Magnetizing characteristic evaluation method and system for high-voltage shunt reactor based on direct current method

Abstract

This invention discloses a method and system for evaluating the magnetization characteristics of high-voltage parallel reactors based on the DC method. The method includes: setting a DC test scheme and building a reactor test network; conducting magnetization characteristic tests to obtain a multi-source test monitoring dataset; performing dynamic inductance and flux linkage sequence calculations to reconstruct a current-flux linkage curve characterizing the nonlinear magnetization characteristics of the core; extracting intrinsic magnetization features using the current-flux linkage curve and extracting environmental covariance features from the multi-source test monitoring dataset to evaluate the magnetization characteristics of the target high-voltage parallel reactor and generate magnetization characteristic evaluation information; acquiring magnetization test data of faulty reactors, modeling the correlation between fault modes and magnetization characteristics, and generating a fault-magnetization knowledge base; establishing a reactor state analysis model using the fault-magnetization knowledge base, and performing state diagnosis by combining the magnetization characteristic evaluation information and multi-modal test feature vectors. This achieves accurate evaluation of magnetization characteristics and intelligent diagnosis of reactor state.

Description

Technical Field

[0001] This invention relates to the field of reactor analysis technology, and in particular to a method and system for evaluating the magnetization characteristics of high-voltage parallel reactors based on the DC method. Background Technology

[0002] High-voltage shunt reactors, as crucial reactive power compensation devices in power systems, require accurate assessment of their magnetization characteristics to ensure stable grid operation. Currently, the AC method is primarily used in engineering practice for magnetization characteristic testing. This method establishes a magnetization curve by applying an AC excitation signal and measuring the response. However, the AC method has significant limitations in practical applications: First, the eddy current effect generated in the core under AC excitation leads to additional losses, causing the measured magnetization curve to deviate from the material's true characteristics; second, the existence of hysteresis effects affects the measurement results due to historical magnetization states; furthermore, the accuracy of the test results is easily affected by factors such as power frequency fluctuations and waveform distortion. These factors inherently limit the AC method's ability to accurately assess the magnetic properties of the core.

[0003] As an alternative, the DC method assesses the condition of reactor cores by applying DC voltage and observing transient processes, offering advantages such as simple equipment, high safety, and strong anti-interference capabilities. However, existing DC testing methods primarily focus on measuring basic parameters such as winding inductance and inter-turn insulation, with a relatively superficial evaluation of core magnetization characteristics. They lack a systematic approach to fully reconstruct the magnetization characteristic curve from transient response data and have failed to establish an effective quantitative correlation between magnetization characteristics and core health. Current technologies struggle to accurately extract the nonlinear characteristic parameters of the magnetization curve and cannot effectively identify early signs of core degradation, thus limiting the application of the DC method in the accurate evaluation of reactor conditions.

[0004] Therefore, a method and system for evaluating the magnetization characteristics of high-voltage parallel reactors based on the DC method is proposed. This method can completely reconstruct the core magnetization characteristic curve from the DC transient response data and achieve accurate diagnosis of the core health status, providing reliable technical support for reactor condition-based maintenance. Summary of the Invention

[0005] This invention overcomes the shortcomings of the prior art and provides a method and system for evaluating the magnetization characteristics of high-voltage parallel reactors based on the DC method.

[0006] To achieve the above objectives, the first aspect of this invention provides a method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method, comprising: A DC step test scheme was set up and a reactor test network was built based on the specification parameters of the high voltage parallel reactor to be tested. The magnetization characteristics were tested and the entire test process was monitored using the reactor test network to obtain a multi-source test monitoring dataset. The multi-source test and monitoring dataset is preprocessed, and dynamic inductance and flux linkage sequence calculations are performed after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core. Intrinsic magnetization features are extracted using current-magnetic flux curves, and environmental covariance features are extracted from multi-source test and monitoring datasets to construct multi-modal test feature vectors. The magnetization features of the target high-voltage parallel reactor are evaluated using the multi-modal test feature vectors to generate magnetization feature evaluation information. High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, and magnetization test data of faulty reactors are obtained by connecting them to a pre-built reactor test network. The correlation between fault modes and magnetization characteristics is modeled to generate a fault-magnetization knowledge base. A reactor condition analysis model is established using the fault-magnetization knowledge base, and the condition diagnosis of the target high-voltage parallel reactor is performed by combining the magnetization feature evaluation information and multi-mode test feature vectors.

[0007] In this scheme, the DC step test scheme is set according to the specifications of the high-voltage parallel reactor under test, and a reactor test network is built. The magnetization characteristics are tested using the reactor test network, and the entire test process is monitored to obtain a multi-source test monitoring dataset, which specifically includes: Obtain the specification sheet of the high-voltage parallel reactor to be tested, calculate the upper limit of the DC excitation voltage required to make the iron core reach the expected magnetization level based on the law of electromagnetic induction, determine the maximum allowable operating current of the test system through safety margin calculation based on the rated current value and insulation withstand voltage level, determine the optimal voltage-current combination parameters according to the power supply output characteristic curve, and generate a DC step test scheme. Subsequently, a reactor test network consisting of a programmable DC power supply, a power adjustable resistor array, a solid-state power switch, and a data acquisition system was built. The power adjustable resistor array adopts a multi-stage parallel structure to achieve fine adjustment of the resistance value. The solid-state power switch uses an insulated gate bipolar transistor and is equipped with an overcurrent protection circuit. The data acquisition system consists of a multi-source heterogeneous sensor array. After the reactor test network is built, a self-calibration process is performed on the reactor test network, including zero-point calibration and proportional coefficient calibration of the voltage measurement channel, frequency response characteristic testing of the current sensor, and thermocouple cold junction compensation calibration of the temperature monitoring unit to ensure the accuracy and synchronization of each measurement channel. In the test network of the target high-voltage parallel reactor and the calibrated reactor, the control unit outputs a DC excitation signal based on the DC step test scheme. After each current step reaches the target value, it enters the stable monitoring stage to monitor the fluctuation of the winding terminal voltage in real time. The sliding window algorithm is used to calculate the voltage change rate. When the voltage change rate of multiple consecutive sampling periods is lower than the preset stability judgment threshold, the steady-state operating point is confirmed. At this time, the steady-state voltage reading, current reading, and temperature data of each measuring point are recorded. During the transient process of current step change, the dynamic response waveforms of voltage and current are captured completely at the highest sampling rate. At the same time, the change trajectory of ambient temperature is recorded. Finally, the collected data are integrated to generate a multi-source test and monitoring dataset.

[0008] In this scheme, the preprocessing of the multi-source test and monitoring dataset, followed by dynamic inductance and flux linkage sequence calculation after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization properties of the iron core, specifically includes: Data quality improvement processing is performed on the original sampled signals in the multi-source test and monitoring dataset. Discrete wavelet transform is used to decompose the voltage and current signals into multiple scales. The wavelet coefficients are subjected to soft thresholding by setting an adaptive threshold to suppress high-frequency noise. At the same time, an outlier detection algorithm is used to identify abnormal temperature monitoring data and to compensate for missing values. Based on the preprocessed multi-source test and monitoring dataset, pure voltage and pure current sequences are extracted. The instantaneous rate of change of the current signal is calculated using a five-point center difference algorithm. Two adjacent sampling points are taken before and after each sampling point to construct a five-point difference template. The weighted average is calculated using preset difference coefficients to obtain the rate of change of current. The current rate of change is then divided point by point with the voltage measurement value at the corresponding time to obtain the dynamic inductance sequence. Simultaneously calculate the flux linkage parameters, use the composite Simpson numerical integration algorithm to numerically integrate the pure voltage sequence, divide the entire integration interval into several equal intervals, approximate the integrand function in each sub-interval using a quadratic interpolation polynomial, and calculate the cumulative flux linkage value at each sampling time by weighted summation to obtain the flux linkage change sequence. The calculated dynamic inductance sequence and flux linkage change sequence are spatiotemporally aligned and fused. A time-scale alignment algorithm based on cross-correlation function is used to correct the sampling time deviation between the two sequences. The dynamic inductance sequence is resampled to the corresponding time point of the flux linkage change sequence through Hermit interpolation algorithm to establish the mapping relationship between dynamic inductance value and flux linkage value, forming a dynamic inductance-flux linkage feature sequence arranged in order of flux linkage value.

