A method, medium and system for evaluating initial breakdown damage of a dry-type air-core reactor caused by a turn-to-turn short circuit fault

Abstract

The application provides a kind of dry type hollow reactor interturn short circuit fault initial electric breakdown damage evaluation method, medium and system, belong to dry type hollow reactor technical field, including: respectively at T1, T0 and T2 Three times of winding infrared image and instantaneous magnetic field data are obtained, and temperature field and magnetic field distribution model are established;Using the pre-trained interturn short circuit fault positioning model, determine the fault area;Calculate the temperature and magnetic field change of the fault area in T1 to T0 and T0 to T1 Two time periods, form the corresponding change matrix;These change matrix is input into the pre-trained electric breakdown damage evaluation model, and the electric breakdown damage degree of reactor is obtained.The electric breakdown damage evaluation method of the application gives full play to the potential of temperature and magnetic field data, realizes the accurate positioning of interturn short circuit fault area and the quantitative evaluation of damage degree through the application of fine thermal magnetic field analysis and deep learning technology.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of dry-type air-core reactors, and in particular, relates to a method for evaluating initial electrical breakdown damage of a dry-type air-core reactor caused by inter-turn short circuit faults, a medium and a system. BACKGROUND

[0002] As an important reactive power compensation device in power systems, dry-type air-core reactors play a key role in grid operation. Compared with oil-immersed reactors, dry-type air-core reactors have advantages such as small size, light weight, low operating loss, and environmental friendliness, and are widely used in urban power grids, wind farms, and photovoltaic power stations. However, due to the fragile insulation structure of dry-type reactors, inter-turn short circuit faults between windings are prone to occur during long-term operation, seriously affecting the safe and stable operation of the power grid.

[0003] The formation mechanism of inter-turn short circuit faults usually includes two stages: the initial stage and the development stage. In the initial stage, the local winding temperature rise and magnetic field distortion caused by the voltage difference between windings will cause local electrical breakdown damage to the insulation material. Although this electrical breakdown damage does not cause the overall failure of the reactor, it will cause a certain degree of change in the reactance value. If this fault is not discovered and addressed in a timely manner, the local breakdown will further spread over time, eventually leading to the overall insulation failure of the reactor and causing serious faults. Therefore, accurately evaluating the degree of electrical breakdown damage in the initial stage of inter-turn short circuit faults is crucial for improving the reliability and economy of the reactor.

[0004] Currently, the main technical means for diagnosing inter-turn short circuit faults include:

[0005] (1) Temperature rise detection method: By measuring the temperature rise of the reactor winding, the heat dissipation condition of the winding can be indirectly reflected, thereby inferring the possibility of fault occurrence. However, this method can only give a rough fault judgment and is difficult to locate the fault area and evaluate the damage degree.

[0006] (2) Reactance value detection method: By measuring the series reactance value of the reactor, the inter-turn short circuit condition of the winding can be reflected. However, the change in the reactance value usually occurs in the later stage of fault development, making it difficult to achieve early warning.

[0007] (3) Partial discharge detection method: Using partial discharge detection technology, local defects in the insulation system can be found. However, this method requires special detection equipment, and the detection accuracy is affected by environmental interference.

[0008] (4) Electromagnetic field detection method: Using a magnetic probe to measure the magnetic field distribution around the reactor winding, the fault condition of the winding can be analyzed. However, this method can only give a qualitative judgment of the fault and lacks quantitative damage evaluation basis.

[0009] In summary, the existing fault diagnosis technology still has deficiencies in early warning and damage quantitative evaluation of inter-turn short circuit fault, and it is difficult to provide strong support for the state monitoring and preventive maintenance of the reactor. SUMMARY

[0010] Therefore, the present application provides a method, medium and system for evaluating initial electrical breakdown damage of inter-turn short circuit fault of dry-type air-core reactor, which can solve the technical problem of the existing fault diagnosis technology still having deficiencies in early warning and damage quantitative evaluation of inter-turn short circuit fault.

[0011] The present application is implemented as follows:

[0012] The first aspect of the present application provides a method for evaluating initial electrical breakdown damage of inter-turn short circuit fault of dry-type air-core reactor, which comprises the following steps:

[0013] S10, acquiring, at a large current moment T0 of the reactor, a winding infrared image collected by an infrared thermal imaging device and instantaneous magnetic field data of a plurality of magnetic probes distributed around the winding, and recording them as T0 infrared image and T0 instantaneous magnetic field group at moment T0, respectively;

[0014] S20, acquiring winding infrared images and instantaneous voltage groups at each of T1 moment and T2 moment of the reactor, wherein T1=T0-Tz and T2=T0+Tz, and recording them as T1 infrared image and T1 instantaneous magnetic field group and T2 infrared image and T2 instantaneous magnetic field group, respectively, and the Tz is the period of input current;

[0015] S30, modeling the static thermal distribution and magnetic field distribution of the reactor winding by using the infrared image and the instantaneous magnetic field group at each of T1 moment, T0 moment and T2 moment, respectively, to obtain T1 temperature field, T1 magnetic field, T0 temperature field, T0 magnetic field and T2 temperature field and T2 magnetic field at the corresponding moment, respectively;

[0016] S40, determining the occurrence area of the inter-turn short circuit fault by using a pre-trained inter-turn short circuit fault positioning model according to the T1 temperature field, the T1 magnetic field, the T0 temperature field, the T0 magnetic field, the T2 temperature field and the T2 magnetic field, and recording it as a fault area;

[0017] S50, acquiring temperature changes and magnetic field changes of the fault area at T1 moment, T0 moment and T2 moment, respectively, to form temperature change matrix and magnetic change matrix between T1 moment and T0 moment and temperature change matrix and magnetic change matrix between T0 moment and T1 moment;

[0018] S60, inputting the temperature change matrix and the magnetic change matrix between the T1 moment and the T0 moment and the temperature change matrix and the magnetic change matrix from T0 to T1, obtaining the electric breakdown damage degree and outputting.

