A method and device for dynamic compensation and correction of mutual inductor loop resistance

Abstract

The application provides a mutual inductor loop resistance dynamic compensation correction method and device, relates to the technical field of circuit test, and the method comprises the following steps: according to the structure parameter information and environmental parameters of primary winding and secondary winding, a structure characteristic tensor and an environmental interference tensor are constructed; based on the structure characteristic tensor, the environmental interference tensor and historical loop resistance test data, a dynamic compensation model is trained; a set direct current is injected into the primary winding, an initial voltage signal sequence of the two ends of the primary winding is obtained, a test state tensor is constructed in combination with the current loop current and real-time environmental parameters; the test state tensor is input into the dynamic compensation model, and a non-target resistance compensation value at the current time is output; based on the non-target resistance compensation value, the initial loop resistance value calculated from the initial voltage signal sequence is corrected, and a compensation resistance value of the target mutual inductor loop is obtained. The application can improve the accuracy of mutual inductor loop resistance test.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of circuit testing, in particular to a mutual inductor loop resistance dynamic compensation correction method and device. BACKGROUND

[0002] Mutual inductor loop resistance testing refers to injecting a stable direct current into a primary winding of a target mutual inductor, and synchronously measuring the voltage drop across the primary winding, combining Ohm's law to calculate the resistance value of the primary loop of the mutual inductor, so as to evaluate the contact integrity and current-carrying capacity of the conductive loop. In order to improve the measurement accuracy, a four-terminal measurement method is usually used, that is, the current injection and voltage sampling channels are independently arranged, so as to avoid introducing additional errors by lead and contact resistance. This test is a key means to determine whether there are defects such as poor contact, oxidation of welding points, loose bolts or other factors causing the degradation of the conductive performance of the primary winding and the external connection loop of the mutual inductor, and is widely used in mutual inductor factory acceptance, on-site handover test and operation and maintenance state evaluation. However, since the measured resistance value is generally low, it is easily affected by non-target interference factors such as test environment temperature and humidity fluctuations, electromagnetic interference, terminal contamination and human wiring errors, causing data deviation, and thus weakening the effective characterization ability of the mutual inductor loop to the real state.

[0003] In the existing mutual inductor loop resistance testing technology, a direct current injection and four-terminal measurement method are generally used to calculate the resistance of the target mutual inductor loop. Although relatively accurate measurement results can be obtained under ideal conditions, the actual operation site is significantly affected by multiple factors, including environmental temperature and humidity changes, which often lead to problems such as instability, large deviation and poor repeatability of the test data. Especially in complex site environments such as substations, the traditional static measurement method cannot effectively identify and eliminate non-target resistance components, and the measurement results cannot truly reflect the conductive performance of the mutual inductor loop itself, affecting the accuracy of operation and maintenance judgment. SUMMARY

[0004] The present application provides a mutual inductor loop resistance dynamic compensation correction method and device, which can improve the accuracy of mutual inductor loop resistance testing.

[0005] In a first aspect of the present application, a mutual inductor loop resistance dynamic compensation correction method is provided, the method comprising:

[0006] Obtaining structure parameter information of a primary winding and a secondary winding of a target mutual inductor, and collecting environmental parameters of a target mutual inductor loop, to construct a structure feature tensor and an environmental interference tensor;

[0007] Training a dynamic compensation model based on the structure feature tensor and the environmental interference tensor, and combining historical loop resistance test data, the dynamic compensation model being constructed based on a recurrent neural network;

[0008] Injecting a set direct current into the primary winding, obtaining an initial voltage signal sequence at both ends of the primary winding, combining the current loop and real-time environmental parameters to construct a test state tensor;

[0009] Inputting the test state tensor into the dynamic compensation model to output a non-target resistance compensation value at the current time;

[0010] Based on the non-target resistance compensation value, the initial loop resistance value calculated from the initial voltage signal sequence is corrected to obtain the compensation resistance value of the target transformer loop.

[0011] Based on the above technical solutions, preferably, after the non-target resistance compensation value is obtained, the initial loop resistance value calculated from the initial voltage signal sequence is corrected to obtain the compensation resistance value of the target transformer loop, the method further comprises:

[0012] Extracting a plurality of historical compensation resistance values within the current test environmental parameter range to construct a time sequence sample containing time stamp, loop current amplitude, environmental interference parameter and corresponding compensation resistance value;

[0013] According to the time sequence sample, an exponential weighted moving average strategy is used to fit the deviation of the compensation resistance value, and an error function is constructed based on the least mean square criterion;

[0014] Combining the error function and the weight decay algorithm to correct the recurrent neural network weight matrix in the dynamic compensation model by back propagation, realizing the adaptive parameter update of the dynamic compensation model under the current test environment.

[0015] Based on the above technical solutions, preferably, the structure parameter information of the primary winding and the secondary winding of the target transformer is obtained, and the environmental parameters of the target transformer loop are collected to construct a structure feature tensor and an environmental interference tensor, which specifically includes:

[0016] Collecting the structure parameter information of the primary winding and the secondary winding of the target transformer, the structure parameter information includes the number of turns, wire diameter, winding arrangement, material resistivity, core shape, core magnetic permeability, winding cross-sectional area of the primary winding, and the ratio of the secondary winding, tap position and internal insulation structure;

[0017] By discretizing and standardizing the structure parameter information, it is mapped to a multi-dimensional structure feature space to construct a structure feature tensor;

[0018] Collecting the environmental parameters of the target transformer loop, the environmental parameters include winding surface temperature, environmental humidity, power frequency, electromagnetic interference amplitude, voltage transient rate and noise spectrum energy density distribution;

[0019] extracting statistical feature quantities of the environment parameters based on the time synchronization window, and constructing an environment interference tensor;

[0020] concatenating the structure feature tensor and the environment interference tensor to form a composite input feature tensor as an input variable of the dynamic compensation model.

[0021] On the basis of the above technical solutions, preferably, the structure feature tensor and the environment interference tensor are combined with historical loop resistance test data to train a dynamic compensation model, specifically including:

[0022] constructing a historical training sample set containing the structure feature tensor and the environment interference tensor, each data in the historical training sample set including a structure feature tensor, an environment interference tensor and a corresponding historical loop resistance test value at the same time step;

[0023] concatenating the structure feature tensor and the environment interference tensor to form a composite input feature tensor as a model input, and constructing the dynamic compensation model, the dynamic compensation model being composed of a plurality of stacked long short-term memory units and introducing a residual connection and a gating mechanism to enhance the historical disturbance dependent modeling capability;

[0024] using mean square error as a loss function, using an adaptive learning rate optimization algorithm to iteratively update the network weight parameters of the dynamic compensation model, and simultaneously introducing a dropout strategy and an L2 regularization term to prevent overfitting;

[0025] in the training process, the historical training sample set is divided using a cross-validation method to verify the model convergence performance, and the parameter optimization of the dynamic compensation model and the generalization training of the non-target resistance compensation prediction capability are completed.

[0026] On the basis of the above technical solutions, preferably, the set direct current is injected into the primary winding to obtain an initial voltage signal sequence across the primary winding, and a test state tensor is constructed in combination with a current loop and real-time environment parameters, specifically including:

[0027] controlling a constant current source control module to inject a preset direct current into the primary winding, the preset direct current being determined according to a primary rated current range of the target transformer and an estimated loop resistance;

[0028] continuously sampling the voltage drop across the primary winding to form an initial voltage signal sequence;

[0029] synchronously collecting a real-time loop current value of the current loop, the loop current value being obtained by a current sampling unit based on a low-impedance shunt detector;

[0030] acquire the environmental parameters collected by the environment perception module;

[0031] pair the initial voltage signal sequence with the loop current value to generate an instantaneous initial loop resistance sequence, and jointly construct the test state tensor with the environmental parameters.