[0009] In this scheme, the preprocessing of the multi-source test and monitoring dataset, followed by dynamic inductance and flux linkage sequence calculation to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization properties of the iron core, further includes: A dynamic inductance-magnetic flux linkage feature sequence is obtained. A sliding window curvature detection algorithm is used to identify the characteristic inflection points of the magnetization curve. Based on a preset window size, the curvature value of the neighboring region is calculated at each data point in the dynamic inductance-magnetic flux linkage feature sequence. When the curvature value of 5 consecutive data points exceeds a preset curvature threshold, it is determined as the boundary point between the linear region and the saturation region, and the sequence segmentation result is generated. Based on the sequence segmentation results, the current optimal linear parameters, including slope and intercept, are solved by constructing a system of normal equations and using the Cholesky decomposition algorithm. In the saturation region, the hyperbolic tangent function is selected for nonlinear fitting, and the Levenberg-Marquardt algorithm is used to solve the nonlinear fitting coefficients. An initial magnetization characteristic model is established to obtain an initial reconstructed parameter combination containing linear and nonlinear parameters. A genetic algorithm is introduced to globally optimize the initial reconstructed parameter combination, and an objective function including a fitting error term and a curve smoothness term is constructed. The fitting error term is characterized by the root mean square error between the measured data and the fitted curve, and the curve smoothness term is defined by the sum of squares of the differences of the first derivatives at the connection points. During the optimization process, the constraints are that the piecewise functions have the same function value at the connection point and the difference in the first derivative value is less than a preset range. Genetic operations are performed by roulette wheel selection algorithm, arithmetic crossover operator and Gaussian mutation operator. The best individual is retained in each generation and an elite retention strategy is adopted. After multiple generations of evolutionary calculation, the optimal combination of reconstruction parameters is obtained. Based on the optimal parameter combination, the complete current-magnetic flux linkage curve is reconstructed. The fitting results of the linear region and the saturation region are combined in the form of a piecewise function, and the residual of each sampling point is calculated. The distribution characteristics of the residual are analyzed to verify the rationality of the reconstruction, and finally the current-magnetic flux linkage curve is obtained.

[0010] In this scheme, the intrinsic magnetization characteristics are extracted using the current-flux flux curve, and environmental covariance characteristics are extracted from the multi-source test and monitoring dataset to construct a multi-modal test feature vector. This multi-modal test feature vector is then used to evaluate the magnetization characteristics of the target high-voltage parallel reactor, generating magnetization characteristic evaluation information. Specifically, this includes: The current-magnetic flux characteristic curve is obtained, and the magnetization characteristic parameters are extracted using an automatic feature point detection algorithm. The differential permeability of the curve at each operating point is calculated using the five-point central difference method. The saturation flux value and the corresponding saturation current value are identified using the golden section search algorithm. The coercivity characteristic points are solved using the cubic spline interpolation method. At the same time, the permeability slope in the linear region and the permeability attenuation coefficient in the saturation region are calculated, and the hysteresis loop area and remanence ratio parameters are extracted to form the intrinsic magnetization characteristic vector. Simultaneously, environmental covariance features are extracted from the multi-source test and monitoring dataset. The average temperature, temperature change gradient, and steady-state temperature value throughout the test are extracted as thermal environment features. Spectral analysis is performed on the background electromagnetic noise data to extract the amplitude of the main interference frequency components as electromagnetic environment features. Together, they constitute the environmental covariance feature vector. The intrinsic magnetization feature vector and the environmental covariance feature vector are fused in a multimodal manner. First, the Z-score normalization method is used to normalize the feature vector and map it to a high-dimensional space. Then, the principal component score of each feature is calculated by principal component analysis. The principal component features with a score greater than the preset threshold are retained to generate a multimodal test feature vector. A magnetization characteristic evaluation model was constructed using SVDD. Magnetization characteristic data of high-voltage parallel reactors in healthy and abnormal states were obtained by retrieving historical data. The magnetization characteristic evaluation model was then trained to obtain the minimum hypersphere boundary. Based on the boundary of the minimum hypersphere, the healthy state and the abnormal state are divided. The state inside the boundary of the minimum hypersphere is the healthy state, and the state outside the boundary of the minimum hypersphere is the abnormal state. At the same time, the center of the hypersphere is used as the reference point for the healthy state, and the distance to the center of the hypersphere is used as the score. The magnetization characteristic evaluation model is repeatedly trained. The multimodal test feature vector is input into the magnetization characteristic evaluation model after training to calculate the Mahalanobis distance with the center of the hypersphere, and compared with the preset hypersphere radius. If it is less than the preset hypersphere radius, it represents a healthy state of magnetization characteristics. If it is greater than the preset hypersphere radius, it represents an abnormal state of magnetization characteristics. The magnetization characteristic score is output based on the inverse ratio of the corresponding Mahalanobis distance, and finally the magnetization characteristic evaluation information is obtained.

[0011] In this solution, the process of acquiring high-voltage parallel reactors of the same specifications with different fault types and severity, connecting them to a pre-built reactor testing network to obtain magnetization test data of faulty reactors, modeling the correlation between fault modes and magnetization characteristics, and generating a fault-magnetization knowledge base specifically includes: High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, connected to a pre-built reactor test network, and magnetization characteristics are tested according to the set DC step test scheme to obtain the magnetization test dataset of the faulty reactors. Based on the fault reactor magnetization test dataset, intrinsic magnetization feature vectors and environmental covariance feature vectors are extracted from the current-magnetic flux linkage curves of each fault reactor magnetization test sample. A fault feature vector set is generated through multimodal feature fusion processing, and each fault sample is labeled with fault type and fault degree. The fault feature vector set is mapped to a high-dimensional vector space. A feature space similarity matrix is ​​constructed by calculating the Euclidean distance between vectors. The DBSCAN clustering algorithm is introduced to identify the distribution pattern of sample points in the feature space. The sample points with density connection are divided into the same cluster by the preset neighborhood radius and minimum number of samples. At the same time, noise points in sparse areas are identified and removed to generate several sample clusters that represent different fault modes. Geometric feature analysis is performed on each cluster to calculate the centroid location, distribution range, and mean and variance statistical indicators of each dimension of the samples within the cluster. A mapping table between failure modes and feature clusters is established. The internal compactness and inter-cluster separation of each cluster are evaluated by cluster effectiveness indicators to optimize cluster merging and ultimately form feature clusters for different failure modes. Subsequently, diagnostic rules are generated based on the feature clustering results. Multi-dimensional statistical distribution analysis is performed on the sample features within each cluster. The probability distribution parameters of each feature dimension are calculated. The effective value range of each feature dimension is determined by the interval estimation method. Feature judgment rules are extracted by the decision tree induction algorithm to generate a multi-level decision rule set. Finally, the multi-level decision rule set and the feature clustering of different fault modes are integrated to generate a fault-magnetization knowledge base.

[0012] In this solution, the step of establishing a reactor condition analysis model using the fault-magnetization knowledge base, and combining the magnetization feature evaluation information and multi-mode test feature vectors to perform condition diagnosis on the target high-voltage parallel reactor under test, specifically includes: A reactor state analysis model was constructed using the K-nearest neighbor classification algorithm and the random forest algorithm. The main classifier was trained using the random forest algorithm, and the node splitting was optimized by the Gini coefficient. At the same time, a K-nearest neighbor auxiliary classifier was constructed, and the feature-weighted Euclidean distance was used as the similarity measure. The two classifiers shared the same training set, and finally the reactor state analysis model that met the expectations was obtained. Magnetization feature evaluation information and multimodal test feature vectors are obtained and input into the reactor state analysis model. The main classifier performs multi-tree voting prediction through random forest and outputs preliminary fault type identification results and confidence scores. The auxiliary classifier adopts a dynamic K-value strategy. It determines the optimal K-value by analyzing the sample distribution density in the feature space through the elbow rule, calculates the weighted Euclidean distance between the multimodal test feature vector and the centroids of various faults in the knowledge base, selects the K nearest neighbor samples for distance-weighted voting, and generates auxiliary recognition results. The classifier outputs are fused for decision-making. When the confidence of the main classifier is higher than the preset threshold, the preliminary fault type identification result is directly adopted. When it is lower than the preset threshold, the Jaccard similarity coefficient between the outputs of the main classifier and the auxiliary classifier is calculated based on the conflict resolution mechanism. The health score in the magnetization feature evaluation information is combined for weighted arbitration to generate the final fault type identification result. Based on the fault type identification results, the corresponding multi-level decision rule set in the fault-magnetization knowledge base is called for state verification. The matching degree between the multi-modal test feature vector and the rule preconditions is calculated step by step. The rule weight is dynamically adjusted using the information gain ratio. The decision path is traversed by depth-first search, and the feature deviation degree and rule matching status of each node are recorded. Finally, the fault diagnosis information is obtained and a reactor state diagnosis report is generated and pushed.

