A passive vector parameter real-time monitoring method in AC-DC conversion process

Abstract

The application provides a passive vector parameter real-time monitoring method in AC-DC conversion process, and belongs to the field of power system detection, comprising: real-time monitoring of the operation data of the AC-DC conversion system in the low-voltage metering cabinet, constructing a hexagon diagram according to the operation data, extracting the characteristic parameters of the hexagon diagram, the characteristic parameters of the hexagon diagram including voltage, current phase relationship and harmonic component; if it is determined that the passive device parameter change has an influence on the AC-DC conversion system, a passive device degradation early warning signal is generated, the reactive power compensation capacity, the reactive power compensation device parameters and the filter parameters are adjusted, and the adjustment amount is determined; real-time monitoring of the change of the characteristic parameters of the hexagon diagram, trend fitting of the historical characteristic parameters of the hexagon diagram is carried out by using a time series analysis method, the passive device degradation acceleration point is judged, and the operation parameters of the reactive power compensation device are adjusted according to the passive device degradation acceleration point; the passive device degradation can be effectively warned, the reactive power compensation strategy can be optimized, and the operation reliability and the measurement accuracy of the low-voltage metering cabinet can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to a passive vector parameter real-time monitoring method in AC-DC conversion process, belonging to the field of power system detection. BACKGROUND

[0002] In the low-voltage metering cabinet system, by sampling the voltage and current signals of the AC-DC conversion system, and converting the three-phase static coordinate system AC voltage and current into two-phase rotating coordinate system DC, then the obtained voltage and current vectors are plotted on the complex plane, with the passage of time, these vector points will form a nearly hexagonal trajectory, that is, a hexagon diagram, the shape, size and position of the hexagon diagram reflect the running state of the AC-DC conversion system and the parameter change of the passive device, the characteristic parameters of the hexagon diagram of the AC-DC conversion system are important indicators reflecting the state of the passive device.

[0003] However, due to the complex electrical coupling relationship between multiple low-voltage metering cabinets, the passive device parameter change of a single metering cabinet will affect the running state of the entire cluster, which also leads to the following technical problems in the extraction and analysis process of the hexagon diagram characteristic parameters:

[0004] Firstly, the passive device parameters of different metering cabinets present nonlinear changes under different load conditions, making it difficult to accurately capture the dynamic characteristics of the hexagon diagram characteristic parameters.

[0005] Secondly, the electrical coupling effect between multiple metering cabinets makes the hexagon diagram characteristic parameter change of a single metering cabinet transmitted to adjacent metering cabinets, forming a chain reaction, further aggravating the complexity of parameter identification.

[0006] In addition, the degradation process of the passive device has gradualness and concealment, and its parameter change is often difficult to be found by conventional monitoring means in the early stage, leading to the gradual deterioration of the performance of the AC-DC conversion system, and in the process of cluster cooperative operation, due to the difference in the state of the passive device of each metering cabinet, the coordinated control of the reactive power compensation device is challenged.

[0007] For example, when the passive device parameter of a certain metering cabinet is abnormal, the adjustment capacity of its reactive power compensation device may be limited, thereby affecting the power factor optimization and measurement accuracy of the entire cluster. At the same time, the degradation of the passive device will also increase the potential failure risk of the metering cabinet, but due to the concealment of its degradation process, the establishment of early warning mechanism faces difficulties.

[0008] The existence of these technical problems makes the research on the passive device state evaluation of the low-voltage metering cabinet based on the hexagon diagram characteristic parameters and the cluster cooperative optimization operation strategy face many challenges. SUMMARY

[0009] According to the problems described in the background, the problem to be solved by the present application is to provide a passive vector parameter real-time monitoring method in an AC-DC conversion process to solve the problems mentioned above.

[0010] To achieve the above-mentioned purpose, the present application provides the following technical solutions: a passive vector parameter real-time monitoring method in an AC-DC conversion process, comprising the following steps:

[0011] (1) Collecting the operation data of the AC-DC conversion system in the low-voltage metering cabinet and generating a hexagon diagram to extract its characteristic parameters, analyzing the parameters to evaluate the state of the passive device;

[0012] (2) Combining and processing the historical operation parameters of the passive device with the characteristic parameters of the hexagon diagram, establishing a mapping relationship with the AC-DC conversion system, and obtaining the failure mode of the passive device, and judging whether the parameter change of the passive device affects the AC-DC conversion system according to the preset performance threshold;

[0013] (3) If it is determined that the parameter change of the passive device affects the AC-DC conversion system, a passive device degradation warning signal is generated, and the reactive power compensation capacity, the reactive power compensation device parameters and the filter parameters are adjusted to determine the adjustment amount;

[0014] (4) Using a machine learning algorithm, the adjustment amount and the historical parameter data of the passive device are used as input to predict the trend of the parameter change of the passive device, and the compensation mode, compensation capacity configuration and switching mode of the reactive power compensation device are determined according to the prediction result, and the power factor of the whole cluster is dynamically compensated and coordinated;

[0015] (5) According to the compensation mode, compensation capacity configuration and switching mode of the reactive power compensation device, as well as the power factor of the whole cluster and the passive device degradation warning signal, the particle swarm optimization algorithm is used to coordinate the reactive power compensation devices of several low-voltage metering cabinets, and the output capacity and operating state of the reactive power compensation device are adjusted;

[0016] (6) Real-time monitoring of the change of the hexagon diagram characteristic parameters, using time series analysis method to trend fit the historical hexagon diagram characteristic parameters, judging the acceleration point of the passive device degradation, and adjusting the operating parameters of the reactive power compensation device according to the acceleration point of the passive device degradation.