[0032] On the basis of the above technical solutions, preferably, the test state tensor is input into the dynamic compensation model, and a non-target resistance compensation value at a current time is output, specifically including:

[0033] The test state tensor is input into the dynamic compensation model, the dynamic compensation model is composed of multiple long short-term memory units, a model input layer receives the test state tensor and is transmitted to the long short-term memory unit in a time sequence manner, a time sequence feature and an amplitude response of a test disturbance are extracted by using a gating mechanism, and a hidden state vector reflecting dynamic changes of the non-target resistance is generated.

[0034] An output resistance compensation value at a current time is output based on the hidden state vector after being processed by a fully connected mapping layer and a nonlinear activation function.

[0035] The output resistance compensation values in the sliding prediction window are fused by using a weighted median filtering method, and the non-target resistance compensation value is obtained.

[0036] On the basis of the above technical solutions, preferably, the initial loop resistance value calculated based on the initial voltage signal sequence is corrected based on the non-target resistance compensation value, and a compensation resistance value of the target transformer loop is obtained, specifically including:

[0037] Based on the initial voltage signal sequence and the loop current value, an initial loop resistance instantaneous value sequence in an uncorrected state is obtained.

[0038] The non-target resistance compensation value corresponding to the initial loop resistance instantaneous value sequence in the time stamp is matched point by point, and the initial loop resistance instantaneous value sequence is deducted by the non-target resistance compensation value item by item, to generate a compensation loop resistance instantaneous value sequence.

[0039] The compensation loop resistance instantaneous value sequence is filtered to obtain the compensation resistance value of the target transformer loop.

[0040] In a second aspect of the present application, a transformer loop resistance dynamic compensation correction device is provided, which is used to execute any one of the transformer loop resistance dynamic compensation correction methods described above. The device includes an acquisition module, a processing module and an output module, wherein:

[0041] The acquisition module is configured to acquire structural parameter information of a primary winding and a secondary winding of a target transformer, and collect environmental parameters of a loop of the target transformer, and construct a structural feature tensor and an environmental interference tensor;

[0042] The processing module is configured to train a dynamic compensation model based on the structural feature tensor and the environmental interference tensor, and in combination with historical loop resistance test data, and the dynamic compensation model is constructed based on a recurrent neural network;

[0043] The processing module is configured to inject a set direct current into the primary winding, acquire an initial voltage signal sequence at both ends of the primary winding, and construct a test state tensor in combination with a current loop and real-time environmental parameters;

[0044] The processing module is configured to input the test state tensor into the dynamic compensation model, and output a non-target resistance compensation value at a current time point;

[0045] The output module is configured to correct an initial loop resistance value calculated based on the initial voltage signal sequence based on the non-target resistance compensation value, and obtain a compensation resistance value of the loop of the target transformer.

[0046] On the basis of the above technical solutions, preferably, the acquisition module is configured to extract a plurality of historical compensation resistance values within a current test environmental parameter range, and construct a time sequence sample containing a time stamp, a loop current amplitude, an environmental interference parameter, and a corresponding compensation resistance value;

[0047] The processing module is configured to perform bias fitting on the compensation resistance value using an exponential weighted moving average strategy according to the time sequence sample, and construct an error function based on a minimum mean square error criterion;

[0048] The processing module is configured to perform back propagation correction on a recurrent neural network weight matrix in the dynamic compensation model in combination with the error function and a weight decay algorithm, and realize adaptive parameter updating of the dynamic compensation model under a current test environment.

[0049] On the basis of the above technical solutions, preferably, the acquisition module is configured to collect structural parameter information of a primary winding and a secondary winding of a target transformer, and the structural parameter information includes the number of turns, the wire diameter, the winding arrangement, the material resistivity, the core shape, the core magnetic permeability, the winding cross-sectional area of the primary winding, and the transformation ratio coefficient, the tap position, and the internal insulation structure of the secondary winding;

[0050] The processing module is configured to map to a multi-dimensional structural feature space to construct a structural feature tensor by discretizing and encoding and standardizing the structural parameter information;

[0051] The acquisition module is configured to collect environmental parameters of the target mutual inductor circuit, and the environmental parameters include winding surface temperature, environmental humidity, power frequency, electromagnetic interference amplitude, voltage transient rate, and noise frequency spectrum energy density distribution.

[0052] The processing module is configured to extract statistical characteristic quantities of the environmental parameters based on a time synchronization window, and construct an environmental interference tensor.

[0053] The processing module is configured to splice the structural feature tensor and the environmental interference tensor to form a composite input feature tensor as an input variable of the dynamic compensation model.

[0054] On the basis of the above technical solutions, preferably, the processing module is configured to construct a historical training sample set containing the structural feature tensor and the environmental interference tensor, and each item of data in the historical training sample set includes a structural feature tensor, an environmental interference tensor, and a corresponding historical loop resistance test value at the same time step.

[0055] The processing module is configured to splice the structural feature tensor and the environmental interference tensor to form a composite input feature tensor as a model input, and construct the dynamic compensation model, wherein the dynamic compensation model is composed of a plurality of stacked long short-term memory units, and residual connection and gating mechanisms are introduced to enhance the historical disturbance dependent modeling capability.

[0056] The processing module is configured to use mean square error as a loss function, use an adaptive learning rate optimization algorithm to iteratively update network weight parameters of the dynamic compensation model, and introduce a dropout strategy and an L2 regularization term to prevent overfitting.

[0057] The processing module is configured to divide the historical training sample set using a cross-validation method to verify the model convergence performance during the training process, and complete parameter optimization of the dynamic compensation model and generalization training of the non-target resistance compensation prediction capability.

[0058] On the basis of the above technical solutions, preferably, the processing module is configured to control a constant current source control module to inject a preset direct current into the primary winding, and the preset direct current is determined according to a primary rated current range and an estimated loop resistance of the target mutual inductor.

[0059] The processing module is configured to continuously sample a voltage drop across the primary winding to form an initial voltage signal sequence.

[0060] The acquisition module is configured to synchronously collect a real-time loop current value of a current loop, and the loop current value is obtained by a current sampling unit based on a low-impedance shunt.

[0061] The acquisition module is configured to acquire the environmental parameters collected by the environmental perception module.

[0062] The output module is configured to pair the initial voltage signal sequence with the loop current value to generate a transient initial loop resistance sequence, and to jointly construct the test state tensor with the environmental parameters.

[0063] On the basis of the above technical solutions, preferably, the output module is configured to input the test state tensor into the dynamic compensation model, the dynamic compensation model is composed of multiple long short-term memory units, the model input layer receives the test state tensor and transmits the test state tensor to the long short-term memory unit in a time sequence manner, the time sequence characteristics and amplitude response of the test disturbance are extracted by using a gating mechanism, and an implicit state vector reflecting the dynamic change of the non-target resistance is generated.

[0064] The acquisition module is configured to acquire the output resistance compensation value at the current time based on the implicit state vector, a fully connected mapping layer, and a non-linear activation function.

[0065] The processing module is configured to fuse the output resistance compensation values in the sliding prediction window by using a weighted median filtering method to obtain the non-target resistance compensation value.

[0066] On the basis of the above technical solutions, preferably, the processing module is configured to calculate an initial loop resistance transient value sequence in an uncorrected state based on the initial voltage signal sequence and the loop current value.

[0067] The processing module is configured to point-by-point match the non-target resistance compensation value corresponding to the initial loop resistance transient value sequence in the time stamp, and to subtract the non-target resistance compensation value from the initial loop resistance transient value sequence item by item to generate a compensated loop resistance transient value sequence.

[0068] The output module is configured to perform filtering processing on the compensated loop resistance transient value sequence to obtain the compensated resistance value of the target transformer loop.

[0069] In a third aspect of the present application, an electronic device is provided, which includes a processor, a memory, a user interface, and a network interface, the memory is configured to store instructions, the user interface and the network interface are configured to communicate with other devices, and the processor is configured to execute the instructions stored in the memory to enable the electronic device to perform the method according to any one of the above aspects.

[0070] In a fourth aspect of the present application, a computer readable storage medium is provided, which stores instructions, when the instructions are executed, the method according to any one of the above aspects is performed.