[0013] A second aspect of the present invention provides a system for evaluating the magnetization characteristics of a high-voltage shunt reactor based on a DC method. The system includes a memory, a processor, and a communication interface. The memory contains a program for evaluating the magnetization characteristics of a high-voltage shunt reactor based on a DC method. When the processor executes the program for evaluating the magnetization characteristics of a high-voltage shunt reactor based on a DC method, it performs the following steps: A DC step test scheme was set up and a reactor test network was built based on the specification parameters of the high voltage parallel reactor to be tested. The magnetization characteristics were tested and the entire test process was monitored using the reactor test network to obtain a multi-source test monitoring dataset. The multi-source test and monitoring dataset is preprocessed, and dynamic inductance and flux linkage sequence calculations are performed after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core. Intrinsic magnetization features are extracted using current-magnetic flux curves, and environmental covariance features are extracted from multi-source test and monitoring datasets to construct multi-modal test feature vectors. The magnetization features of the target high-voltage parallel reactor are evaluated using the multi-modal test feature vectors to generate magnetization feature evaluation information. High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, and magnetization test data of faulty reactors are obtained by connecting them to a pre-built reactor test network. The correlation between fault modes and magnetization characteristics is modeled to generate a fault-magnetization knowledge base. A reactor condition analysis model is established using the fault-magnetization knowledge base, and the condition diagnosis of the target high-voltage parallel reactor is performed by combining the magnetization feature evaluation information and multi-mode test feature vectors.

[0014] This invention discloses a method and system for evaluating the magnetization characteristics of high-voltage parallel reactors based on the DC method. The method includes: setting a DC test scheme and building a reactor test network; conducting magnetization characteristic tests to obtain a multi-source test monitoring dataset; performing dynamic inductance and flux linkage sequence calculations to reconstruct a current-flux linkage curve characterizing the nonlinear magnetization characteristics of the core; extracting intrinsic magnetization features using the current-flux linkage curve and extracting environmental covariance features from the multi-source test monitoring dataset to evaluate the magnetization characteristics of the target high-voltage parallel reactor and generate magnetization characteristic evaluation information; acquiring magnetization test data of faulty reactors, modeling the correlation between fault modes and magnetization characteristics, and generating a fault-magnetization knowledge base; establishing a reactor state analysis model using the fault-magnetization knowledge base, and performing state diagnosis by combining the magnetization characteristic evaluation information and multi-modal test feature vectors. This achieves accurate evaluation of magnetization characteristics and intelligent diagnosis of reactor state. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0016] Figure 1 A flowchart of the first method for evaluating the magnetization characteristics of a high-voltage parallel reactor based on the DC method, provided in an embodiment of the present invention; Figure 2 A flowchart of the second method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method, provided as an embodiment of the present invention; Figure 3 A block diagram of a high-voltage shunt reactor magnetization characteristic evaluation system based on the DC method is provided in one embodiment of the present invention. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0019] Figure 1 A flowchart of the first method for evaluating the magnetization characteristics of a high-voltage parallel reactor based on the DC method, provided in an embodiment of the present invention; like Figure 1 As shown, the present invention provides a first method flowchart for evaluating the magnetization characteristics of high-voltage shunt reactors based on the DC method, including: S102, set up a DC step test scheme and build a reactor test network based on the specification parameters of the high voltage parallel reactor to be tested, use the reactor test network to test the magnetization characteristics and monitor the entire test process to obtain a multi-source test monitoring dataset. S104, preprocess the multi-source test and monitoring dataset, and after the preprocessing is completed, perform dynamic inductance and flux linkage sequence calculation to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core. S106. The intrinsic magnetization features are extracted using the current-magnetic flux curve, and the environmental covariance features are extracted using the multi-source test and monitoring dataset to construct a multi-mode test feature vector. The magnetization features of the target high-voltage parallel reactor are evaluated using the multi-mode test feature vector, and magnetization feature evaluation information is generated. S108: Obtain high-voltage parallel reactors of the same specification with different fault types and fault degrees, connect them to the pre-built reactor test network to obtain magnetization test data of faulty reactors, model the correlation between fault modes and magnetization characteristics, and generate a fault-magnetization knowledge base. S110, using the fault-magnetization knowledge base to establish a reactor condition analysis model, and combining the magnetization feature evaluation information and multi-mode test feature vectors to perform condition diagnosis on the target high-voltage parallel reactor under test.

[0020] Furthermore, in a preferred embodiment of the present invention, the step of setting a DC step test scheme and building a reactor test network based on the specifications of the high-voltage parallel reactor under test, and using the reactor test network to perform magnetization characteristic testing and monitor the entire test process to obtain a multi-source test monitoring dataset, specifically includes: Obtain the specification sheet of the high-voltage parallel reactor to be tested, calculate the upper limit of the DC excitation voltage required to make the iron core reach the expected magnetization level based on the law of electromagnetic induction, determine the maximum allowable operating current of the test system through safety margin calculation based on the rated current value and insulation withstand voltage level, determine the optimal voltage-current combination parameters according to the power supply output characteristic curve, and generate a DC step test scheme. Subsequently, a reactor test network consisting of a programmable DC power supply, a power adjustable resistor array, a solid-state power switch, and a data acquisition system was built. The power adjustable resistor array adopts a multi-stage parallel structure to achieve fine adjustment of the resistance value. The solid-state power switch uses an insulated gate bipolar transistor and is equipped with an overcurrent protection circuit. The data acquisition system consists of a multi-source heterogeneous sensor array. After the reactor test network is built, a self-calibration process is performed on the reactor test network, including zero-point calibration and proportional coefficient calibration of the voltage measurement channel, frequency response characteristic testing of the current sensor, and thermocouple cold junction compensation calibration of the temperature monitoring unit to ensure the accuracy and synchronization of each measurement channel. In the test network of the target high-voltage parallel reactor and the calibrated reactor, the control unit outputs a DC excitation signal based on the DC step test scheme. After each current step reaches the target value, it enters the stable monitoring stage to monitor the fluctuation of the winding terminal voltage in real time. The sliding window algorithm is used to calculate the voltage change rate. When the voltage change rate of multiple consecutive sampling periods is lower than the preset stability judgment threshold, the steady-state operating point is confirmed. At this time, the steady-state voltage reading, current reading, and temperature data of each measuring point are recorded. During the transient process of current step change, the dynamic response waveforms of voltage and current are captured completely at the highest sampling rate. At the same time, the change trajectory of ambient temperature is recorded. Finally, the collected data are integrated to generate a multi-source test and monitoring dataset.

[0021] It should be noted that, based on the specifications of the high-voltage shunt reactor under test, the first step involves analyzing key parameters such as rated voltage, winding resistance, and core cross-sectional area. The upper limit of the DC excitation voltage required for the core to reach the expected magnetization level is calculated using the law of electromagnetic induction. Simultaneously, considering the equipment's insulation withstand voltage rating and rated current, a safety margin factor is introduced to determine the maximum allowable operating current of the test system. Based on this, and combined with the output characteristic curve of the programmable DC power supply, the voltage-current operating point combination is optimized, and a DC step test scheme including starting current, ending current, step size, and hold time for each step is developed. Subsequently, a complete reactor test network is constructed. This system consists of a programmable DC power supply, a power-adjustable resistor array with a multi-stage parallel structure for fine adjustment, a solid-state power switch based on an insulated-gate bipolar transistor design and equipped with overcurrent protection circuitry, and a data acquisition system containing multi-source heterogeneous sensors for voltage, current, and temperature, forming a stable and reliable test platform.