[0019] The step S10 specifically comprises: acquiring, at the large-current moment T0, temperature field data of the winding using an infrared thermal imaging device, denoted as a T0 infrared image; meanwhile, a plurality of magnetic probes are distributed around the winding to acquire instantaneous magnetic field data at the T0 moment, denoted as a T0 instantaneous magnetic field group. The infrared thermal imaging device can non-contact measure the temperature distribution on the surface of the winding, while the magnetic probes can acquire the magnetic field change around the winding in real time. The two types of data can completely reflect the heat and magnetic field characteristics of the reactor winding under the large-current condition.

[0020] The step S20 specifically comprises: on the basis of the T0 moment, acquiring the winding temperature field data and the instantaneous voltage data at two moments, T1=T0-Tz and T2=T0+Tz. Wherein, Tz represents the period of the input current. The T1 temperature field data and the instantaneous voltage group are denoted as a T1 infrared image and a T1 instantaneous magnetic field group, and the T2 temperature field data and the instantaneous voltage group are denoted as a T2 infrared image and a T2 instantaneous magnetic field group. By acquiring the temperature and magnetic field data at the T1 and T2 moments, the temperature and magnetic field change trend of the reactor winding before and after the large current can be observed, which provides an important basis for subsequent damage analysis.

[0021] The step S30 specifically comprises: respectively using the infrared image data and the instantaneous magnetic field data at the T1, T0 and T2 moments to construct temperature field and magnetic field distribution models. Specifically, the T1 temperature field model T1 temperature field and the T1 magnetic field model T1 magnetic field are established according to the T1 infrared image and the T1 instantaneous magnetic field group data by using finite element analysis or contour interpolation method; similarly, the T0 temperature field model T0 temperature field and the T0 magnetic field model T0 magnetic field are established according to the T0 infrared image and the T0 instantaneous magnetic field group data; and the T2 temperature field model T2 temperature field and the T2 magnetic field model T2 magnetic field are established according to the T2 infrared image and the T2 instantaneous magnetic field group data. By constructing the temperature and magnetic field distribution models at the three moments, the heat and magnetic field change of the reactor winding under the large-current condition can be analyzed more finely.

[0022] The step S40 specifically comprises: inputting the data of six channels of T1 temperature field, T1 magnetic field, T0 temperature field, T0 magnetic field, and T2 temperature field and T2 magnetic field into the pre-trained inter-turn short circuit fault positioning model, and using the model to position and identify the fault area. The inter-turn short circuit fault positioning model adopts a multi-layer tower convolutional neural network (Tower CNN) structure, including an input layer, an encoder part, a decoder part and an output layer, can effectively extract multi-scale features from the temperature and magnetic field data at multiple times, and accurately position the area where the fault occurs.

[0023] The step S50 specifically comprises: calculating the temperature change and the magnetic field change of the area at T1 to T0 and T0 to T1 times according to the fault area identified in the step S40, and forming two temperature change matrices and two magnetic change matrices. Specifically, the temperature values of the fault area in the T1 temperature field model and the T0 temperature field model are recorded as T1(x, y) and T0(x, y) respectively, and the temperature change matrix ΔT1-0(x, y) at T1 to T0 times is calculated as T0(x, y)-T1(x, y); the magnetic field values of the fault area in the T1 magnetic field model and the T0 magnetic field model are recorded as H1(x, y) and H0(x, y) respectively, and the magnetic field change matrix ΔH1-0(x, y) at T1 to T0 times is calculated as H0(x, y)-H1(x, y); similarly, the temperature change matrix ΔT0-1(x, y) at T0 to T1 times is calculated as T1(x, y)-T0(x, y), and the magnetic field change matrix ΔH0-1(x, y) at T0 to T1 times is calculated as H1(x, y)-H0(x, y).

[0024] The step S60 specifically comprises: inputting the temperature change matrices and the magnetic change matrices at T1 to T0 and T0 to T1 times calculated in the step S50 into the pre-trained electric breakdown damage evaluation model, and outputting the electric breakdown damage degree of the reactor winding. The electric breakdown damage evaluation model adopts a fusion evaluation network structure, including a temperature change damage evaluation subnetwork, a magnetic change damage evaluation subnetwork and a fusion evaluation subnetwork, can make full use of the information of temperature and magnetic field, and give more accurate electric breakdown damage evaluation results.

[0025] The training of the inter-turn short circuit fault positioning model specifically comprises four steps of data collection, data preprocessing, model training and model evaluation. The data collection comprises collecting a large amount of temperature field and magnetic field data of the dry-type air-core reactor under normal working and inter-turn short circuit fault conditions, and corresponding fault area labeling information; the data preprocessing comprises normalizing the data and generating a binary segmentation graph; the model training adopts a cross-entropy loss function, and uses a back propagation algorithm to optimize the Tower CNN network parameters; the model evaluation adopts an independent test set to calculate the accuracy and other indicators, and further adjusts the network structure and hyperparameters according to the results.

[0026] The training of the electric breakdown damage evaluation model specifically comprises four steps of data collection, data preprocessing, model training and model evaluation. The data collection comprises obtaining temperature variation and magnetic field variation data of the reactor winding at different times and corresponding electric breakdown damage degrees. The data preprocessing comprises normalizing the data to form temperature variation matrices and magnetic variation matrices. The model training adopts a mean square error loss function and optimizes fusion evaluation network parameters by using a back propagation algorithm. The model evaluation calculates indexes such as mean square error by using an independent test set, and further adjusts network structure and hyperparameters according to the results.

[0027] The inter-turn short circuit fault positioning model adopts a multi-layer tower convolutional neural network.

[0028] The specific structure of the inter-turn short circuit fault positioning model is as follows:

[0029] The input layer receives inputs of six channels of T1 temperature field, T1 magnetic field, T0 temperature field, T0 magnetic field, T2 temperature field and T2 magnetic field.

[0030] The encoder part comprises three convolutional blocks, each of which is composed of a convolutional layer (kernel_size=3, padding=1), a batch normalization layer and a ReLU activation layer. The convolutional blocks are down-sampled by using a 2x2 maximum pooling layer.

[0031] The decoder part comprises three deconvolutional blocks, each of which is composed of a transposed convolutional layer (kernel_size=3, padding=1), a batch normalization layer and a ReLU activation layer. The deconvolutional blocks are up-sampled by using a 2x2 up-sampling operation.

[0032] The output layer outputs the occurrence area of the inter-turn short circuit fault, and uses a Sigmoid activation function.