[0071] To sum up, the one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0072] 1. The present application quantifies the physical structure difference of the target transformer and the environmental disturbance factors in the test site by constructing the structural feature tensor and the environmental disturbance tensor, and trains a dynamic compensation model combined with historical loop resistance test data to realize real-time prediction and dynamic elimination of non-target resistance components; in the actual test process, the test state tensor is constructed by flow sampling and input into the compensation model, and the non-target resistance compensation value is output to finely correct the original measurement results, significantly improving the identification and suppression ability of disturbance factors such as lead resistance, poor contact, electromagnetic interference, etc. in complex site environment, so as to realize accurate extraction of the real conductive performance of the transformer loop, and effectively improve the measurement accuracy and engineering applicability of the loop resistance test.

[0073] 2. The historical compensation resistance value is introduced to construct a time series sample, and an error function is established combined with the exponential weighted moving average and the least mean square criterion, and the dynamic compensation model is further corrected by using the weight decay algorithm for back propagation, so that the model can continuously optimize its non-target resistance prediction performance in a new environment, thereby enhancing the adaptive ability and site adaptability of the model, and improving the overall prediction accuracy and robustness.

[0074] 3. The structural parameters and environmental parameters of the primary winding and secondary winding of the transformer are comprehensively collected and converted into structural feature tensors and environmental disturbance tensors, so that the subsequent compensation model has the ability to quantitatively identify and express the differences in physical configuration and environmental disturbance changes, providing a complete multi-dimensional input data basis for dynamic compensation modeling, thereby improving the perception dimension and compensation expression ability of the model to interference characteristics.

[0075] 4. The historical training sample set containing the structural feature tensor and the environmental disturbance tensor is constructed, and the multi-layer long short-term memory structure combined with the residual connection and the gating mechanism is used to model the nonlinear disturbance behavior, while the adaptive learning rate optimization algorithm and the regularization strategy are used for parameter update, and the model generalization training is realized through cross-validation, which significantly improves the learning ability and generalization performance of the dynamic compensation model to non-target resistance disturbance, and ensures its stable output effect in multiple scenarios.

[0076] 5. Constant current source flow control, four-terminal method voltage sampling, accurate current detection and synchronous perception of environmental parameters, construction of test state tensor, so that the electrical measurement data and disturbance state have high time sequence synchronization and multi-modal fusion characteristics, providing input samples with strong timeliness, disturbance sensitivity and stable structure for the dynamic compensation model, effectively improving the input integrity and measurement scene reconstruction ability in the non-target resistance prediction process.

[0077] 6. The test state tensor is input into a dynamic compensation model composed of multiple layers of long short-term memory units, the disturbance evolution features are extracted by combining the gating mechanism, and the non-target resistance compensation value is output through the mapping layer. Further introduce the weighted median filter to stabilize the prediction results in the sliding window, enhance the stability and anti-interference ability of the compensation value, and ensure the accuracy and volatility control of the final correction amount.

[0078] 7. The initial voltage signal sequence and loop current value are calculated in real time, and combined with the non-target resistance compensation value for point-by-point correction to generate a compensated loop resistance instantaneous value sequence, and the stable compensation resistance value is output through the filtering algorithm, realizing a complete loop resistance recovery path from interference identification, error elimination to stable correction, significantly improving the usability and reliability of the target transformer loop resistance measurement in complex environments. BRIEF DESCRIPTION OF DRAWINGS

[0079] Figure 1 is a flowchart of a transformer loop resistance dynamic compensation correction method disclosed in an embodiment of the present application;

[0080] Figure 2 is a module schematic diagram of a transformer loop resistance dynamic compensation correction device disclosed in an embodiment of the present application;

[0081] Figure 3 is a structural schematic diagram of an electronic device disclosed in an embodiment of the present application.

[0082] BRIEF DESCRIPTION OF DRAWINGS DETAILED DESCRIPTION

[0083] In order for those skilled in the art to better understand the technical solutions in the specification, the technical solutions in the specification will be described clearly and completely in the following with reference to the drawings in the embodiments of the specification. Obviously, the described embodiments are only some of the embodiments of the present application, not all.

[0084] In the description of the embodiments of the present application, the words such as "for example" or "for instance" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "for example" or "for instance" in the embodiments of the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the words such as "for example" or "for instance" are intended to present the relevant concept in a specific way.

[0085] In the description of the embodiments of the present application, the term "a plurality of" means two or more. For example, a plurality of systems means two or more systems, and a plurality of screen terminals means two or more screen terminals. In addition, the terms "first" and "second" are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more features. The terms "include", "contain", "have" and their variants mean "include but are not limited to", unless otherwise specifically emphasized.

[0086] The transformer loop resistance test is a key means to identify the defects of poor contact, solder oxidation and bolt loosening and other conductive performance degradation by injecting a stable DC constant current into the primary winding of the target transformer and measuring the voltage drop across it using the four-terminal measurement method, combining Ohm's law to calculate the loop resistance value, to evaluate the loop contact integrity and current-carrying capacity. It is widely used in factory acceptance, handover test and operation and maintenance evaluation. However, due to the low test resistance value, actual field testing is easily affected by non-target interference such as changes in environmental temperature and humidity, electromagnetic interference and human wiring errors, resulting in unstable data, large deviation and poor repeatability. The traditional static measurement method cannot effectively eliminate the non-target resistance component in the complex field environment, limiting the accurate representation of the measurement results on the real conductive performance of the transformer loop.

[0087] The embodiment discloses a transformer loop resistance dynamic compensation correction method, referring to Figure 1 , comprising the following steps S110-S150:

[0088] S110, obtaining the structure parameter information of the primary winding and the secondary winding of the target transformer, and collecting the environmental parameters of the target transformer loop, constructing the structure feature tensor and the environmental interference tensor.

[0089] The transformer loop resistance dynamic compensation correction method disclosed in the embodiments of the present application is applied to a server. The server includes but is not limited to electronic devices such as mobile phones, tablet computers, wearable devices, PC (Personal Computer), and the like. It can also be a background server running a transformer loop resistance dynamic compensation correction method. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0090] In a possible implementation, structural parameter information of a primary winding and a secondary winding of a target transformer is acquired, and environmental parameters of a loop of the target transformer are collected, a structural feature tensor and an environmental interference tensor are constructed, and specifically, the structural parameter information of the primary winding and the secondary winding of the target transformer is collected, the structural parameter information includes the number of turns of the primary winding, the wire diameter, the winding arrangement mode, the material resistivity, the core shape, the core magnetic permeability, the winding cross-sectional area, and the transformation ratio coefficient, the tap position, and the internal insulation structure of the secondary winding; the structural parameter information is discretely coded and standardized, and is mapped to a multi-dimensional structural feature space to construct the structural feature tensor; the environmental parameters of the loop of the target transformer are collected, the environmental parameters include the winding surface temperature, the environmental humidity, the power frequency, the electromagnetic interference amplitude, the voltage transient rate, and the noise spectrum energy density distribution; statistical characteristic quantities of the environmental parameters are extracted based on a time synchronization window, and the environmental interference tensor is constructed; the structural feature tensor and the environmental interference tensor are spliced to form a composite input feature tensor, which is used as an input variable of a dynamic compensation model.

[0091] Specifically, in the implementation process, first, the structural parameter information of the primary winding and the secondary winding of the target transformer is collected, and specifically includes the number of turns of the primary winding, the wire diameter, the winding arrangement mode, the material resistivity, the core shape, the core magnetic permeability, the winding cross-sectional area, and the transformation ratio coefficient, the tap position, and the internal insulation structure of the secondary winding. The collection of the above parameters can be automatically extracted through the transformer manufacturing account, the structural drawing, or the structure parameter recognition unit integrated in the device.