[0022] After the test network was built, a self-calibration process was executed. Zero-point calibration and proportional coefficient calibration were performed on the voltage measurement channels, frequency response characteristic tests were conducted on the current sensor, and thermocouple cold junction compensation calibration was performed on the temperature monitoring unit. Synchronization adjustments to each measurement channel ensured the consistency of data acquisition time. The target reactor was connected to the calibrated test network. The control unit output an excitation signal according to the DC step test scheme. After each current step reached a set value, a stable monitoring phase was initiated, continuously monitoring the fluctuation of the winding terminal voltage. A sliding window algorithm was used to calculate the voltage change rate in real time. When the change rate for multiple consecutive sampling cycles was lower than the preset stability threshold, the system was confirmed to have reached a steady-state operating point. At this point, steady-state voltage, current readings, and temperature data at each measurement point were recorded. During the transient process of current step changes, a high-speed acquisition mode was activated to fully capture the dynamic response waveforms of voltage and current, simultaneously recording the ambient temperature change trajectory. Finally, through data integration and processing, a multi-source test monitoring dataset containing electrical parameter sequences, temperature sequences, and timestamp information was formed, providing a complete and reliable raw data foundation for subsequent magnetization characteristic analysis.

[0023] Furthermore, in a preferred embodiment of the present invention, the preprocessing of the multi-source test monitoring dataset, followed by dynamic inductance and flux linkage sequence calculation after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization properties of the iron core, specifically includes: Data quality improvement processing is performed on the original sampled signals in the multi-source test and monitoring dataset. Discrete wavelet transform is used to decompose the voltage and current signals into multiple scales. The wavelet coefficients are subjected to soft thresholding by setting an adaptive threshold to suppress high-frequency noise. At the same time, an outlier detection algorithm is used to identify abnormal temperature monitoring data and to compensate for missing values. Based on the preprocessed multi-source test and monitoring dataset, pure voltage and pure current sequences are extracted. The instantaneous rate of change of the current signal is calculated using a five-point center difference algorithm. Two adjacent sampling points are taken before and after each sampling point to construct a five-point difference template. The weighted average is calculated using preset difference coefficients to obtain the rate of change of current. The current rate of change is then divided point by point with the voltage measurement value at the corresponding time to obtain the dynamic inductance sequence. Simultaneously calculate the flux linkage parameters, use the composite Simpson numerical integration algorithm to numerically integrate the pure voltage sequence, divide the entire integration interval into several equal intervals, approximate the integrand function in each sub-interval using a quadratic interpolation polynomial, and calculate the cumulative flux linkage value at each sampling time by weighted summation to obtain the flux linkage change sequence. The calculated dynamic inductance sequence and flux linkage change sequence are spatiotemporally aligned and fused. A time-scale alignment algorithm based on cross-correlation function is used to correct the sampling time deviation between the two sequences. The dynamic inductance sequence is resampled to the corresponding time point of the flux linkage change sequence through Hermit interpolation algorithm to establish the mapping relationship between dynamic inductance value and flux linkage value, forming a dynamic inductance-flux linkage feature sequence arranged in order of flux linkage value.

[0024] It should be noted that data quality enhancement processing is performed on the original sampled signals in the multi-source test and monitoring dataset. First, discrete wavelet transform is used to decompose the voltage and current signals into multiple scales. Adaptive thresholds are set by analyzing the characteristics of wavelet coefficients at each scale, and soft thresholding is applied to the wavelet coefficients to effectively suppress high-frequency noise interference. At the same time, an outlier detection algorithm based on statistical principles is used to identify outliers in the temperature monitoring data, and linear interpolation is used to compensate for and repair missing data to ensure the integrity and reliability of the data quality. Based on the preprocessed dataset, pure voltage and pure current sequences are extracted. The instantaneous rate of change of the current signal is calculated using a five-point central difference algorithm. Two adjacent sampling points are taken before and after each sampling point to construct a five-point difference template. The accurate rate of change of the current is calculated by weighted averaging using preset difference coefficients. This rate of change is then divided point by point with the voltage measurement value at the corresponding time to derive the dynamic inductance sequence, which accurately reflects the changes in the electromagnetic characteristics of the reactor at different operating points.

[0025] Simultaneously, flux linkage parameters are calculated. A composite Simpson numerical integration algorithm is used to integrate the preprocessed pure voltage sequence. The entire integration interval is divided into several equal-numerical intervals. In each sub-interval, a quadratic interpolation polynomial is used to approximate the integrand. A weighted summation method is employed to calculate the cumulative flux linkage value at each sampling time. Reasonable initial conditions are set during the integration process to eliminate accumulated errors, resulting in an accurate flux linkage change sequence. The calculated dynamic inductance sequence and flux linkage change sequence undergo spatiotemporal alignment and data fusion. First, a time-scale alignment algorithm based on cross-correlation functions is used to correct the sampling time deviation between the two sequences, eliminating data asynchrony caused by sampling delay. Then, a Hermitian interpolation algorithm is used to resample the dynamic inductance sequence to the corresponding time points of the flux linkage change sequence, ensuring strict synchronization of the two sequences in the time domain. Finally, a one-to-one mapping relationship between dynamic inductance values ​​and flux linkage values ​​is established, forming a dynamic inductance-flux linkage characteristic sequence arranged in order of flux linkage values.

[0026] Furthermore, in a preferred embodiment of the present invention, the step of preprocessing the multi-source test monitoring dataset, and then performing dynamic inductance and flux linkage sequence calculations after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core, further includes: A dynamic inductance-magnetic flux linkage feature sequence is obtained. A sliding window curvature detection algorithm is used to identify the characteristic inflection points of the magnetization curve. Based on a preset window size, the curvature value of the neighboring region is calculated at each data point in the dynamic inductance-magnetic flux linkage feature sequence. When the curvature value of 5 consecutive data points exceeds a preset curvature threshold, it is determined as the boundary point between the linear region and the saturation region, and the sequence segmentation result is generated. Based on the sequence segmentation results, the current optimal linear parameters, including slope and intercept, are solved by constructing a system of normal equations and using the Cholesky decomposition algorithm. In the saturation region, the hyperbolic tangent function is selected for nonlinear fitting, and the Levenberg-Marquardt algorithm is used to solve the nonlinear fitting coefficients. An initial magnetization characteristic model is established to obtain an initial reconstructed parameter combination containing linear and nonlinear parameters. A genetic algorithm is introduced to globally optimize the initial reconstructed parameter combination, and an objective function including a fitting error term and a curve smoothness term is constructed. The fitting error term is characterized by the root mean square error between the measured data and the fitted curve, and the curve smoothness term is defined by the sum of squares of the differences of the first derivatives at the connection points. During the optimization process, the constraints are that the piecewise functions have the same function value at the connection point and the difference in the first derivative value is less than a preset range. Genetic operations are performed by roulette wheel selection algorithm, arithmetic crossover operator and Gaussian mutation operator. The best individual is retained in each generation and an elite retention strategy is adopted. After multiple generations of evolutionary calculation, the optimal combination of reconstruction parameters is obtained. Based on the optimal parameter combination, the complete current-magnetic flux linkage curve is reconstructed. The fitting results of the linear region and the saturation region are combined in the form of a piecewise function, and the residual of each sampling point is calculated. The distribution characteristics of the residual are analyzed to verify the rationality of the reconstruction, and finally the current-magnetic flux linkage curve is obtained.

[0027] It should be noted that, based on the dynamic inductance-magnetic flux linkage feature sequence, a sliding window curvature detection algorithm is used to automatically identify the characteristic inflection regions of the magnetization curve. By setting an appropriate window size, the curvature value of the neighboring region is calculated at each data point in the sequence. When the curvature values ​​of multiple consecutive data points exceed a preset threshold, the region is determined to be the characteristic boundary between the linear region and the saturation region, generating a sequence segmentation result. Subsequently, mathematical models for the linear region and the saturation region are established based on the segmentation results. In the linear region, a system of normal equations is constructed and the Cholesky decomposition algorithm is used to solve for the optimal linear parameters, including key parameters such as slope and intercept. In the saturation region, a hyperbolic tangent function with saturation characteristics is selected for nonlinear fitting, and the Levenberg-Marquardt algorithm is used to iteratively solve for the nonlinear fitting coefficients, establishing an initial segmented model of magnetization characteristics and obtaining an initial reconstructed parameter combination containing linear and nonlinear parameters.