[0033] The electric breakdown damage evaluation model adopts a fusion evaluation network structure comprising a temperature variation damage evaluation sub-network, a magnetic variation damage evaluation sub-network and a fusion evaluation sub-network.

[0034] Further, the temperature variation damage evaluation sub-network is configured to receive temperature variation matrices of T1 to T0 and T0 to T1, and output a temperature variation damage evaluation result.

[0035] Further, the magnetic variation damage evaluation sub-network is configured to receive magnetic variation matrices of T1 to T0 and T0 to T1, and output a magnetic variation damage evaluation result.

[0036] Further, the fusion evaluation sub-network is configured to receive the temperature variation damage evaluation result and the magnetic variation damage evaluation result, and output a final electric breakdown damage degree.

[0037] The infrared thermal imaging device is used to non-contact measure the temperature distribution of the winding surface, and the magnetic probe is used to collect the magnetic field change around the winding in real time.

[0038] The large current refers to a current flowing through the reactor reaching a current value of 4 times or more than any number of the average current of the reactor, preferably 16 times.

[0039] Specifically: the large current may be a transient or short-time large current caused by system failure, load mutation, short circuit, etc.

[0040] 1. The large current moment (T0) is the key time point of the entire evaluation process. The current value at this moment is much higher than the normal working current of the reactor.

[0041] 2. The large current may be a transient or short-time large current caused by system failure, load mutation, short circuit, etc.

[0042] 3. This large current moment is selected as the core time point for evaluation because:

[0043] The large current causes the temperature of the reactor to rise rapidly

[0044] The large current produces strong magnetic field changes

[0045] These changes help better detect and locate potential turn-to-turn short circuit faults

[0046] 4. The method uses the moments before and after T0 (T1 and T2) for comparative analysis to capture the temperature and magnetic field changes caused by the large current.

[0047] The second aspect of the application provides a computer readable storage medium, wherein the computer readable storage medium stores program instructions, and the program instructions are used to execute the above-mentioned preliminary electrical breakdown damage evaluation method for turn-to-turn short circuit faults of a dry-type air-core reactor.

[0048] The third aspect of the application provides a preliminary electrical breakdown damage evaluation system for turn-to-turn short circuit faults of a dry-type air-core reactor, which comprises the above-mentioned computer readable storage medium.

[0049] Compared with the prior art, the preliminary electrical breakdown damage evaluation method, medium and system for turn-to-turn short circuit faults of a dry-type air-core reactor provided by the application have the following beneficial effects:

[0050] (1) The region where the turn-to-turn short circuit fault occurs can be accurately identified, providing reliable positioning basis for subsequent maintenance. The existing temperature rise detection method, reactance value detection method, etc. can only give rough fault judgment and cannot accurately locate the fault region.

[0051] (2) It can quantitatively assess the degree of electrical breakdown damage to reactor windings, providing an important reference for maintenance decisions. However, existing methods such as electromagnetic field detection can only provide qualitative fault analysis and lack quantitative damage assessment basis.

[0052] (3) The temperature and magnetic field monitoring method used does not require contact with the reactor body and does not interfere with the operation of the equipment. In contrast, methods such as partial discharge detection require special detection equipment and are easily affected by environmental interference.

[0053] (4) The fault diagnosis model based on deep learning has strong generalization ability and can be applied to different types of dry reactors. Traditional empirical models usually need to be corrected and optimized for specific equipment.

[0054] In summary, the electrical breakdown damage assessment method proposed in this invention can provide strong support for condition monitoring and preventive maintenance of dry-type air-core reactors, and solves the technical problem that existing fault diagnosis technologies still have shortcomings in early warning and quantitative damage assessment of inter-turn short-circuit faults. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 A flowchart of the method provided by the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0058] like Figure 1 The diagram shown is a flowchart of an initial electrical breakdown damage assessment method for inter-turn short-circuit faults in a dry-type air-core reactor provided by this invention. The method includes the following steps:

[0059] S10. When the reactor is at its high current T0, the infrared image of the winding acquired by the infrared thermal imaging device and the instantaneous magnetic field data of multiple magnetic probes distributed around the winding are recorded as the T0 infrared image and the T0 instantaneous magnetic field group at time T0, respectively.

[0060] S20, obtaining winding infrared images and instantaneous voltage groups of the reactor at each of T1 and T2, where T1=T0-Tz and T2=T0+Tz, respectively denoted as T1 infrared image and T1 instantaneous magnetic field group and T2 infrared image and T2 instantaneous magnetic field group, and the Tz is a period of the input current;

[0061] S30, respectively using the infrared image and the instantaneous magnetic field group at each of T1, T0 and T2, to model the static thermal distribution and the magnetic field distribution of the reactor winding, to obtain T1 temperature field, T1 magnetic field, T0 temperature field, T0 magnetic field and T2 temperature field and T2 magnetic field at the corresponding time;

[0062] S40, according to the T1 temperature field, T1 magnetic field, T0 temperature field, T0 magnetic field and T2 temperature field and T2 magnetic field, using a pre-trained inter-turn short circuit fault positioning model, to determine the occurrence area of the inter-turn short circuit fault, denoted as the fault area;

[0063] S50, obtaining the temperature change and the magnetic field change of the fault area at T1, T0 and T2 respectively, to form the temperature change matrix and the magnetic change matrix between T1 and T0 and the temperature change matrix and the magnetic change matrix between T0 and T1;

[0064] S60, using a pre-trained electrical breakdown damage evaluation model, inputting the temperature change matrix and the magnetic change matrix between T1 and T0 and the temperature change matrix and the magnetic change matrix between T0 and T1, to obtain the electrical breakdown damage degree and output.

[0065] The specific embodiments of the above steps are described in detail as follows:

[0066] The specific embodiments of step S10 are as follows:

[0067] At the large current time T0, the temperature field data of the winding is collected using an infrared thermal imaging device, denoted as T0 infrared image. At the same time, a plurality of magnetic probes are distributed around the winding to collect the instantaneous magnetic field data at T0, denoted as T0 instantaneous magnetic field group. The infrared thermal imaging device can non-contact measure the temperature distribution of the winding surface, and the magnetic probe can real-time collect the magnetic field change around the winding. These two types of data can completely reflect the heat and magnetic field characteristics of the reactor winding under large current conditions.