[0092] The structural parameter information of the primary winding and the secondary winding is basic data representing the electric conduction performance and the electromagnetic coupling characteristics of the transformer, wherein the number of turns of the primary winding determines the loop length and the induction intensity, the wire diameter affects the current-carrying capacity and the resistance level, the winding arrangement mode reflects the compactness of the winding structure and the magnetic flux distribution, the material resistivity determines the resistance value of the winding conductor per unit length, the core shape and the core magnetic permeability jointly determine the magnetic circuit characteristics and the magnetic saturation behavior of the transformer, and the winding cross-sectional area directly affects the resistance loss and the thermal stability per unit current; the transformation ratio coefficient of the secondary winding represents the proportional relationship between the primary and secondary sides, is a key parameter of the transmission accuracy of the transformer, the tap position reflects the electrical configuration flexibility of the secondary side structure, and the internal insulation structure constitutes the insulation strength and the voltage withstand capability between the windings and the core. The above parameters jointly constitute a complete structural expression of the internal electromagnetic characteristics, the conduction capacity, and the insulation performance of the transformer.

[0093] After the collection is completed, in order to realize the modeling of the structure feature tensor, the various structure parameters are discretized and coded and standardized. The discretized coding assigns a numerical label or one-hot coding to each category variable, such as mapping "layer winding", "pie winding" and "concentric winding" in "winding arrangement" to a multi-dimensional vector; for continuous variables such as number of turns, resistivity, cross-sectional area, etc., the Z-score standardization method is used to remove the dimension effect and maintain the relative distribution consistency. Finally, all the structure parameters are mapped to a unified multi-dimensional structure feature space to form a structure feature tensor, and each dimension of the tensor corresponds to a certain structure feature variable, which retains the information expression ability of the physical structure difference of the target transformer.

[0094] Subsequently, the environmental parameters of the target transformer loop are collected, including: winding surface temperature, which is collected in real time using a surface thermocouple sensor or an infrared temperature measurement module; environmental humidity is obtained by a relative humidity sensor; power frequency is measured by analyzing the zero-crossing period through a voltage sampling circuit; electromagnetic interference amplitude is obtained by a near-field interference probe combined with a spectrum analyzer to obtain electric field or magnetic field strength; voltage transient rate is monitored by a high-speed sampling module to monitor the voltage slope change in the power supply waveform; and noise spectrum energy density distribution is obtained by frequency domain analysis of the loop noise signal using the fast Fourier transform method to obtain the frequency component energy. All collected data are attached with high-precision time stamps.

[0095] After obtaining the above environmental parameters, the statistical characteristic quantities thereof are extracted based on a time synchronization window. The specific operation is to calculate the mean, variance, maximum value, change rate and power spectral density within a set sliding time window (such as 2 seconds or 1000 sampling points) to convert the instantaneous environmental fluctuations into quantitative data with representation ability, thereby forming an environmental disturbance tensor. Each dimension in the tensor corresponds to a statistical characteristic of an environmental disturbance factor, and the structure of each dimension is aligned with that of the structure feature tensor.

[0096] Finally, the structure feature tensor and the environmental disturbance tensor are spliced, and a vector-level splicing method is used to combine the two tensors along the feature dimension to form a composite input feature tensor, which is used as the input variable of the dynamic compensation model. The composite input feature tensor is used as the input of the subsequent recurrent neural network model to learn the non-target resistance change rule of the target transformer under different structure compositions and disturbance environments, thereby providing a unified modeling input basis for realizing high-precision non-target resistance compensation. The entire process realizes comprehensive quantitative representation of the state of the target transformer in both time axis and feature space.

[0097] S120, based on the structure feature tensor and the environmental disturbance tensor, and combined with historical loop resistance test data, a dynamic compensation model is trained.

[0098] In a possible implementation, the dynamic compensation model is trained based on the structure feature tensor and the environmental disturbance tensor, and in combination with historical loop resistance test data, specifically including: constructing a historical training sample set containing the structure feature tensor and the environmental disturbance tensor, each data in the historical training sample set including the structure feature tensor, the environmental disturbance tensor at the same time step, and the corresponding historical loop resistance test value; concatenating the structure feature tensor and the environmental disturbance tensor to form a composite input feature tensor as the model input, constructing the dynamic compensation model, the dynamic compensation model being composed of multiple long short-term memory units stacked in depth, and introducing a residual connection and a gating mechanism to enhance the historical disturbance dependent modeling capability; using mean square error as a loss function, using an adaptive learning rate optimization algorithm to iteratively update the network weight parameters of the dynamic compensation model, while introducing a dropout strategy and an L2 regularization term to prevent overfitting; in the training process, the historical training sample set is divided using a cross-validation method to verify the model convergence performance, and the parameter optimization of the dynamic compensation model and the generalization training of the non-target resistance compensation prediction capability are completed.

[0099] Specifically, in the implementation process, first, a historical training sample set containing the structure feature tensor and the environmental disturbance tensor is constructed, the historical training sample set being indexed by a uniform time step, wherein each data is composed of three parts: one is the structure feature tensor formed by the discretization and standardization of the structure parameters of the primary winding and the secondary winding of the target transformer at the time step; the second is the environmental disturbance tensor composed of the environmental parameters collected and extracted in the same time window; and the third is the historical loop resistance test value obtained by four-terminal measurement of the loop resistance of the target transformer at the same time, which constitutes a supervision label. All samples are timestamped to ensure the consistency of the structure feature tensor, the environmental disturbance tensor, and the loop resistance test value in the time dimension, forming a standard supervised learning sample set.

[0100] Subsequently, each group of structure feature tensor and environmental disturbance tensor is concatenated in the feature dimension to form a composite input feature tensor, which encodes the structure information and external disturbance state of the target transformer in a unified vector space to capture their joint action on the non-target resistance interference. On this basis, the dynamic compensation model is constructed, which adopts a long short-term memory neural network structure. The structure effectively handles the long-term dependence problem existing in the input data by introducing a gating mechanism composed of an input gate, a forgetting gate, and an output gate; multiple long short-term memory units are stacked in the depth direction to build the network backbone, and a residual connection mechanism is introduced to relieve the gradient vanishing problem and enhance the feature transmission path, thereby improving the fitting capability of the multi-time scale disturbance signal.

[0101] In the model training process, the mean square error is used as the loss function to calculate the average square deviation between the non-target resistance compensation value output by the model and the true historical loop resistance test value. The loss function has the properties of continuous derivability and sensitivity to abnormal values, which helps to guide the model to converge to the direction of the minimum global error. The adaptive learning rate optimization algorithm based on gradient descent optimization strategy (such as Adam) is used for network parameter iterative update. During the optimization process, the learning rate corresponding to each parameter is automatically adjusted to adapt to the gradient change amplitude, improving the convergence speed and stability. At the same time, the dropout strategy is introduced to randomly shield part of the neuron activation response in each iteration to prevent the model from over-memorizing the training samples and enhance its generalization ability. The L2 regularization term is introduced to constrain the network weight size, inhibit the formation of high complexity model, and reduce the risk of overfitting.

[0102] Finally, in the entire training phase, the historical training sample set is divided by cross-validation, for example, the sample data set is divided into a training set and a validation set, and the model performance is trained and validated on multiple divided subsets. The cross-validation technique can be used to evaluate the stability and convergence performance of the model under different sample distributions. By analyzing the validation error trend, the number of training rounds and the model depth are dynamically adjusted to achieve dynamic compensation of the model in the full sample space. The parameter optimization and generalization training output of the non-target resistance compensation prediction ability, so as to ensure its robustness and adaptability in subsequent field tests.

[0103] S130, injecting a set direct current into the primary winding, obtaining an initial voltage signal sequence at both ends of the primary winding, and constructing a test state tensor in combination with the current loop current and real-time environmental parameters.