[0028] To improve the accuracy of curve reconstruction, a genetic algorithm is introduced to globally optimize the initial parameter combination, constructing a multi-objective optimization function that includes a fitting error term and a curve smoothness term. The fitting error term is quantified by calculating the root mean square error between the measured data and the fitted curve, while the curve smoothness term is defined by evaluating the continuity of the first derivative at the connection points. Constraints are set during the optimization process, requiring each piecewise function to have continuous function values ​​and first derivative values ​​at the connection points. Genetic evolution is performed using a roulette wheel selection mechanism, arithmetic crossover, and Gaussian mutation operator. An elite retention strategy is employed in each generation to ensure the inheritance of superior characteristics. After multiple generations of iterative calculations, the optimal parameter combination is obtained. Based on the optimized parameter combination, the complete current-magnetic flux linkage characteristic curve is reconstructed. A piecewise function is used to smoothly connect the fitting results in the linear and saturation regions, ensuring that the curve maintains continuous differentiability throughout the overall range. Finally, the reconstruction results are verified and analyzed. The residual value of each sampling point is calculated, and the curve fitting quality is evaluated by analyzing the distribution characteristics and statistical properties of the residuals. This includes checking the randomness and systematic bias of the residuals to ensure that the reconstruction curve accurately reflects the true magnetization characteristics of the reactor core, providing a reliable data foundation for subsequent magnetization feature extraction and condition diagnosis.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of extracting intrinsic magnetization features using current-flux flux curves and constructing a multi-modal test feature vector by extracting environmental covariance features from a multi-source test monitoring dataset, and then using the multi-modal test feature vector to evaluate the magnetization features of the target high-voltage parallel reactor and generate magnetization feature evaluation information, specifically includes: The current-magnetic flux characteristic curve is obtained, and the magnetization characteristic parameters are extracted using an automatic feature point detection algorithm. The differential permeability of the curve at each operating point is calculated using the five-point central difference method. The saturation flux value and the corresponding saturation current value are identified using the golden section search algorithm. The coercivity characteristic points are solved using the cubic spline interpolation method. At the same time, the permeability slope in the linear region and the permeability attenuation coefficient in the saturation region are calculated, and the hysteresis loop area and remanence ratio parameters are extracted to form the intrinsic magnetization characteristic vector. Simultaneously, environmental covariance features are extracted from the multi-source test and monitoring dataset. The average temperature, temperature change gradient, and steady-state temperature value throughout the test are extracted as thermal environment features. Spectral analysis is performed on the background electromagnetic noise data to extract the amplitude of the main interference frequency components as electromagnetic environment features. Together, they constitute the environmental covariance feature vector. The intrinsic magnetization feature vector and the environmental covariance feature vector are fused in a multimodal manner. First, the Z-score normalization method is used to normalize the feature vector and map it to a high-dimensional space. Then, the principal component score of each feature is calculated by principal component analysis. The principal component features with a score greater than the preset threshold are retained to generate a multimodal test feature vector. A magnetization characteristic evaluation model was constructed using SVDD. Magnetization characteristic data of high-voltage parallel reactors in healthy and abnormal states were obtained by retrieving historical data. The magnetization characteristic evaluation model was then trained to obtain the minimum hypersphere boundary. Based on the boundary of the minimum hypersphere, the healthy state and the abnormal state are divided. The state inside the boundary of the minimum hypersphere is the healthy state, and the state outside the boundary of the minimum hypersphere is the abnormal state. At the same time, the center of the hypersphere is used as the reference point for the healthy state, and the distance to the center of the hypersphere is used as the score. The magnetization characteristic evaluation model is repeatedly trained. The multimodal test feature vector is input into the magnetization characteristic evaluation model after training to calculate the Mahalanobis distance with the center of the hypersphere, and compared with the preset hypersphere radius. If it is less than the preset hypersphere radius, it represents a healthy state of magnetization characteristics. If it is greater than the preset hypersphere radius, it represents an abnormal state of magnetization characteristics. The magnetization characteristic score is output based on the inverse ratio of the corresponding Mahalanobis distance, and finally the magnetization characteristic evaluation information is obtained.

[0030] It should be noted that, based on the current-magnetic flux linkage characteristic curve, a feature point automatic detection algorithm is used to systematically extract magnetization characteristic parameters. The differential permeability of the curve at each operating point is accurately calculated using the five-point central difference method. The golden section search algorithm is used to quickly locate the saturation flux linkage value and its corresponding saturation current value within a preset range. The coercivity characteristic points are solved using the cubic spline interpolation method. Simultaneously, the permeability slope in the linear region and the permeability attenuation coefficient in the saturation region are calculated, and key parameters such as the hysteresis loop area and remanence ratio are extracted. These parameters, reflecting the essential magnetization characteristics of the iron core, are integrated to form the intrinsic magnetization feature vector. Meanwhile, environmental covariance features are extracted from the multi-source test and monitoring dataset, including calculating the average temperature throughout the test, analyzing the temperature change gradient curve, recording the steady-state temperature value as a thermal environment characteristic parameter, and performing spectral analysis on the background electromagnetic noise data to extract the amplitude characteristics of the main interference frequency components as electromagnetic environment assessment indicators. These environmental influencing factors together constitute the environmental covariance feature vector.

[0031] Multimodal feature fusion processing is performed on the intrinsic magnetization feature vector and the environmental covariance feature vector. First, Z-score normalization is used to normalize each feature component to eliminate the influence of dimensions. Then, kernel function mapping is used to transform the feature vector to a high-dimensional space to enhance feature separability. Principal component analysis is used to calculate the principal component scores of each feature. Based on a preset variance contribution rate threshold, the main feature components are selected and retained to generate multimodal test feature vectors. A magnetization characteristic evaluation model is constructed using the Support Vector Data Description (SVDD) algorithm. Magnetization feature data of high-voltage parallel reactors in healthy and abnormal states are retrieved from the historical database as training samples. The SVDD model is trained using the Sequence Minimum Optimization (SVDD) algorithm to obtain the minimum hypersphere decision boundary that can include most normal samples. Based on the minimum hypersphere boundary obtained by training, healthy and abnormal states are divided. Samples inside the hypersphere boundary are defined as healthy states, and samples outside the boundary are judged as abnormal states. At the same time, the center of the hypersphere is used as the benchmark reference point for healthy states. The magnetization characteristic score is quantified by calculating the distance between the sample feature and the center of the hypersphere. During model optimization, a repetitive training strategy is employed, adjusting the kernel function parameters and penalty coefficients to optimize the hypersphere boundary. The multimodal test feature vectors to be evaluated are input into the trained magnetization characteristic evaluation model. The Mahalanobis distance between the vector and the center of the hypersphere is calculated, and this distance is compared with the preset hypersphere radius. When the distance is less than the radius, the magnetization characteristic is considered healthy; when it is greater than the radius, it is considered abnormal. Based on the reciprocal of the Mahalanobis distance, a magnetization characteristic score from 0 to 100 is calculated. Finally, complete magnetization characteristic evaluation information, including state determination, quantitative score, and feature analysis results, is generated.

[0032] Furthermore, in a preferred embodiment of the present invention, the step of acquiring high-voltage parallel reactors of the same specification with different fault types and fault degrees, connecting them to a pre-built reactor test network to obtain magnetization test data of faulty reactors, modeling the correlation between fault modes and magnetization characteristics, and generating a fault-magnetization knowledge base specifically includes: High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, connected to a pre-built reactor test network, and magnetization characteristics are tested according to the set DC step test scheme to obtain the magnetization test dataset of the faulty reactors. Based on the fault reactor magnetization test dataset, intrinsic magnetization feature vectors and environmental covariance feature vectors are extracted from the current-magnetic flux linkage curves of each fault reactor magnetization test sample. A fault feature vector set is generated through multimodal feature fusion processing, and each fault sample is labeled with fault type and fault degree. The fault feature vector set is mapped to a high-dimensional vector space. A feature space similarity matrix is ​​constructed by calculating the Euclidean distance between vectors. The DBSCAN clustering algorithm is introduced to identify the distribution pattern of sample points in the feature space. The sample points with density connection are divided into the same cluster by the preset neighborhood radius and minimum number of samples. At the same time, noise points in sparse areas are identified and removed to generate several sample clusters that represent different fault modes. Geometric feature analysis is performed on each cluster to calculate the centroid location, distribution range, and mean and variance statistical indicators of each dimension of the samples within the cluster. A mapping table between failure modes and feature clusters is established. The internal compactness and inter-cluster separation of each cluster are evaluated by cluster effectiveness indicators to optimize cluster merging and ultimately form feature clusters for different failure modes. Subsequently, diagnostic rules are generated based on the feature clustering results. Multi-dimensional statistical distribution analysis is performed on the sample features within each cluster. The probability distribution parameters of each feature dimension are calculated. The effective value range of each feature dimension is determined by the interval estimation method. Feature judgment rules are extracted by the decision tree induction algorithm to generate a multi-level decision rule set. Finally, the multi-level decision rule set and the feature clustering of different fault modes are integrated to generate a fault-magnetization knowledge base.