[0068] The specific embodiments of step S20 are as follows:

[0069] On the basis of T0, winding temperature field data and instantaneous voltage data at T1 = T0-Tz and T2 = T0+Tz are obtained respectively. Wherein, Tz represents the period of input current. T1 temperature field data and instantaneous voltage group are recorded as T1 infrared image and T1 instantaneous magnetic field group, and T2 temperature field data and instantaneous voltage group are recorded as T2 infrared image and T2 instantaneous magnetic field group. By collecting the temperature and magnetic field data at T1 and T2, the temperature and magnetic field change trend of the reactor winding before and after the large current can be observed, which provides an important basis for subsequent damage analysis.

[0070] The specific implementation of step S30 is as follows:

[0071] The temperature field and magnetic field distribution model is constructed by using the infrared image data and instantaneous magnetic field data at T1, T0 and T2 respectively. The specific method is as follows:

[0072] (1) According to T1 infrared image and T1 instantaneous magnetic field group data, T1 temperature field model T1 temperature field and magnetic field model T1 magnetic field at T1 are established by using finite element analysis or contour interpolation method.

[0073] (2) Similarly, according to T0 infrared image and T0 instantaneous magnetic field group data, T0 temperature field model T0 temperature field and magnetic field model T0 magnetic field at T0 are established.

[0074] (3) According to T2 infrared image and T2 instantaneous magnetic field group data, T2 temperature field model T2 temperature field and magnetic field model T2 magnetic field at T2 are established.

[0075] By constructing the temperature and magnetic field distribution model at the three moments, the heat and magnetic field change of the reactor winding under the condition of large current can be analyzed more finely.

[0076] The specific implementation of step S40 is as follows:

[0077] The pre-trained inter-turn short circuit fault positioning model is used to input the data of T1 temperature field, T1 magnetic field, T0 temperature field, T0 magnetic field, and T2 temperature field and T2 magnetic field, and the model is used for positioning and identifying the fault area.

[0078] The inter-turn short circuit fault positioning model adopts a multi-layer tower convolutional neural network (Tower CNN) structure, which specifically includes:

[0079] Input layer: receiving 6 channels of temperature and magnetic field data, size HxW.

[0080] Encoder part: containing 3 convolution blocks, each convolution block is composed of convolution layer (kernel_size = 3, padding = 1), batch normalization layer and ReLU activation layer, and 2x2 maximum pooling layer is used for down sampling between convolution blocks.

[0081] Decoder: contains 3 deconvolutional blocks, each of which consists of a transpose convolutional layer (kernel_size = 3, padding = 1), a batch normalization layer and a ReLU activation layer, and 2x2 up-sampling operation is used between deconvolutional blocks.

[0082] Output layer: outputs an HxW segmentation map, which identifies the region where the inter-turn short circuit fault occurs using a Sigmoid activation function.

[0083] The Tower CNN model can effectively extract multi-scale features from multi-time temperature and magnetic field data, and accurately locate the region where the fault occurs.

[0084] The specific implementation of step S50 is as follows:

[0085] According to the fault region identified in step S40, the temperature change and magnetic field change of the region at T1 to T0 and T0 to T1 are calculated respectively, forming two temperature change matrices and two magnetic change matrices.

[0086] The specific calculation process is as follows:

[0087] (1) Record the temperature values of the fault region in the T1 temperature field model and the T0 temperature field model as T1(x, y) and T0(x, y) respectively, and then calculate the temperature change matrix ΔT1-0(x, y) = T0(x, y) - T1(x, y) at T1 to T0.

[0088] (2) Record the magnetic field values of the fault region in the T1 magnetic field model and the T0 magnetic field model as H1(x, y) and H0(x, y) respectively, and then calculate the magnetic field change matrix ΔH1-0(x, y) = H0(x, y) - H1(x, y) at T1 to T0.

[0089] (3) Similarly, calculate the temperature change matrix ΔT0-1(x, y) = T1(x, y) - T0(x, y) and the magnetic field change matrix ΔH0-1(x, y) = H1(x, y) - H0(x, y) at T0 to T1.

[0090] By calculating these temperature and magnetic field change matrices, the heat and magnetic field change characteristics of the inter-turn short circuit fault region at different times can be reflected, providing key inputs for subsequent electrical breakdown damage assessment.

[0091] The specific implementation of step S60 is as follows:

[0092] Use the pre-trained electrical breakdown damage assessment model to input the temperature change matrices and magnetic change matrices at T1 to T0 and T0 to T1 calculated in step S50, and output the electrical breakdown damage degree of the reactor winding.

[0093] The electric breakdown damage evaluation model adopts a fusion evaluation network structure, including:

[0094] The temperature change damage evaluation sub-network:

[0095] Input layer: receives the temperature change matrix from T1 to T0 and from T0 to T1.

[0096] 3 residual modules, each module contains 2 convolution layers (kernel_size = 3, padding = 1), batch normalization layer and ReLU activation layer, residual connection is used to improve performance.

[0097] Fully connected layer: outputs the temperature change damage evaluation result.

[0098] The magnetic change damage evaluation sub-network:

[0099] Input layer: receives the magnetic change matrix from T1 to T0 and from T0 to T1.

[0100] The network structure is similar to the temperature change damage evaluation sub-network, containing 3 residual modules.

[0101] Fully connected layer: outputs the magnetic change damage evaluation result.

[0102] The fusion evaluation sub-network:

[0103] Input layer: receives the temperature change damage evaluation result and the magnetic change damage evaluation result.

[0104] 2 fully connected layers, using ReLU activation function in the middle.

[0105] Output layer: outputs the final electric breakdown damage degree.

[0106] This fusion evaluation network can fully utilize the information of temperature change and magnetic field change, comprehensively analyze the electric breakdown damage degree of the reactor winding, and provide reliable basis for fault diagnosis.

[0107] For the training of the turn-to-turn short circuit fault positioning model, the specific steps are as follows:

[0108] (1) Data collection: collect a large amount of temperature field and magnetic field data of dry-type air-core reactor under normal working and turn-to-turn short circuit fault conditions, and record the fault area label information corresponding to each group of data.

[0109] (2) Data preprocessing: normalize the collected temperature field and magnetic field data, map them to the range of 0-1. According to the label information, generate the corresponding binary segmentation map for each group of data, and identify the fault area. Pack the temperature field, magnetic field data and segmentation map into training samples.