[0104] In one possible implementation, a set direct current is injected into the primary winding, an initial voltage signal sequence at both ends of the primary winding is obtained, and a test state tensor is constructed in combination with the current loop current and real-time environmental parameters, specifically including: controlling the constant current source control module to inject a preset direct current into the primary winding, the preset direct current being determined according to the primary rated current range of the target transformer and the estimated loop resistance; continuously sampling the voltage drop at both ends of the primary winding to form an initial voltage signal sequence; synchronously collecting the real-time loop current value of the current loop current, the loop current value being obtained by the current sampling unit based on a low-impedance shunt; obtaining the environmental parameters collected by the environmental perception module; pairing the initial voltage signal sequence with the loop current value to generate an instantaneous initial loop resistance sequence, and constructing a test state tensor together with the environmental parameters.

[0105] Specifically, in the implementation process, first, the constant current source control module injects a preset direct current into the primary winding of the target transformer. The constant current source control module is composed of a high-stability programmable direct current power supply and has a closed-loop current control capability, which can adjust the output current in real time to match the set value. The size of the preset direct current is determined according to the rated current range of the target transformer and the estimated loop resistance, and is usually 5% to 10% of the rated value, so as to avoid heating or magnetic saturation of the transformer while ensuring the measurement sensitivity. For example, for a transformer with a rated primary current of 1000A and an estimated loop resistance of 200μΩ, the injected current is set to 50A, and the corresponding voltage drop is 10mV, which is within the sensitivity range of conventional measuring instruments. The current is set by a digital controller and output to the primary winding through an optical isolation channel to ensure signal stability and system electrical safety.

[0106] Subsequently, the voltage drop across the primary winding is continuously sampled through the voltage acquisition channel. The sampling adopts a four-terminal measurement method, that is, independent voltage sampling leads are connected to the conductors at both ends of the primary winding of the transformer, thereby avoiding additional errors introduced by the contact resistance and wire resistance of the injection lead. The voltage signal is sampled by a high-precision analog-to-digital converter, and the sampling rate is usually set to more than 10kHz to capture transient disturbances. The collected voltage data forms a time series, constituting an initial voltage signal sequence, and through a band-pass filter and noise suppression algorithm, power frequency interference and spike noise are eliminated to ensure the stability of the subsequent loop resistance calculation.

[0107] At the same time of voltage sampling, the loop current value is synchronously collected through the current sampling unit. The current sampling unit is composed of a low-impedance shunt and a differential amplifier. The low-impedance shunt is a standard metal foil shunt with a resistance of several dozen to several hundred micro-ohms, which has good thermal stability and linearity. When the current passes through the shunt, a small voltage drop is generated, which is sent to the analog-to-digital conversion module after differential amplification to obtain the real-time value of the injected current. The sampling clock is synchronized with the voltage sampling to ensure the one-to-one correspondence of voltage and current data on the time axis.

[0108] At the same time, the environmental parameters at the current test time are obtained from the environmental perception module. The environmental perception module is composed of multiple sensors, including a temperature sensor (such as a PT100 thermal resistor or a thermistor) for collecting the winding surface temperature, a humidity sensor (such as a capacitive humidity sensor) for collecting the relative humidity of the environment, a power frequency monitoring module based on voltage zero-crossing detection to measure frequency drift, electromagnetic interference amplitude coupled to the frequency spectrum acquisition circuit through a near-field probe and processed by short-time Fourier transform to obtain the disturbance energy distribution; the voltage transient rate is detected by a high-speed comparator in real time to detect the change of the waveform steepness and compared with a preset slope threshold; the noise spectrum energy density is obtained by frequency domain analysis of the collected loop voltage high-frequency signal. All environmental parameters are timestamped and cached in the test data set.

[0109] The initial voltage signal sequence is paired with the synchronously collected loop current value, and the instantaneous initial loop resistance value at each sampling time is calculated point by point according to Ohm's law to form an initial loop resistance sequence; and then it is aligned with the environmental parameters at the corresponding time according to the time index to form a test state tensor with a standard structure, consistent dimensions and synchronous time sequence. The test state tensor contains three types of features: the instantaneous loop resistance value reflecting the conductivity characteristic, the loop current value and the voltage sequence reflecting the excitation condition, and the environmental parameter reflecting the disturbance environment, which is used as the input of the dynamic compensation model for subsequent non-target resistance prediction and compensation correction. This process forms a complete closed loop in terms of measurement accuracy, data synchronization and disturbance modeling, ensuring the stability and reliability of the test results of the transformer loop resistance in a complex environment.

[0110] S140, inputting the test state tensor into the dynamic compensation model to output the non-target resistance compensation value at the current time.

[0111] In one possible implementation, the test state tensor is input into the dynamic compensation model to output the non-target resistance compensation value at the current time, specifically including: inputting the test state tensor into the dynamic compensation model, the dynamic compensation model being composed of multiple layers of long short-term memory units, the model input layer receiving the test state tensor and being transmitted to the long short-term memory unit in a time sequence manner, using a gating mechanism to extract the time sequence characteristics and amplitude response of the test disturbance, and generating a hidden state vector reflecting the dynamic change of the non-target resistance; obtaining an output resistance compensation value at the current time after the hidden state vector is processed by a fully connected mapping layer and a nonlinear activation function; and fusing the output resistance compensation values in the sliding prediction window by a weighted median filtering method to obtain the non-target resistance compensation value.

[0112] Specifically, in the implementation process, first, the test state tensor is input into the dynamic compensation model. The test state tensor is a multi-dimensional time sequence data structure containing the initial voltage signal sequence, loop current value and environmental parameter vector, representing the comprehensive state of the target transformer in the current test period. The dynamic compensation model is composed of multiple layers of long short-term memory units (Long Short-Term Memory Unit), which has the ability to model time-dependent relationships and nonlinear sequences. The model input layer rearranges the feature dimensions of the test state tensor to ensure that the tensor is gradually passed to each long short-term memory unit in the time dimension. Each long short-term memory unit is composed of an input gate, a forget gate, and an output gate, which are used to control the writing of current input information, the retention of old state information, and the generation of output results, respectively. The gating mechanism filters and compresses the disturbance features in the test state tensor by adjusting the activation function and weight matrix, extracts dynamic time sequence features representing non-target disturbances, and finally generates a corresponding hidden state vector at each time step. The hidden state vector is a compressed representation of the test disturbance in the time evolution process.

[0113] After the hidden state vector is generated, a fully connected mapping layer (Fully Connected Layer) is connected, which maps the hidden state vector to the target output space. The specific operation is to perform linear transformation on the hidden state through the weight matrix and bias vector to generate an intermediate representation, and then process it through a nonlinear activation function. Common nonlinear activation functions are ReLU (Rectified Linear Unit) or Tanh function, which can effectively enhance the model's fitting ability for nonlinear features. The final output is the predicted output resistance compensation value at the current time, which is used to approximately estimate the resistance offset caused by non-target factors (such as lead resistance, poor contact, electromagnetic interference, etc.) on the actual measurement. For example, if the actual value of the primary loop resistance is 200 μΩ, and a 20 μΩ deviation is introduced due to lead and environmental disturbances, the model output resistance compensation value should tend to be close to 20 μΩ.

[0114] Since the model prediction result may have short-term fluctuations on a single sample, to improve stability and robustness, a weighted median filtering method is introduced at the output end of the model to fuse the output resistance compensation values within a sliding prediction window. The sliding prediction window is set to cover several time steps before and after the current time, for example, expanding 2 time points forward and backward with the current time as the center, forming a 5-point sliding window. In this window, all output resistance compensation values are sorted by size, and the median value is extracted as the basis, and the edge points are set to a lower weight combined with the weighting coefficient, thereby realizing dynamic suppression of short-term abnormal values and prediction mutations. Weighted median filtering has better anti-exception ability than traditional mean filtering, and is suitable for electrical measurement scenarios affected by strong noise interference.

[0115] Finally, the compensation result after the weighted median filtering fusion processing is the non-target resistance compensation value, which truly reflects the resistance deviation caused by systematic disturbance under the current test state, and can be directly used for subsequent correction operation on the initial loop resistance, thereby improving the stability, accuracy and anti-interference ability of the test result of the target transformer loop resistance. The process realizes a complete non-target resistance compensation reasoning path from test data input, time sequence modeling, output prediction to steady-state fusion.

