric converter; the optical fiber is laid on the surface of the cable to be assessed. The optical fiber is used to transmit the optical signal injected by the control center to the photoelectric converter; The photoelectric converter is used to convert optical signals into electrical signals and transmit the electrical signals to the dielectric loss signal monitoring module, so that the dielectric loss signal monitoring module can synchronously collect dielectric loss data of the cable to be evaluated according to the time difference of the electrical signals.

7. The online assessment method for the insulation condition of high-voltage cables as described in claim 6, characterized in that, Also includes: If the insulation condition assessment result is good, continue to monitor the dielectric loss data of the cable to be assessed; If the insulation condition assessment indicates a minor fault, increase the frequency of optical signal injection; If the insulation condition assessment result is a moderate fault, local repair measures shall be taken for the cable to be assessed. If the insulation condition assessment result is a severe fault, disconnect the power supply circuit of the cable being assessed and replace the cable.

8. An online evaluation system for the insulation condition of high-voltage cables, characterized in that, include: Control center, low-frequency signal injection source module, and dielectric loss signal monitoring module; The low-frequency signal injection source module is used to receive low-frequency signal injection instructions from the control center, and after receiving the low-frequency signal injection instructions from the control center, inject a test signal of a preset frequency into the cable to be evaluated. The dielectric loss signal monitoring module is used to collect dielectric loss data of the cable to be evaluated after injecting a test signal of a preset frequency into the cable to be evaluated. The control center is used to send low-frequency signal injection commands to the low-frequency signal injection source module; Real-time acquisition of dielectric loss data for the cable to be evaluated; Dynamic modeling is performed based on the dielectric loss data to obtain a hidden state sequence; and sequence feature extraction is performed based on the hidden state sequence to obtain dynamic change rate, trend features, and key point features. Convolutional feature extraction is performed based on the hidden state sequence and the preset temporal convolutional network model to obtain local features for characterizing short-term changes in dielectric loss data, long-term dependency features for characterizing long-term trends in dielectric loss data, and multi-scale features. The feature matrix is ​​obtained by integrating dynamic change rate, trend features, key point features, local features, long-term dependence features, and multi-scale features. The feature matrix is ​​input into a preset cable insulation condition assessment model to perform insulation condition assessment, and the insulation condition assessment result of the cable to be assessed is obtained.

9. A terminal device, characterized in that, The device includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements an online assessment method for the insulation condition of a high-voltage cable as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform an online assessment method for the insulation status of a high-voltage cable as described in any one of claims 1 to 7.

Images