[0110] (3) Model training: The pre-processed training sample set is input into the Tower CNN network, and the cross-entropy loss function is used to optimize the network parameters using the back propagation algorithm. The validation set is used to monitor the model performance, and the training is stopped in time and the optimal model is saved.

[0111] (4) Model evaluation: The trained model is evaluated using an independent test set, and accuracy, recall rate and other indicators are calculated. Based on the evaluation results, the network structure and hyperparameters are further adjusted until the performance requirements are met.

[0112] For the training of the electric breakdown damage evaluation model, the specific steps are as follows:

[0113] (1) Data collection: Under the condition of simulated inter-turn short circuit fault, the temperature change and magnetic field change data of the reactor winding at T0-Tz, T0, T0+Tz are obtained, and the corresponding electric breakdown damage degree at these times is measured and recorded.

[0114] (2) Data preprocessing: Normalize the temperature change and magnetic field change data to form the temperature change matrix and magnetic change matrix from T1 to T0 and T0 to T1. Pack the temperature change matrix, magnetic change matrix and corresponding electric breakdown damage degree label into training samples.

[0115] (3) Model training: Construct a fusion evaluation network, including a temperature change damage evaluation sub-network, a magnetic change damage evaluation sub-network and a fusion evaluation sub-network. Use the mean square error loss function to optimize the network parameters using the back propagation algorithm. Use the validation set to monitor the model performance, and stop the training in time and save the optimal model.

[0116] (4) Model evaluation: The trained model is evaluated using an independent test set, and the mean square error, correlation coefficient and other indicators are calculated. Based on the evaluation results, the network structure and hyperparameters are further adjusted until the accuracy requirements are met.

[0117] The determination method of the electric breakdown damage degree label is based on the actual observation or accurate calculation of the test method. One optional specific determination step is as follows:

[0118] 1. Partial discharge measurement:

[0119] Use a partial discharge detection device to measure the partial discharge level of the reactor winding. Partial discharge is an important indicator of electric breakdown damage.

[0120] 2. Calculate the electric breakdown damage degree:

[0121] Based on the partial discharge measurement results, the electric breakdown damage degree can be calculated using the following formula:

[0122]

[0123] In the formula: D eb is the electric breakdown damage degree, ranging from 0 to 1; Q m is the measured partial discharge quantity (pC); Q0 is the background partial discharge quantity (pC) in the normal state; Q c is the critical partial discharge quantity, i.e. the discharge quantity (pC) that causes complete breakdown of insulation;

[0124] 3. Determine the parameters: Q0 can be obtained by measuring new or undamaged reactors; Q c can be determined by historical data, manufacturer specifications or accelerated aging tests;

[0125] 4. Data label method process: measure the partial discharge of the reactor under simulated inter-turn short circuit fault conditions; record the partial discharge quantity Q m at different times and different fault degrees; calculate the corresponding electric breakdown damage degree D eb using the above formula.

[0126] Through the above steps, the measured data can be fully utilized to train excellent inter-turn short circuit fault positioning models and electric breakdown damage evaluation models, providing reliable technical support for state monitoring and fault diagnosis of dry-type air-core reactors.

[0127] In the early stage of inter-turn short circuit fault, local temperature rise of the winding and distortion of the magnetic field distribution will occur, thereby causing electric breakdown damage of the insulation material. Although this electric breakdown damage does not cause the overall failure of the reactor, it will cause a certain degree of change in the reactance value. Specifically, the increase in the resistance of the winding material caused by the temperature rise will cause the reactance value to rise, and the change in the magnetic flux chain will cause the reactance value to drop. The combined effect of temperature change and magnetic field change determines the trend of the reactance value change of the reactor in the early stage of inter-turn short circuit fault. Therefore, the present application proposes an electric breakdown damage evaluation method based on temperature and magnetic field changes, which can accurately evaluate the electric breakdown damage degree of the winding of the reactor by analyzing and fusing the change laws of temperature and magnetic field through a neural network model, thereby providing technical support for timely discovery and processing of faults.

[0128] In the specific implementation process of the present application, the temperature and magnetic field data at T0 in step S10 are obtained by using an infrared thermal imaging device and a magnetic probe to measure the temperature field and the magnetic field of the winding, respectively, which can reflect the heat and magnetic field characteristics of the reactor under large current conditions. The temperature and magnetic field data at T1 and T2 in step S20 are obtained, which can observe the temperature and magnetic field changes of the winding of the reactor before and after the large current, thereby providing important data for subsequent damage analysis.

[0129] In step S30, the temperature field and magnetic field distribution models at T1, T0 and T2 are constructed using the obtained temperature and magnetic field data, which can more finely analyze the heat and magnetic field variation law of the reactor winding under large current conditions. In step S40, the pre-trained Tower CNN model is used to analyze the temperature and magnetic field data, which can accurately locate the region where the inter-turn short circuit fault occurs.

[0130] In step S50, the temperature change matrix and magnetic field change matrix at different times are calculated for the fault region identified in step S40. These temperature and magnetic field change characteristics provide key input data for subsequent electrical breakdown damage assessment.

[0131] Finally, in step S60, a fusion evaluation network is used to comprehensively analyze the temperature and magnetic field changes, and output the electrical breakdown damage degree of the reactor winding. The network consists of a temperature change damage evaluation sub-network, a magnetic change damage evaluation sub-network and a fusion evaluation sub-network, which can fully utilize the information of temperature and magnetic field to give more accurate electrical breakdown damage assessment results.

[0132] The second aspect of the present application provides a computer readable storage medium, wherein the computer readable storage medium stores program instructions, and the program instructions are used to execute the above-mentioned method for initial electrical breakdown damage assessment of dry-type air-core reactor inter-turn short circuit fault.

[0133] The third aspect of the present application provides a system for initial electrical breakdown damage assessment of dry-type air-core reactor inter-turn short circuit fault, which comprises the above-mentioned computer readable storage medium.