[0033] It should be noted that samples of high-voltage parallel reactors of the same specifications with different fault types and degrees were obtained, including typical fault modes such as inter-turn short circuits, core grounding, and insulation degradation. These faulty reactors were sequentially connected to a pre-built reactor test network, and their magnetization characteristics were systematically tested according to a standardized DC step test scheme. The voltage, current, and temperature response data of each fault sample under various levels of DC excitation were fully recorded, establishing a faulty reactor magnetization test dataset containing multiple fault modes. Based on this dataset, the current-magnetic flux linkage curves of each fault sample were feature-extracted to obtain intrinsic magnetization feature vectors that reflect the essence of magnetization characteristics. At the same time, environmental covariant feature vectors such as temperature and electromagnetic interference were extracted from the test environment monitoring data. The two types of feature vectors were integrated into a fault feature vector set, and each sample was accurately labeled with its corresponding fault type and fault degree level.

[0034] Subsequently, the fault feature vector set is mapped to a high-dimensional feature space. A sample similarity matrix is ​​constructed by calculating the Euclidean distance between vectors. The DBSCAN algorithm, based on density clustering principles, is used to analyze the sample distribution patterns in the feature space. Neighborhood radius and minimum sample number parameters are set to group density-connected sample points into the same cluster. Simultaneously, noise points in sparse regions are identified and removed, forming several fault mode sample clusters. Furthermore, in-depth geometric feature analysis is performed on each cluster, calculating the centroid location, distribution range, and statistical indicators of each feature dimension, including central tendency and dispersion, to establish a mapping relationship between fault modes and feature clusters. The internal compactness and inter-class separation of each cluster are evaluated using clustering effectiveness indicators. Similar clusters are merged and optimized, ultimately forming feature clusters for different fault modes. Building upon feature clustering, further diagnostic rule mining is performed. Multi-dimensional statistical distribution analysis is conducted on the sample features within each cluster, calculating the probability distribution parameters for each feature dimension. Interval estimation methods are used to determine the effective value range for each feature dimension. A decision tree induction algorithm is then used to extract discriminative feature judgment rules, generating a multi-level decision rule set. Finally, the feature clustering results are integrated with the multi-level decision rule set to construct a hierarchical fault-magnetization knowledge base containing a fault feature database and diagnostic rules, providing complete knowledge support and decision-making basis for subsequent reactor condition diagnosis.

[0035] Figure 2 A flowchart of the second method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method, provided as an embodiment of the present invention; like Figure 2 As shown, the present invention provides a second method flowchart for evaluating the magnetization characteristics of high-voltage shunt reactors based on the DC method, including: S202, a reactor state analysis model is constructed using the K-nearest neighbor classification algorithm and the random forest algorithm. The main classifier is trained using the random forest algorithm, and the node splitting is optimized by the Gini coefficient. At the same time, a K-nearest neighbor auxiliary classifier is constructed, and the feature-weighted Euclidean distance is used as the similarity measure. The two classifiers share the same training set, and finally a reactor state analysis model that meets the expectations is obtained. S204, obtain magnetization feature evaluation information and multi-modal test feature vector, input them into the reactor state analysis model, the main classifier performs multi-tree voting prediction through random forest, and outputs the preliminary identification result of fault type and confidence score; S206, the auxiliary classifier adopts a dynamic K-value strategy. It determines the optimal K-value by analyzing the sample distribution density in the feature space through the elbow rule, calculates the weighted Euclidean distance between the multimodal test feature vector and the centroids of various faults in the knowledge base, selects the K nearest neighbor samples for distance-weighted voting, and generates auxiliary recognition results. S208: Decision fusion is performed on the classifier output. When the confidence of the main classifier is higher than the preset threshold, the preliminary fault type identification result is directly adopted. When it is lower than the preset threshold, the Jaccard similarity coefficient between the output results of the main classifier and the auxiliary classifier is calculated based on the conflict resolution mechanism. The health score in the magnetization feature evaluation information is combined for weighted arbitration to generate the final fault type identification result. S210, based on the fault type identification result, calls the corresponding multi-level decision rule set in the fault-magnetization knowledge base for state verification. By calculating the matching degree between the multi-modal test feature vector and the rule premise conditions step by step, the rule weight is dynamically adjusted using the information gain ratio. The decision path is traversed by depth-first search, and the feature deviation degree and rule matching status of each node are recorded. Finally, the fault diagnosis information is obtained and a reactor state diagnosis report is generated and pushed.

[0036] It should be noted that, firstly, a random forest algorithm is used to train the main classifier. The Gini coefficient is used as the node splitting criterion to optimize the construction process of the decision tree, ensuring that each node split minimizes data impurity. Simultaneously, a K-nearest neighbor auxiliary classifier is constructed as a supplementary identification mechanism, using feature-weighted Euclidean distance as the similarity metric. Weights are assigned based on the importance of each feature dimension in historical diagnoses. The two classifiers share the same training dataset but employ different learning strategies, ultimately forming a complementary hybrid diagnostic model. The magnetization feature evaluation information and multimodal test feature vectors of the reactor under test are input into the trained state analysis model. The main classifier predicts the fault type using the multi-tree voting mechanism of the random forest. Each decision tree independently generates classification results based on its self-sampled training subset. Finally, the outputs of all decision trees are integrated, and a majority voting method is used to determine the preliminary fault type identification result. The confidence score of this result is calculated based on the voting distribution to assess the diagnostic reliability. The auxiliary classifier adopts a dynamic K-value strategy. It analyzes the sample distribution density in the feature space using the elbow rule to adaptively determine the optimal K-value parameter. It calculates the weighted Euclidean distance between the multimodal test feature vector and the centroids of various fault clusters in the knowledge base. Based on the distance, it selects the K nearest samples for distance-weighted class voting to generate auxiliary recognition results as a supplementary verification of the main classifier.

[0037] Subsequently, the outputs of the two classifiers are fused at the decision level. When the confidence score of the main classifier is higher than a preset reliability threshold, its fault type identification result is directly adopted. When the confidence score is lower than the threshold, a conflict resolution mechanism is activated. The Jaccard similarity coefficient between the outputs of the main classifier and the auxiliary classifier is calculated to assess their consistency. Simultaneously, a weighted arbitration is performed using the health score from the magnetization feature evaluation information to generate the final fault type identification result. Based on the confirmed fault type identification result, the corresponding multi-level decision rule set in the fault-magnetization knowledge base is invoked for state verification. Verification paths are constructed according to the rule confidence levels from high to low. Rule activation is achieved by calculating the matching degree between the multimodal test feature vector and the preconditions of each rule at each level. The information gain ratio is used to dynamically adjust the rule weights to optimize the decision-making effect. A depth-first search algorithm is used to systematically traverse all possible decision paths, recording the feature deviation and rule matching status of each node in real time during the traversal. Finally, all diagnostic information is integrated to generate a complete state diagnostic report containing fault type determination, severity assessment, feature parameter analysis, and maintenance recommendations, providing assistance for subsequent maintenance decisions.