[0116] In S150, based on the non-target resistance compensation value, the initial loop resistance value calculated from the initial voltage signal sequence is corrected to obtain the compensation resistance value of the target transformer loop.

[0117] In a possible implementation, based on the non-target resistance compensation value, the initial loop resistance value calculated from the initial voltage signal sequence is corrected to obtain the compensation resistance value of the target transformer loop, specifically including: based on the initial voltage signal sequence and the loop current value, the initial loop resistance instantaneous value sequence in the uncorrected state is calculated; the non-target resistance compensation value corresponding to the initial loop resistance instantaneous value sequence in the time stamp is matched point by point, and the initial loop resistance instantaneous value sequence is deducted from the non-target resistance compensation value item by item to generate a compensation loop resistance instantaneous value sequence; the compensation loop resistance instantaneous value sequence is filtered to obtain the compensation resistance value of the target transformer loop.

[0118] Specifically, based on the initial voltage signal sequence and the corresponding loop current value, the initial loop resistance instantaneous value sequence in the uncorrected state is calculated. The initial voltage signal sequence is composed of continuous voltage sampling points in the constant current injection process, and the loop current value is the constant direct current collected at the same time step. According to Ohm's law, the loop resistance is equal to the voltage divided by the current, that is, at each sampling time, the voltage value at the time is divided by the corresponding current value to obtain the instantaneous initial loop resistance value. For example, if the voltage is 10 mV and the current is 50 A at t1, the instantaneous initial loop resistance value is 200 μΩ. Repeating the calculation process can form a complete initial loop resistance instantaneous value sequence, reflecting the loop conduction state without compensation processing in the test process.

[0119] Subsequently, the non-target resistance compensation value output by the dynamic compensation model is matched point by point to the initial loop resistance instantaneous value sequence. Since the non-target resistance compensation value corresponds one-to-one with the test state tensor output, its timestamp is completely consistent with the initial loop resistance instantaneous value. In the point-by-point matching process, for each time point, the corresponding non-target resistance compensation value is directly deducted from the initial loop resistance value to obtain the pure compensation value after removing the lead resistance, contact resistance and external electromagnetic interference. For example, if the initial loop resistance at a certain time is 210 μΩ and the model predicts a non-target resistance compensation value of 15 μΩ, then the compensation loop resistance value at that time is 195 μΩ. The processing result generates a compensation loop resistance instantaneous value sequence, which is used to express the real conduction performance of the loop after interference factors are excluded.

[0120] Finally, in order to improve the stability and reliability of the final output loop resistance, the compensation loop resistance instantaneous value sequence is filtered. The sliding average filter or Kalman filter is selected for filtering. The sliding average filter suppresses high-frequency noise by calculating the average of compensation values within a certain window, for example, setting the window length to 5, taking the average of the two compensation values before and after the current value and the current value to obtain the smoothed output value. If there are mutations or measurement spikes in the compensation value sequence, the Kalman filter can be used to establish a state prediction model, fuse the current measurement value and the predicted value according to the weight, and output the optimal estimation result. This filtering process not only eliminates the random fluctuations of single-point measurement, but also improves the trend stability of the resistance sequence.

[0121] The single output value after filtering is the compensation resistance value of the target transformer loop, which represents the real resistance level of the primary winding loop of the target transformer under the current test state and the identified disturbance condition. The compensation resistance value can be directly used for device contact integrity evaluation, conduction reliability discrimination or abnormality identification by comparing with the standard value in the account, and is a highly reliable electrical characteristic index in transformer operation decision-making. The overall process realizes a complete correction path from raw measurement value calculation, model compensation, error elimination to stable output.

[0122] In one possible implementation, after the initial loop resistance value calculated based on the non-target resistance compensation value is modified to obtain the compensation resistance value of the target transformer loop, the method further comprises: extracting a plurality of historical compensation resistance values within the current test environment parameter range, and constructing a time sequence sample containing the timestamp, loop current amplitude, environmental disturbance parameter and corresponding compensation resistance value; fitting the compensation resistance value using the exponential weighted moving average strategy according to the time sequence sample, constructing an error function based on the least mean square criterion; and combining the error function and the weight decay algorithm to correct the recurrent neural network weight matrix in the dynamic compensation model through back propagation, realizing adaptive parameter update of the dynamic compensation model under the current test environment.

[0123] Specifically, the historical compensation resistance value records matching the current test environment parameters are extracted to form a time series sample. Specifically, according to the environment parameter vector at the current test time, including temperature, humidity, power frequency, electromagnetic interference amplitude, voltage transient rate and noise spectrum energy density distribution, the historical test records similar to the environment parameter characteristics are retrieved in the historical database, and the similarity matching is realized by multidimensional Euclidean distance or Mahalanobis distance measurement. For each historical record that matches successfully, the corresponding compensation resistance value, loop current amplitude, environmental disturbance parameter and test time stamp are extracted to constitute a complete time series sample. The time series sample records the historical prediction results of the dynamic compensation model under the condition of approximate environmental disturbance, and provides reference data for subsequent model performance evaluation and adaptive update.

[0124] Subsequently, the compensation resistance value at the current test point in the time series sample is subjected to deviation fitting processing. An exponential weighted moving average strategy is adopted, and by assigning an exponential decay weight to the historical multiple compensation resistance values according to the time distance, the influence of the compensation value at the recent time on the current prediction result deviation is made greater, and a dynamic weighted deviation fitting curve is constructed. The process is realized by the formula:

[0125]

[0126] Among them, represents the current weighted fitting compensation resistance value, is the actual compensation resistance value at the current time, is a smoothing factor, usually set to 0.2 to 0.3. After obtaining the weighted fitting value, it is subjected to difference calculation with the actual measured current compensation resistance value to construct an error function. The error function is defined as the minimum mean square error criterion, that is, the target is to minimize the mean square error between the historical reference fitting value and the current model output value, and the formula is:

[0127]

[0128] Among them, is the error function value, is the model output value, is the weighted average fitting value, is the sample number.

[0129] Finally, based on the error function, the weight decay algorithm is combined to correct the recursive neural network weight matrix in the dynamic compensation model. The weight decay algorithm adds a weight L2 norm constraint term in the loss function to suppress the phenomenon of model overfitting caused by too large weight amplitude, and the corrected loss function is:

[0130]

[0131] in As the attenuation factor, For the first in the model Each weight, This represents the total number of weights involved in the update. Through the backpropagation algorithm of the error function, the weight parameters within the recurrent neural network are dynamically updated, making the model output more closely reflect the perturbation characteristics of the current test environment. This adaptive parameter update mechanism allows the model to adjust its response sensitivity to perturbation features online, achieving dynamic adaptability to complex test environments.

[0132] For example, at a substation site, if the current test compensation resistance value deviates significantly from the statistical mean under similar humid and hot conditions in the past, this deviation will guide the recurrent neural network to adjust its weight allocation for humidity or interference spectrum dimensions, making subsequent predictions more consistent with the non-target resistance behavior characteristics under specific disturbance scenarios, thereby improving the overall loop resistance correction accuracy and the intelligent robustness of the system.

[0133] This embodiment also discloses a dynamic compensation and correction device for the circuit resistance of a current transformer, referring to... Figure 2 The device includes an acquisition module 201, a processing module 202, and an output module 203. It is used to execute any of the above-described methods for dynamic compensation and correction of transformer loop resistance, wherein:

[0134] The acquisition module 201 is used to acquire the structural parameter information of the primary and secondary windings of the target transformer, and to collect the environmental parameters of the target transformer circuit to construct the structural feature tensor and the environmental disturbance tensor.

[0135] The processing module 202 is used to train a dynamic compensation model based on the structural feature tensor and the environmental disturbance tensor, combined with historical loop resistance test data. The dynamic compensation model is constructed based on a recurrent neural network.

[0136] The processing module 202 is used to inject a set DC current into the primary winding, obtain the initial voltage signal sequence at both ends of the primary winding, and construct a test state tensor by combining the current circuit current and real-time environmental parameters.