[0134] In order to better understand and implement the present application, the following provides a specific embodiment 1 of the present application in a computer readable storage medium or a computer system for a computer program: the specific implementation of step S10 is as follows:

[0135] At time T0 when the reactor is in a large current working state, an infrared thermal imaging device is used to collect the temperature field data of the winding, denoted as T0 infrared image. At the same time, a plurality of magnetic probes are distributed around the winding to collect the instantaneous magnetic field data at T0, denoted as T0 instantaneous magnetic field group.

[0136] The infrared thermal imaging device can non-contact measure the temperature distribution of the winding surface, i.e. obtain two-dimensional temperature field data T0(x, y), wherein x and y represent the horizontal and vertical coordinates of the winding plane respectively. The magnetic probe can real-time collect the instantaneous magnetic field data H0(x, y) around the winding, wherein H0 represents the magnetic field intensity at T0. These two types of data can completely reflect the heat and magnetic field characteristics of the reactor winding under large current conditions.

[0137] The specific implementation of step S20 is as follows:

[0138] On the basis of T0 moment, winding temperature field data and instantaneous voltage data at T1 = T0-Tz and T2 = T0+Tz moments are obtained respectively. Wherein, Tz represents the period of input current. T1 temperature field data and instantaneous voltage group are recorded as T1 infrared image and T1 instantaneous magnetic field group, and T2 temperature field data and instantaneous voltage group are recorded as T2 infrared image and T2 instantaneous magnetic field group.

[0139] Specifically, the temperature field data at T1 moment is T1(x,y), and the instantaneous magnetic field data is H1(x,y); the temperature field data at T2 moment is T2(x,y), and the instantaneous magnetic field data is H2(x,y). By collecting the temperature and magnetic field data at T1 and T2 moments, the temperature and magnetic field change trend of the reactor winding before and after the large current can be observed, which provides an important basis for subsequent damage analysis.

[0140] The specific implementation of step S30 is as follows:

[0141] The temperature field and magnetic field distribution models are constructed by using the infrared image data and instantaneous magnetic field data at T1, T0 and T2 moments respectively.

[0142] For T1 moment, the temperature field model T1(x,y) and the magnetic field model H1(x,y) can be established according to T1(x,y) and H1(x,y) data by using finite element analysis or contour interpolation method. Similarly, the temperature field model T0(x,y) and the magnetic field model H0(x,y) at T0 moment are established according to T0(x,y) and H0(x,y) data; the temperature field model T2(x,y) and the magnetic field model H2(x,y) at T2 moment are established according to T2(x,y) and H2(x,y) data.

[0143] By constructing the temperature and magnetic field distribution models at the three moments, the heat and magnetic field change of the reactor winding under the condition of large current can be analyzed more finely.

[0144] The specific implementation of step S40 is as follows:

[0145] The pre-trained turn-to-turn short circuit fault positioning model is used to input the data of six channels of T1(x,y), H1(x,y), T0(x,y), H0(x,y), T2(x,y) and H2(x,y), and the model is used for positioning and identifying the fault area.

[0146] The turn-to-turn short circuit fault positioning model adopts a multi-layer tower convolutional neural network (Tower CNN) structure, which can be represented as:

[0147] Input layer:

[0148] Encoder part:

[0149] Convolution block 1: C1 = ReLU(BN(Conv(X, W1, b1)))

[0150] Pooling layer 1: P1 = MaxPool(C1)

[0151] Convolution block 2: C2 = ReLU(BN(Conv(P1, W2, b2)))

[0152] Pooling layer 2: P2 = MaxPool(C2)

[0153] Convolution block 3: C3 = ReLU(BN(Conv(P2, W3, b3)))

[0154] Decoder part:

[0155] Transposed convolution block 1: T1 = ReLU(BN(TransposeConv(C3, W4, b4)))

[0156] Up-sampling 1: U1 = Upsample(T1)

[0157] Transposed convolution block 2: T2 = ReLU(BN(TransposeConv(U1, W5, b5)))

[0158] Up-sampling 2: U2 = Upsample(T2)

[0159] Transposed convolution block 3: T3 = ReLU(BN(TransposeConv(U2, W6, b6)))

[0160] Output layer: T = Sigmoid(T3)

[0161] where W i and b i are the weights and biases of the i-th convolutional / transposed convolutional layer, BN represents the batch normalization layer, and ReLU is the activation function. Y is a HxW segmentation map that identifies the region where the interturn short circuit fault occurs using the Sigmoid function.

[0162] The Tower CNN model can effectively extract multi-scale features from multi-time temperature and magnetic field data and accurately locate the region where the fault occurs.

[0163] The specific implementation of step S50 is as follows:

[0164] According to the fault region identified in step S40, the temperature change and the magnetic field change of the region at T1 to T0 and T0 to T1 are calculated respectively, forming two temperature change matrices and two magnetic change matrices.

[0165] The specific calculation process is as follows:

[0166] (1) Record the temperature values of the fault area in the T1 temperature field model T1(x, y) and the T0 temperature field model T0(x, y) as T1(x, y) and T0(x, y) respectively. Then calculate the temperature change matrix from T1 to T0:

[0167] ΔT 1-0 (x, y) = T0(x, y) - T1(x, y)

[0168] (2) Record the magnetic field values of the fault area in the T1 magnetic field model H1(x, y) and the T0 magnetic field model H0(x, y) as H1(x, y) and H0(x, y) respectively. Then calculate the magnetic field change matrix from T1 to T0:

[0169] ΔH 1-0 (x, y) = H0(x, y) - H1(x, y)

[0170] (3) Similarly, calculate the temperature change matrix from T0 to T1:

[0171] ΔT 0-1 (x, y) = T1(x, y) - T0(x, y)

[0172] And the magnetic field change matrix:

[0173] ΔH 0-1 (x, y) = H1(x, y) - H0(x, y)

[0174] By calculating these temperature and magnetic field change matrices, the heat and magnetic field change characteristics of the inter-turn short circuit fault area at different times can be reflected, providing key inputs for subsequent electrical breakdown damage assessment.

[0175] The specific implementation of step S60 is as follows:

[0176] Using the pre-trained electrical breakdown damage assessment model, input ΔT 1-0 , ΔH 1-0 , ΔT 0-1 and ΔH 0-1 calculated in step S50, and output the electrical breakdown damage degree of the reactor winding.