[0038] Figure 3 A DC-based high-voltage shunt reactor magnetization characteristic evaluation system 3 is provided as an embodiment of the present invention. The system includes a memory 301, a processor 302, and a communication interface 303. The memory 301 contains a DC-based high-voltage shunt reactor magnetization characteristic evaluation method program. When the processor 302 executes the DC-based high-voltage shunt reactor magnetization characteristic evaluation method program, it performs the following steps: A DC step test scheme was set up and a reactor test network was built based on the specification parameters of the high voltage parallel reactor to be tested. The magnetization characteristics were tested and the entire test process was monitored using the reactor test network to obtain a multi-source test monitoring dataset. The multi-source test and monitoring dataset is preprocessed, and dynamic inductance and flux linkage sequence calculations are performed after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core. Intrinsic magnetization features are extracted using current-magnetic flux curves, and environmental covariance features are extracted from multi-source test and monitoring datasets to construct multi-modal test feature vectors. The magnetization features of the target high-voltage parallel reactor are evaluated using the multi-modal test feature vectors to generate magnetization feature evaluation information. High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, and magnetization test data of faulty reactors are obtained by connecting them to a pre-built reactor test network. The correlation between fault modes and magnetization characteristics is modeled to generate a fault-magnetization knowledge base. A reactor condition analysis model is established using the fault-magnetization knowledge base, and the condition diagnosis of the target high-voltage parallel reactor is performed by combining the magnetization feature evaluation information and multi-mode test feature vectors.

[0039] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0040] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0041] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0042] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0043] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method, characterized in that, include: A DC step test scheme was set up and a reactor test network was built based on the specification parameters of the high voltage parallel reactor to be tested. The magnetization characteristics were tested and the entire test process was monitored using the reactor test network to obtain a multi-source test monitoring dataset. The multi-source test and monitoring dataset is preprocessed, and dynamic inductance and flux linkage sequence calculations are performed after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core. Intrinsic magnetization features are extracted using current-magnetic flux curves, and environmental covariance features are extracted from multi-source test and monitoring datasets to construct multi-modal test feature vectors. The magnetization features of the target high-voltage parallel reactor are evaluated using the multi-modal test feature vectors to generate magnetization feature evaluation information. High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, and magnetization test data of faulty reactors are obtained by connecting them to a pre-built reactor test network. The correlation between fault modes and magnetization characteristics is modeled to generate a fault-magnetization knowledge base. A reactor condition analysis model is established using the fault-magnetization knowledge base, and the condition diagnosis of the target high-voltage parallel reactor is performed by combining the magnetization feature evaluation information and multi-mode test feature vectors.

2. The method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method according to claim 1, characterized in that, The process involves setting up a DC step test scheme and building a reactor test network based on the specifications of the high-voltage parallel reactor under test. The reactor test network is then used to perform magnetization characteristic testing and monitor the entire testing process, resulting in a multi-source test monitoring dataset. Specifically, this dataset includes: Obtain the specification sheet of the high-voltage parallel reactor to be tested, calculate the upper limit of the DC excitation voltage required to make the iron core reach the expected magnetization level based on the law of electromagnetic induction, determine the maximum allowable operating current of the test system through safety margin calculation based on the rated current value and insulation withstand voltage level, determine the optimal voltage-current combination parameters according to the power supply output characteristic curve, and generate a DC step test scheme. Subsequently, a reactor test network consisting of a programmable DC power supply, a power adjustable resistor array, a solid-state power switch, and a data acquisition system was built. The power adjustable resistor array adopts a multi-stage parallel structure to achieve fine adjustment of the resistance value. The solid-state power switch uses an insulated gate bipolar transistor and is equipped with an overcurrent protection circuit. The data acquisition system consists of a multi-source heterogeneous sensor array. After the reactor test network is built, a self-calibration process is performed on the reactor test network, including zero-point calibration and proportional coefficient calibration of the voltage measurement channel, frequency response characteristic testing of the current sensor, and thermocouple cold junction compensation calibration of the temperature monitoring unit to ensure the accuracy and synchronization of each measurement channel. In the test network of the target high-voltage parallel reactor and the calibrated reactor, the control unit outputs a DC excitation signal based on the DC step test scheme. After each current step reaches the target value, it enters the stable monitoring stage to monitor the fluctuation of the winding terminal voltage in real time. The sliding window algorithm is used to calculate the voltage change rate. When the voltage change rate of multiple consecutive sampling periods is lower than the preset stability judgment threshold, the steady-state operating point is confirmed. At this time, the steady-state voltage reading, current reading, and temperature data of each measuring point are recorded. During the transient process of current step change, the dynamic response waveforms of voltage and current are captured completely at the highest sampling rate. At the same time, the change trajectory of ambient temperature is recorded. Finally, the collected data are integrated to generate a multi-source test and monitoring dataset.

3. The method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method according to claim 1, characterized in that, The preprocessing of the multi-source test and monitoring dataset, followed by dynamic inductance and flux linkage sequence calculations to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization properties of the iron core, specifically includes: Data quality improvement processing is performed on the original sampled signals in the multi-source test and monitoring dataset. Discrete wavelet transform is used to decompose the voltage and current signals into multiple scales. The wavelet coefficients are subjected to soft thresholding by setting an adaptive threshold to suppress high-frequency noise. At the same time, an outlier detection algorithm is used to identify abnormal temperature monitoring data and to compensate for missing values. Based on the preprocessed multi-source test and monitoring dataset, pure voltage and pure current sequences are extracted. The instantaneous rate of change of the current signal is calculated using a five-point center difference algorithm. Two adjacent sampling points are taken before and after each sampling point to construct a five-point difference template. The weighted average is calculated using preset difference coefficients to obtain the rate of change of current. The current rate of change is then divided point by point with the voltage measurement value at the corresponding time to obtain the dynamic inductance sequence. Simultaneously calculate the flux linkage parameters, use the composite Simpson numerical integration algorithm to numerically integrate the pure voltage sequence, divide the entire integration interval into several equal intervals, approximate the integrand function in each sub-interval using a quadratic interpolation polynomial, and calculate the cumulative flux linkage value at each sampling time by weighted summation to obtain the flux linkage change sequence. The calculated dynamic inductance sequence and flux linkage change sequence are spatiotemporally aligned and fused. A time-scale alignment algorithm based on cross-correlation function is used to correct the sampling time deviation between the two sequences. The dynamic inductance sequence is resampled to the corresponding time point of the flux linkage change sequence through Hermit interpolation algorithm to establish the mapping relationship between dynamic inductance value and flux linkage value, forming a dynamic inductance-flux linkage feature sequence arranged in order of flux linkage value.

4. The method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method according to claim 1, characterized in that, The preprocessing of the multi-source test and monitoring dataset, followed by dynamic inductance and flux linkage sequence calculations to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization properties of the iron core, further includes: A dynamic inductance-magnetic flux linkage feature sequence is obtained. A sliding window curvature detection algorithm is used to identify the characteristic inflection points of the magnetization curve. Based on a preset window size, the curvature value of the neighboring region is calculated at each data point in the dynamic inductance-magnetic flux linkage feature sequence. When the curvature value of 5 consecutive data points exceeds a preset curvature threshold, it is determined as the boundary point between the linear region and the saturation region, and the sequence segmentation result is generated. Based on the sequence segmentation results, the current optimal linear parameters, including slope and intercept, are solved by constructing a system of normal equations and using the Cholesky decomposition algorithm. In the saturation region, the hyperbolic tangent function is selected for nonlinear fitting, and the Levenberg-Marquardt algorithm is used to solve the nonlinear fitting coefficients. An initial magnetization characteristic model is established to obtain an initial reconstructed parameter combination containing linear and nonlinear parameters. A genetic algorithm is introduced to globally optimize the initial reconstructed parameter combination, and an objective function including a fitting error term and a curve smoothness term is constructed. The fitting error term is characterized by the root mean square error between the measured data and the fitted curve, and the curve smoothness term is defined by the sum of squares of the differences of the first derivatives at the connection points. During the optimization process, the constraints are that the piecewise functions have the same function value at the connection point and the difference in the first derivative value is less than a preset range. Genetic operations are performed by roulette wheel selection algorithm, arithmetic crossover operator and Gaussian mutation operator. The best individual is retained in each generation and an elite retention strategy is adopted. After multiple generations of evolutionary calculation, the optimal combination of reconstruction parameters is obtained. Based on the optimal parameter combination, the complete current-magnetic flux linkage curve is reconstructed. The fitting results of the linear region and the saturation region are combined in the form of a piecewise function, and the residual of each sampling point is calculated. The distribution characteristics of the residual are analyzed to verify the rationality of the reconstruction, and finally the current-magnetic flux linkage curve is obtained.