[0137] The processing module 202 is used to input the test state tensor into the dynamic compensation model and output the non-target resistance compensation value at the current moment.

[0138] The output module 203 is used to correct the initial loop resistance value calculated from the initial voltage signal sequence based on the non-target resistance compensation value, so as to obtain the compensation resistance value of the target transformer loop.

[0139] In a possible implementation, the acquisition module 201 is configured to extract a plurality of historical compensation resistance values in a current test environment parameter range, and construct a time sequence sample containing a timestamp, a loop current amplitude, an environmental interference parameter, and a corresponding compensation resistance value.

[0140] The processing module 202 is configured to perform bias fitting on the compensation resistance value according to the time sequence sample by using an exponential weighted moving average strategy, and construct an error function based on a minimum mean square error criterion.

[0141] The processing module 202 is configured to perform back propagation correction on a recursive neural network weight matrix in the dynamic compensation model in combination with the error function and a weight decay algorithm, so as to realize adaptive parameter updating of the dynamic compensation model under the current test environment.

[0142] In a possible implementation, the acquisition module 201 is configured to collect structural parameter information of a primary winding and a secondary winding of a target transformer, and the structural parameter information includes the number of turns of the primary winding, the diameter of the wire, the winding arrangement mode, the material resistivity, the core shape, the core magnetic permeability, the winding cross-sectional area, and the ratio coefficient, the tap position, and the internal insulation structure of the secondary winding.

[0143] The processing module 202 is configured to map the structural parameter information to a multi-dimensional structural feature space by performing discretization coding and standardization processing, so as to construct a structural feature tensor.

[0144] The acquisition module 201 is configured to collect environmental parameters of a loop of the target transformer, and the environmental parameters include a winding surface temperature, an environmental humidity, a power supply frequency, an electromagnetic interference amplitude, a voltage transient rate, and a noise frequency spectrum energy density distribution.

[0145] The processing module 202 is configured to extract statistical characteristic quantities of the environmental parameters based on a time synchronization window, and construct an environmental interference tensor.

[0146] The processing module 202 is configured to splice the structural feature tensor and the environmental interference tensor to form a composite input feature tensor as an input variable of the dynamic compensation model.

[0147] In a possible implementation, the processing module 202 is configured to construct a historical training sample set containing the structural feature tensor and the environmental interference tensor, and each item of data in the historical training sample set includes the structural feature tensor, the environmental interference tensor, and a corresponding historical loop resistance test value at a same time step.

[0148] The processing module 202 is configured to splice the structural feature tensor and the environmental interference tensor to form a composite input feature tensor as a model input, and construct a dynamic compensation model, where the dynamic compensation model is composed of a plurality of stacked long short-term memory units, and residual connection and gating mechanisms are introduced to enhance the historical disturbance dependent modeling capability.

[0149] The processing module 202 is configured to use mean square error as a loss function, use an adaptive learning rate optimization algorithm to iteratively update network weight parameters of the dynamic compensation model, and introduce a dropout strategy and an L2 regular term to prevent overfitting.

[0150] The processing module 202 is configured to divide the historical training sample set in a cross-validation manner to verify the model convergence performance during the training process, and complete parameter optimization and generalization training of the non-target resistance compensation prediction ability of the dynamic compensation model.

[0151] In a possible implementation, the processing module 202 is configured to control the constant current source control module to inject a preset direct current into the primary winding, and the preset direct current is determined according to a primary rated current range of the target transformer and an estimated loop resistance.

[0152] The processing module 202 is configured to continuously sample a voltage drop across the primary winding to form an initial voltage signal sequence.

[0153] The acquisition module 201 is configured to synchronously acquire a loop current value of a current loop current in real time, and the loop current value is obtained by a current sampling unit based on a low-impedance shunt.

[0154] The acquisition module 201 is configured to acquire an environmental parameter collected by an environmental perception module.

[0155] The output module 203 is configured to pair the initial voltage signal sequence with the loop current value to generate an instantaneous initial loop resistance sequence, and construct a test state tensor together with the environmental parameter.

[0156] In a possible implementation, the output module 203 is configured to input the test state tensor into a dynamic compensation model, the dynamic compensation model is composed of multiple layers of long short-term memory units, the model input layer receives the test state tensor and transmits the test state tensor to the long short-term memory units in a time sequence manner, extracts time sequence features and amplitude responses of a test disturbance by using a gating mechanism, and generates a hidden state vector reflecting dynamic changes of the non-target resistance.

[0157] The acquisition module 201 is configured to acquire an output resistance compensation value at a current moment based on the hidden state vector, a fully connected mapping layer, and a nonlinear activation function.

[0158] The processing module 202 is configured to fuse the output resistance compensation values in the sliding prediction window by using a weighted median filtering method to obtain a non-target resistance compensation value.

[0159] In a possible implementation, the processing module 202 is configured to calculate and obtain an initial loop resistance instantaneous value sequence in an uncorrected state based on the initial voltage signal sequence and the loop current value.

[0160] The processing module 202 is configured to match the non-target resistance compensation value corresponding to the initial loop resistance instantaneous value sequence in the time stamp point by point, and subtract the non-target resistance compensation value from the initial loop resistance instantaneous value sequence item by item to generate a compensated loop resistance instantaneous value sequence.

[0161] The output module 203 is configured to filter the compensated loop resistance instantaneous value sequence to obtain a compensated resistance value of the target mutual inductor loop.

[0162] It should be noted that: the device provided in the above embodiment is only exemplified by the above division of functional modules when realizing its function. In actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the above described functions. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is described in the method embodiment, which will not be repeated here.

[0163] The embodiment further discloses an electronic device, which refers to Figure 3 The electronic device can include: at least one processor 301, at least one communication bus 302, a user interface 303, a network interface 304, and at least one memory 305.

[0164] The communication bus 302 is configured to realize the connection and communication between the components.

[0165] The user interface 303 can include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 can further include a standard wired interface and a wireless interface.

[0166] The network interface 304 can optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0167] The processor 301 can include one or more processing cores. The processor 301 connects various parts within the server through various interfaces and lines, performs various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 305, and calling data stored in the memory 305. Alternatively, the processor 301 can be implemented in at least one of a hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 301 can integrate a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. Among them, the CPU mainly processes operating systems, user interfaces, and application programs. The GPU is responsible for rendering and drawing the content needed to be displayed on the display screen. The modem is used to process wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 301, but be realized by a separate chip.

[0168] The memory 305 can include a random access memory (RAM) and a read-only memory (ROM). Alternatively, the memory includes a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 can include a program storage area and a data storage area, wherein the program storage area can store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the above-mentioned various method embodiments, etc. The data storage area can store data involved in the above-mentioned various method embodiments, etc. The memory 305 can also be at least one storage device located away from the aforementioned processor 301. As a computer storage medium, the memory 305 can include an operating system, a network communication module, a user interface 303 module, and an application program of a kind of transformer circuit resistance dynamic compensation correction method.

[0169] In Figure 3The electronic device shown, the user interface 303 is mainly used for providing the interface for the user to input, obtaining the data input by the user. The processor 301 can be used to call the application program stored in the memory 305 and store a mutual inductor loop resistance dynamic compensation correction method, when executed by one or more processors 301, so that the electronic device executes the method of one or more of the above embodiments.

[0170] It should be noted that for the foregoing method embodiments, in order to simply describe, they are all described as a series of action combinations, but those skilled in the art should know that the present application is not limited to the order of the actions described, because according to the present application, certain steps can be performed in other order or at the same time. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0171] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0172] In several embodiments provided by the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic. The division of units is only a logical function division. There can be another division manner in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some services interfaces, devices or units, and can be electrical or other forms.

[0173] 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 can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0174] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present alone, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of software functional unit.

[0175] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable memory. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a memory 305 and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned memory 305 includes: a U disk, a mobile hard disk, a magnetic disk or an optical disk and various program code storage media.