[0177] The electrical breakdown damage assessment model uses a fusion evaluation network structure, including:

[0178] Temperature change damage assessment sub-network:

[0179] Input layer: X T = [ΔT 1-0 , ΔT 0-1]

[0180] Residual module 1: R1 = ReLU(BN(Conv(X T ,W T1 ,b T1 ))+X T )

[0181] Residual module 2: R2 = ReLU(BN(Conv(R1,W T2 ,b T2 ))+R1)

[0182] Residual module 3: R3 = ReLU(BN(Conv(R2,W T3 ,b T3 ))+R2)

[0183] Fully connected layer: Y T = FC(R3)

[0184] Magnetic variable damage evaluation subnetwork:

[0185] Input layer: X H = [Delta H 1-0 , Delta H 0-1 ]

[0186] The network structure is similar to that of the temperature variable damage evaluation subnetwork, and contains three residual modules

[0187] Fully connected layer: Y H = FC(R3)

[0188] Fusion evaluation subnetwork:

[0189] Input layer: X F = [Y T , Y H ]

[0190] Fully connected layer 1: Z1 = ReLU(FC(X F ,W1,b1))

[0191] Fully connected layer 2: Y F = FC(Z1,W2,b2)

[0192] The final output Y F is the electric breakdown damage degree of the reactor winding.

[0193] Specifically, the principle of the present application is: based on the in-depth analysis of the temperature field and magnetic field variation law of the reactor winding under the condition of large current work.

[0194] In the initial stage of the inter-turn short circuit fault of the reactor, the local winding temperature rise and magnetic field distortion caused by the voltage difference between the windings will cause local electrical breakdown damage of the insulation material. Although this electrical breakdown damage does not cause the overall failure of the reactor, it will cause a certain degree of change in the reactance value. Specifically, the increase in the resistance of the winding material caused by the temperature rise will cause the reactance value to rise, and the change in the magnetic flux linkage will cause the reactance value to decrease. The combined effect of temperature change and magnetic field change determines the trend of the reactance value change of the reactor in the initial stage of the inter-turn short circuit fault.

[0195] Therefore, the present application adopts an infrared thermal imaging device and a magnetic probe to measure the temperature field and magnetic field distribution of the reactor winding under a large current state. By constructing temperature and magnetic field distribution models at T1, T0 and T2, the heat and magnetic field changes of the winding before and after the large current can be analyzed more finely.

[0196] Further, the present application adopts a pre-trained Tower CNN model to extract features from the temperature and magnetic field data at multiple times, which can quickly locate the region where the inter-turn short circuit fault occurs. Compared with the traditional empirical model, the deep learning model has stronger generalization ability and can be applied to different types of dry-type reactors.

[0197] Finally, the present application proposes a fusion evaluation network that considers both temperature change and magnetic field change information to output the electrical breakdown damage degree of the reactor winding. The network consists of a temperature change damage evaluation sub-network, a magnetic change damage evaluation sub-network and a fusion evaluation sub-network, which can fully utilize multi-source heterogeneous data to give more accurate fault diagnosis results.

[0198] Overall, the electrical breakdown damage evaluation method of the present application fully utilizes the potential of temperature and magnetic field data, and through fine thermal and magnetic field analysis and the application of deep learning technology, it realizes the accurate positioning of the inter-turn short circuit fault region and the quantitative evaluation of the damage degree.

[0199] The following provides an embodiment 2 of a specific application scenario of the present application: the embodiment 2 takes a 1000kvar dry-type air-core reactor as an example to describe the specific implementation process of the method of the present application in detail:

[0200] 1. Data acquisition

[0201] The power company installed an infrared thermal imaging device on the surface of the winding of the reactor, and arranged multiple magnetic probes around the winding. At a certain time T0 under the large current working state of the reactor, the infrared camera captured the temperature field distribution of the winding, and at the same time, the magnetic probes also monitored the instantaneous magnetic field data around the winding; in addition, the temperature field and magnetic field data at T1 = T0-Tz and T2 = T0+Tz were obtained respectively within a current period Tz before and after T0.

[0202] 2. Temperature and magnetic field distribution modeling

[0203] Based on the temperature and magnetic field data obtained above, the engineers of the power company used the finite element analysis method to construct temperature and magnetic field distribution models at T1, T0 and T2 respectively.

[0204] For T1, the temperature field distribution model can be expressed as:

[0205] T1(x, y) = f1(x, y)

[0206] where f1(x, y) is the temperature field interpolation function established based on the T1 infrared image data, and (x, y) is the coordinate of the winding plane. Similarly, the magnetic field distribution model is:

[0207] H1(x, y) = g1(x, y)

[0208] where g1(x, y) is the magnetic field distribution function constructed according to the T1 magnetic probe data.

[0209] For T0 and T2, temperature and magnetic field distribution models can also be similarly established:

[0210] T0(x, y) = f0(x, y)

[0211] H0(x, y) = g0(x, y)

[0212] T2(x, y) = f2(x, y)

[0213] H2(x, y) = g2(x, y)

[0214] With the temperature and magnetic field distribution models at these three times, the engineers can more finely analyze the heat and magnetic field changes of the reactor winding under large current conditions.

[0215] 3. Inter-turn short circuit fault area positioning

[0216] The power company had previously purchased a batch of pre-trained Tower CNN models that could be used for inter-turn short circuit fault positioning of dry-type reactors. The input of this model is 6 channels, which are T1(x, y), H1(x, y) at T1, T0(x, y), H0(x, y) at T0, and T2(x, y), H2(x, y) at T2.

[0217] The temperature and magnetic field data are input into the Tower CNN model. After multi-layer convolution and pooling in the encoder part, and deconvolution and up-sampling in the decoder part, a HxW segmentation map Y is finally output, where the area where the inter-turn short circuit fault occurs is identified by using the Sigmoid activation function.

[0218] 4. Temperature and magnetic field change calculation

[0219] According to the fault area located in step 3, the engineer next calculates the temperature change and magnetic field change of the area at T1 to T0 and T0 to T1.