5. The method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method according to claim 1, characterized in that, The process involves extracting intrinsic magnetization features using current-flux flux curves and constructing a multi-modal test feature vector by extracting environmental covariance features from a multi-source test and monitoring dataset. This multi-modal test feature vector is then used to evaluate the magnetization characteristics of the target high-voltage parallel reactor, generating magnetization feature evaluation information. Specifically, this includes: The current-magnetic flux characteristic curve is obtained, and the magnetization characteristic parameters are extracted using an automatic feature point detection algorithm. The differential permeability of the curve at each operating point is calculated using the five-point central difference method. The saturation flux value and the corresponding saturation current value are identified using the golden section search algorithm. The coercivity characteristic points are solved using the cubic spline interpolation method. At the same time, the permeability slope in the linear region and the permeability attenuation coefficient in the saturation region are calculated, and the hysteresis loop area and remanence ratio parameters are extracted to form the intrinsic magnetization characteristic vector. Simultaneously, environmental covariance features are extracted from the multi-source test and monitoring dataset. The average temperature, temperature change gradient, and steady-state temperature value throughout the test are extracted as thermal environment features. Spectral analysis is performed on the background electromagnetic noise data to extract the amplitude of the main interference frequency components as electromagnetic environment features. Together, they constitute the environmental covariance feature vector. The intrinsic magnetization feature vector and the environmental covariance feature vector are fused in a multimodal manner. First, the Z-score normalization method is used to normalize the feature vector and map it to a high-dimensional space. Then, the principal component score of each feature is calculated by principal component analysis. The principal component features with a score greater than the preset threshold are retained to generate a multimodal test feature vector. A magnetization characteristic evaluation model was constructed using SVDD. Magnetization characteristic data of high-voltage parallel reactors in healthy and abnormal states were obtained by retrieving historical data. The magnetization characteristic evaluation model was then trained to obtain the minimum hypersphere boundary. Based on the boundary of the minimum hypersphere, the healthy state and the abnormal state are divided. The state inside the boundary of the minimum hypersphere is the healthy state, and the state outside the boundary of the minimum hypersphere is the abnormal state. At the same time, the center of the hypersphere is used as the reference point for the healthy state, and the distance to the center of the hypersphere is used as the score. The magnetization characteristic evaluation model is repeatedly trained. The multimodal test feature vector is input into the magnetization characteristic evaluation model after training to calculate the Mahalanobis distance with the center of the hypersphere, and compared with the preset hypersphere radius. If it is less than the preset hypersphere radius, it represents a healthy state of magnetization characteristics. If it is greater than the preset hypersphere radius, it represents an abnormal state of magnetization characteristics. The magnetization characteristic score is output based on the inverse ratio of the corresponding Mahalanobis distance, and finally the magnetization characteristic evaluation information is obtained.

6. The method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method according to claim 1, characterized in that, The process involves acquiring high-voltage parallel reactors of the same specifications with different fault types and severity, connecting them to a pre-built reactor testing network to obtain magnetization test data of faulty reactors, modeling the correlation between fault modes and magnetization characteristics, and generating a fault-magnetization knowledge base. Specifically, this includes: High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, connected to a pre-built reactor test network, and magnetization characteristics are tested according to the set DC step test scheme to obtain the magnetization test dataset of the faulty reactors. Based on the fault reactor magnetization test dataset, intrinsic magnetization feature vectors and environmental covariance feature vectors are extracted from the current-magnetic flux linkage curves of each fault reactor magnetization test sample. A fault feature vector set is generated through multimodal feature fusion processing, and each fault sample is labeled with fault type and fault degree. The fault feature vector set is mapped to a high-dimensional vector space. A feature space similarity matrix is ​​constructed by calculating the Euclidean distance between vectors. The DBSCAN clustering algorithm is introduced to identify the distribution pattern of sample points in the feature space. The sample points with density connection are divided into the same cluster by the preset neighborhood radius and minimum number of samples. At the same time, noise points in sparse areas are identified and removed to generate several sample clusters that represent different fault modes. Geometric feature analysis is performed on each cluster to calculate the centroid location, distribution range, and mean and variance statistical indicators of each dimension of the samples within the cluster. A mapping table between failure modes and feature clusters is established. The internal compactness and inter-cluster separation of each cluster are evaluated by cluster effectiveness indicators to optimize cluster merging and ultimately form feature clusters for different failure modes. Subsequently, diagnostic rules are generated based on the feature clustering results. Multi-dimensional statistical distribution analysis is performed on the sample features within each cluster. The probability distribution parameters of each feature dimension are calculated. The effective value range of each feature dimension is determined by the interval estimation method. Feature judgment rules are extracted by the decision tree induction algorithm to generate a multi-level decision rule set. Finally, the multi-level decision rule set and the feature clustering of different fault modes are integrated to generate a fault-magnetization knowledge base.

7. The method for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method according to claim 1, characterized in that, The process of establishing a reactor condition analysis model using the fault-magnetization knowledge base, and combining the magnetization feature evaluation information and multi-mode test feature vectors to perform condition diagnosis on the target high-voltage parallel reactor under test, specifically includes: A reactor state analysis model was constructed using the K-nearest neighbor classification algorithm and the random forest algorithm. The main classifier was trained using the random forest algorithm, and the node splitting was optimized by the Gini coefficient. At the same time, a K-nearest neighbor auxiliary classifier was constructed, and the feature-weighted Euclidean distance was used as the similarity measure. The two classifiers shared the same training set, and finally the reactor state analysis model that met the expectations was obtained. Magnetization feature evaluation information and multimodal test feature vectors are obtained and input into the reactor state analysis model. The main classifier performs multi-tree voting prediction through random forest and outputs preliminary fault type identification results and confidence scores. The auxiliary classifier adopts a dynamic K-value strategy. It determines the optimal K-value by analyzing the sample distribution density in the feature space through the elbow rule, calculates the weighted Euclidean distance between the multimodal test feature vector and the centroids of various faults in the knowledge base, selects the K nearest neighbor samples for distance-weighted voting, and generates auxiliary recognition results. The classifier outputs are fused for decision-making. When the confidence of the main classifier is higher than the preset threshold, the preliminary fault type identification result is directly adopted. When it is lower than the preset threshold, the Jaccard similarity coefficient between the outputs of the main classifier and the auxiliary classifier is calculated based on the conflict resolution mechanism. The health score in the magnetization feature evaluation information is combined for weighted arbitration to generate the final fault type identification result. Based on the fault type identification results, the corresponding multi-level decision rule set in the fault-magnetization knowledge base is called for state verification. The matching degree between the multi-modal test feature vector and the rule preconditions is calculated step by step. The rule weight is dynamically adjusted using the information gain ratio. The decision path is traversed by depth-first search, and the feature deviation degree and rule matching status of each node are recorded. Finally, the fault diagnosis information is obtained and a reactor state diagnosis report is generated and pushed.

8. A system for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory contains a program for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method. When the processor executes the program for evaluating the magnetization characteristics of a high-voltage shunt reactor based on the DC method, it performs the following steps: A DC step test scheme was set up and a reactor test network was built based on the specification parameters of the high voltage parallel reactor to be tested. The magnetization characteristics were tested and the entire test process was monitored using the reactor test network to obtain a multi-source test monitoring dataset. The multi-source test and monitoring dataset is preprocessed, and dynamic inductance and flux linkage sequence calculations are performed after preprocessing to reconstruct the current-flux linkage curve characterizing the nonlinear magnetization characteristics of the iron core. Intrinsic magnetization features are extracted using current-magnetic flux curves, and environmental covariance features are extracted from multi-source test and monitoring datasets to construct multi-modal test feature vectors. The magnetization features of the target high-voltage parallel reactor are evaluated using the multi-modal test feature vectors to generate magnetization feature evaluation information. High-voltage parallel reactors of the same specification with different fault types and fault degrees are obtained, and magnetization test data of faulty reactors are obtained by connecting them to a pre-built reactor test network. The correlation between fault modes and magnetization characteristics is modeled to generate a fault-magnetization knowledge base. A reactor condition analysis model is established using the fault-magnetization knowledge base, and the condition diagnosis of the target high-voltage parallel reactor is performed by combining the magnetization feature evaluation information and multi-mode test feature vectors.

Images