[0176] The present application also discloses a computer readable storage medium, which stores instructions. When executed by one or more processors 301, the instructions cause an electronic device to perform one or more methods as described in the above embodiments.

[0177] The above are only exemplary embodiments of the present disclosure, and cannot limit the scope of the present disclosure. That is, any equivalent changes and modifications made in accordance with the teachings of the present disclosure are still within the scope of the present disclosure. Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon considering the specification and practicing the true principles of the present disclosure. The present application is intended to cover any variations, uses or adaptive changes of the present disclosure that follow the general principles of the present disclosure and include common knowledge or conventional techniques in the art that are not described in the present disclosure. The specification and examples are only considered as exemplary, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A mutual inductor loop resistance dynamic compensation correction method, characterized in that, The method comprises: obtaining the structure parameter information of the primary winding and the secondary winding of the target transformer, and collecting the environmental parameters of the target transformer loop to construct a structure feature tensor and an environmental interference tensor; based on the structure feature tensor and the environmental interference tensor, and combined with historical loop resistance test data, a dynamic compensation model is trained, and the dynamic compensation model is constructed based on a recurrent neural network; injecting a set direct current into the primary winding, obtaining an initial voltage signal sequence at both ends of the primary winding, combining the current loop and the real-time environmental parameters to construct a test state tensor; inputting the test state tensor into the dynamic compensation model to output the non-target resistance compensation value at the current time; based on the non-target resistance compensation value, the initial loop resistance value calculated from the initial voltage signal sequence is corrected to obtain the compensation resistance value of the target transformer loop; the non-target resistance compensation value corresponding to the initial loop resistance instantaneous value sequence in the time stamp is matched point by point, and the non-target resistance compensation value is deducted from the initial loop resistance instantaneous value sequence item by item to generate a compensation loop resistance instantaneous value sequence; filtering the compensation loop resistance instantaneous value sequence to obtain the compensation resistance value of the target transformer loop. after the non-target resistance compensation value is used to correct the initial loop resistance value calculated from the initial voltage signal sequence to obtain the compensation resistance value of the target transformer loop, the method further comprises: extracting a plurality of historical compensation resistance values within the current test environmental parameter range to construct a time series sample containing time stamp, loop current amplitude, environmental interference parameter and corresponding compensation resistance value; 2. The method for dynamic compensation of the transformer loop resistance according to claim 1, characterized in that, according to the time series sample, an exponential weighted moving average strategy is used to fit the deviation of the compensation resistance value, and an error function is constructed based on the least mean square criterion; combining the error function and the weight decay algorithm to perform back propagation correction on the recurrent neural network weight matrix in the dynamic compensation model to realize adaptive parameter update of the dynamic compensation model under the current test environment. the structure parameter information of the primary winding and the secondary winding of the target transformer is collected, and the environmental parameters of the target transformer loop are collected to construct a structure feature tensor and an environmental interference tensor, which comprises: collecting the structure parameter information of the primary winding and the secondary winding of the target transformer, the structure parameter information including the number of turns, wire diameter, winding arrangement, material resistivity, core shape, core magnetic permeability, winding cross-sectional area of the primary winding, and the ratio coefficient, tap position and internal insulation structure of the secondary winding; 3. The method of claim 1, wherein, through discrete coding and standardization processing of the structure parameter information, it is mapped to a multi-dimensional structure feature space to construct a structure feature tensor; ​ ​ Collecting environmental parameters of the target transformer circuit, the environmental parameters including winding surface temperature, environmental humidity, power frequency, electromagnetic interference amplitude, voltage transient rate and noise spectrum energy density distribution; Extracting statistical feature quantities of the environmental parameters based on a time synchronization window, and constructing an environmental interference tensor; Splicing the structural feature tensor and the environmental interference tensor to form a composite input feature tensor as an input variable of the dynamic compensation model.

4. The method of claim 1, wherein, The dynamic compensation model is trained based on the structural feature tensor and the environmental interference tensor, and combined with historical loop resistance test data, specifically including: A historical training sample set containing the structural feature tensor and the environmental interference tensor is constructed, each data in the historical training sample set including the structural feature tensor, the environmental interference tensor and the corresponding historical loop resistance test value at the same time step; The structural feature tensor and the environmental interference tensor are spliced to form a composite input feature tensor as the model input, and the dynamic compensation model is constructed, which is composed of multiple layers of long short-term memory units, and residual connection and gating mechanism are introduced to enhance the historical disturbance dependent modeling capability; The mean square error is used as the loss function, the adaptive learning rate optimization algorithm is used to iteratively update the network weight parameters of the dynamic compensation model, and the dropout strategy and L2 regularization term are introduced to prevent overfitting; In the training process, the historical training sample set is divided using cross-validation method to verify the model convergence performance, and the parameter optimization and generalization training of the non-target resistance compensation prediction ability of the dynamic compensation model are completed.

5. The method of claim 1, wherein, Injecting a set direct current into the primary winding, obtaining an initial voltage signal sequence at both ends of the primary winding, and constructing a test state tensor combined with the current loop current and real-time environmental parameters, specifically including: Controlling the constant current source control module to inject a preset direct current into the primary winding, the preset direct current being determined according to the primary rated current range and the estimated loop resistance of the target transformer; Continuously sampling the voltage drop at both ends of the primary winding to form an initial voltage signal sequence; Synchronously collecting the real-time loop current value of the current loop current, which is obtained by the current sampling unit based on a low-impedance shunt; Obtaining the environmental parameters collected by the environmental perception module; Pairing the initial voltage signal sequence with the loop current value to generate an instantaneous initial loop resistance sequence, and constructing the test state tensor together with the environmental parameters.

6. The method of claim 1, wherein, Inputting the test state tensor into the dynamic compensation model to output the non-target resistance compensation value at the current time, specifically including: Inputting the test state tensor into the dynamic compensation model, the dynamic compensation model being composed of multiple layers of long short-term memory units, the model input layer receiving the test state tensor and transmitting it to the long short-term memory unit in time series, extracting the time sequence features and amplitude response of the test disturbance using the gating mechanism, and generating a hidden state vector reflecting the dynamic changes of the non-target resistance; An output resistance compensation value at a current time is output after the hidden state vector is mapped through a full connection mapping layer and processed by a nonlinear activation function. The output resistance compensation values in the sliding prediction window are fused by a weighted median filtering method to obtain the non-target resistance compensation value.

7. A mutual inductor loop resistance dynamic compensation correction device, characterized in that, The device is used to execute the mutual inductor loop resistance dynamic compensation correction method in any one of claims 1-6, and the device comprises an acquisition module (201), a processing module (202), and an output module (203), wherein: The acquisition module (201) is used to acquire structure parameter information of a primary winding and a secondary winding of a target mutual inductor, and collect environment parameters of a target mutual inductor loop, and construct a structure feature tensor and an environment interference tensor; The processing module (202) is used to train a dynamic compensation model based on the structure feature tensor and the environment interference tensor and in combination with historical loop resistance test data, and the dynamic compensation model is constructed based on a recurrent neural network; The processing module (202) is used to inject a set direct current into the primary winding, acquire an initial voltage signal sequence at both ends of the primary winding, and construct a test state tensor in combination with a current loop and real-time environment parameters; The processing module (202) is used to input the test state tensor into the dynamic compensation model to output a non-target resistance compensation value at a current time; The output module (203) is used to correct an initial loop resistance value calculated from the initial voltage signal sequence based on the non-target resistance compensation value to obtain a compensation resistance value of the target mutual inductor loop.

8. An electronic device, comprising: The electronic device comprises a processor (301), a communication bus (302), a user interface (303), a network interface (304), and a memory (305), the memory (305) is used to store instructions, the user interface (303) and the network interface (304) are both used to communicate with other devices, the communication bus (302) is used to realize connection and communication between components in the electronic device, and the processor (301) is used to execute the instructions stored in the memory (305) to enable the electronic device to execute the method in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions, when the instructions are executed, the method in any one of claims 1-6 is executed.

Images