[0220] Specifically, the temperature values of the fault area in the T1 temperature field model T1(x,y) and the T0 temperature field model T0(x,y) are recorded as T1(x,y) and T0(x,y) respectively, and the temperature change matrix at T1 to T0 is:

[0221] ΔT 1-0 (x,y) = T0(x,y) - T1(x,y)

[0222] Similarly, the magnetic field values of the fault area in the T1 magnetic field model H1(x,y) and the T0 magnetic field model H0(x,y) are recorded as H1(x,y) and H0(x,y) respectively, and the magnetic field change matrix at T1 to T0 is:

[0223] ΔH 1-0 (x,y) = H0(x,y) - H1(x,y)

[0224] And the temperature change matrix and the magnetic field change matrix at T0 to T1 are respectively:

[0225] ΔT 0-1 (x,y) = T1(x,y) - T0(x,y)

[0226] ΔH 0-1 (x,y) = H1(x,y) - H0(x,y)

[0227] The four change matrices reflect the heat and magnetic field change characteristics of the inter-turn short circuit fault area at different times, providing key inputs for subsequent electrical breakdown damage assessment.

[0228] 5. Electrical breakdown damage assessment

[0229] Finally, the power company inputs the temperature and magnetic field change matrices calculated in step 4, ΔT 1-0 , ΔH 1-0 , ΔT 0-1 and ΔH 0-1 , into the pre-trained fusion assessment network, and outputs the electrical breakdown damage degree of the reactor winding.

[0230] The fusion evaluation network includes:

[0231] The temperature change damage evaluation sub-network: the input layer receives ΔT 1-0 and ΔT 0-1 , passes through 3 residual modules and a fully connected layer, and outputs the temperature change damage evaluation result Y T .

[0232] The magnetic change damage evaluation sub-network: the input layer receives ΔH 1-0 and ΔH 0-1 , similar to the structure of the temperature change damage evaluation sub-network, outputs the magnetic change damage evaluation result Y H .

[0233] The fusion evaluation sub-network: the input layer receives Y T and Y H , passes through 2 fully connected layers, and outputs the final electric breakdown damage degree Y F .

[0234] After the processing of the fusion evaluation network, the output result Y F indicates that the electric breakdown damage degree of the reactor winding is 0.43, belonging to the moderate damage range.

[0235] According to the evaluation result, the power company decides to repair the reactor in time. First, the winding is carefully checked under the condition of power off, and it is confirmed that the inter-turn short circuit fault occurs in the marked area. Then, the insulation material in the area is replaced, and the overall winding is reinforced. After the repair, the reactance value of the reactor returns to the normal level, and the reactor can continue to operate safely in the future period of time.

[0236] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for evaluating initial breakdown damage of a dry-type air-core reactor due to a turn-to-turn short circuit fault, characterized by, The method comprises the following steps: S10, obtaining the winding infrared image collected by the infrared thermal imaging device and the instantaneous magnetic field data of the plurality of magnetic probes distributed around the winding at a large current time T0 of the reactor, and recording the T0 infrared image and the T0 instantaneous magnetic field group at the time T0 as T0 respectively; S20, obtaining the winding infrared image and the instantaneous voltage group at each time T1 and T2 of the reactor, wherein T1=T0-Tz and T2=T0+Tz, and recording the T1 infrared image and the T1 instantaneous magnetic field group and the T2 infrared image and the T2 instantaneous magnetic field group as T2 respectively, and the Tz is the period of the input current; S30, modeling the static thermal distribution and the magnetic field distribution of the reactor winding by using the infrared image and the instantaneous magnetic field group at each time T1, T0 and T2 respectively, and obtaining the T1 temperature field, the T1 magnetic field, the T0 temperature field, the T0 magnetic field, the T2 temperature field and the T2 magnetic field at the corresponding time respectively; S40, determining the occurrence area of the inter-turn short circuit fault by using the pre-trained inter-turn short circuit fault positioning model according to the T1 temperature field, the T1 magnetic field, the T0 temperature field, the T0 magnetic field, the T2 temperature field and the T2 magnetic field, and recording the fault area as fault area; S50, obtaining the temperature change and the magnetic field change of the fault area at the T1 time, the T0 time and the T2 time respectively, and forming the temperature change matrix and the magnetic change matrix between the T1 time and the T0 time and the temperature change matrix and the magnetic change matrix between the T0 time and the T1 time; S60, inputting the temperature change matrix and the magnetic change matrix between the T1 time and the T0 time and the temperature change matrix and the magnetic change matrix between the T0 time and the T1 time into the pre-trained electric breakdown damage evaluation model to obtain the electric breakdown damage degree and output.

2. The method according to claim 1, characterized in that, The inter-turn short circuit fault positioning model adopts a multi-layer tower type convolutional neural network.

3. The method according to claim 1, characterized in that, The electric breakdown damage evaluation model adopts a fusion evaluation network structure including a temperature change damage evaluation sub-network, a magnetic change damage evaluation sub-network and a fusion evaluation sub-network.

4. The method according to claim 3, characterized in that, The temperature change damage evaluation sub-network is used to receive the temperature change matrix between the T1 time and the T0 time and between the T0 time and the T1 time, and output the temperature change damage evaluation result.

5. The method according to claim 3, wherein the method further comprises: determining the initial breakdown damage of the dry-type air-core reactor based on the first voltage and the second voltage. The magnetic change damage evaluation sub-network is used to receive the magnetic change matrix between the T1 time and the T0 time and between the T0 time and the T1 time, and output the magnetic change damage evaluation result.

6. The method according to claim 3, wherein the method is characterized by: The fusion evaluation sub-network is used to receive the temperature change damage evaluation result and the magnetic change damage evaluation result, and output the final electric breakdown damage degree.

7. The method of claim 1, wherein the method further comprises: determining the initial breakdown damage of the dry-type air-core reactor based on the comparison result. The infrared thermal imaging device is used to non-contact measure the temperature distribution of the winding surface, and the magnetic probe is used to collect the magnetic field change around the winding in real time.

8. The method of claim 1, wherein the method further comprises: determining the initial breakdown damage of the dry-type air-core reactor based on the comparison result. The large current refers to the time when the current flowing through the reactor reaches more than 16 times of the average current of the reactor.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores program instructions, and the program instructions are used to execute the method for evaluating the initial electric breakdown damage of the inter-turn short circuit fault of the dry-type air-core reactor according to any one of claims 1-8 when running.

10. A system for assessing the initial electrical breakdown damage of inter-turn short-circuit faults in dry-type air-core reactors, characterized in that, The computer readable storage medium comprises the computer readable storage medium of claim 9.

Images