[0017] Preferably, the step (1) comprises the following steps:

[0018] (1.1) Sampling the AC-DC conversion system of the low-voltage metering cabinet according to a preset sampling frequency to obtain AC voltage and current waveform data;

[0019] (1.2) Processing the AC voltage and current waveform data by Fourier transform to extract the amplitude and phase characteristics of the fundamental signal and harmonic signal from the waveform data;

[0020] (1.3) constructing a hexagon diagram coordinate matrix according to the characteristics of the fundamental wave signal and the harmonic signal, the hexagon diagram coordinate matrix containing a voltage amplitude, a current amplitude, a phase difference value, a fundamental wave frequency, a total harmonic distortion rate, and a power grid frequency deviation value;

[0021] (1.4) performing principal component analysis on the hexagon diagram coordinate matrix, clustering the characteristic components obtained by the principal component analysis by using a support vector machine, and obtaining an evaluation index of a monitoring parameter.

[0022] Preferably, the step (2) comprises the following steps:

[0023] (2.1) obtaining a historical operation record of a passive device in a storage, the historical operation record containing a device operation time, a temperature parameter, a voltage stress parameter, and a current stress parameter;

[0024] (2.2) constructing a device stress parameter matrix according to the historical operation record, the stress parameter matrix obtaining a failure mode classification result by Gaussian mixture clustering;

[0025] (2.3) establishing a mapping relationship between the stress parameter matrix and an operation parameter of an AC-DC conversion system by using a recurrent neural network, the operation parameter containing a power factor value and a measurement accuracy value;

[0026] (2.4) judging an influence degree of the passive device on the performance of the AC-DC conversion system, specifically comprising: constructing a device performance degradation index according to the failure mode classification result;

[0027] (2.5) extracting a temperature threshold parameter, a voltage threshold parameter, and a current threshold parameter according to the performance degradation index;

[0028] (2.6) calculating a power factor deviation value and a measurement accuracy deviation value by using the threshold parameters.

[0029] Preferably, the step (3) comprises the following steps:

[0030] (3.1) receiving a pre-warning data matrix in which at least one of a device temperature change amount, a voltage change amount, and a current change amount exceeds a preset threshold interval, and classifying the pre-warning data matrix by using a multilayer perceptron to obtain a pre-warning level value;

[0031] (3.2) extracting a reactive power compensation capacity reference value, a compensation device reference parameter, and a compensation operating point reference value from a compensation parameter database according to the pre-warning level value, and calculating a compensation capacity adjustment value by using a reactive power compensation algorithm;

[0032] (3.3) extracting a frequency range reference value, a bandwidth reference value and a gain reference value from a filter parameter database according to the early warning level value and the compensation capacity adjustment value, and obtaining a filter parameter adjustment value through least square method operation;

[0033] (3.4) constructing a correction parameter matrix according to the compensation capacity adjustment value and the filter parameter adjustment value, and obtaining a final adjustment value of reactive power compensation capacity, a final adjustment value of compensation device parameters and a final adjustment value of filter parameters through optimization operation of the correction parameter matrix by using a random forest regression method.

[0034] Preferably, the step (4) comprises the following steps:

[0035] (4.1) constructing a training data matrix according to the historical parameter data of the passive device, and obtaining a passive device parameter change trend prediction curve through processing of the training data matrix by using a long short-term memory network;

[0036] (4.2) extracting a parameter change amount time distribution sequence according to the prediction curve, constructing a parameter change matrix, and calculating a compensation mode selection serial number according to the parameter change matrix, wherein the parameter change matrix comprises a temperature change amount, a voltage change amount and a current change amount;

[0037] (4.3) extracting a compensation capacity initial configuration value corresponding to the compensation mode selection serial number from a compensation device parameter database, optimizing the compensation capacity initial configuration value through a Kalman filter, calculating an optimal estimation value according to a measurement noise variance and a process noise variance, and obtaining a compensation capacity optimal configuration value;

[0038] (4.4) generating a compensation device switching time sequence table according to the compensation capacity optimal configuration value, wherein the compensation device switching time sequence table comprises a switching sequence number, a switching time interval and a switching holding time, and adjusting the compensation device switching time sequence table in real time through an adaptive feedback compensation algorithm, and calculating a compensation deviation according to a real-time power factor measurement value;

[0039] (4.5) if the overall power factor is lower than a preset threshold value, determining a reactive power compensation amount adjustment range, calculating a reactive power compensation device operating parameter adjustment value according to the adjustment range, updating the operating strategy of the reactive power compensation device through the parameter adjustment value, iteratively updating the overall power factor of the cluster, judging whether the updated overall power factor reaches a preset optimization target value, and if the optimization target value is not reached, re-updating the operating strategy until the overall power factor reaches the optimization target value.

[0040] Preferably, the step (5) comprises the following steps:

[0041] (5.1) Obtain the operation data of the low-voltage metering cabinet compensation device to construct an operation state matrix, wherein the operation state matrix comprises attribute values of compensation modes and capacity configuration values and switching mode serial numbers;

[0042] (5.2) Construct a fitness function according to the operation state matrix, wherein the fitness function comprises compensation effect constraint terms and device operation constraint terms;

[0043] (5.3) Perform parameter search on the fitness function by a particle swarm optimization algorithm, wherein a particle position represents a compensation capacity distribution ratio, and a particle speed represents a capacity adjustment step;

[0044] (5.4) Calculate a device load rate according to the compensation capacity distribution ratio, wherein the load rate is a ratio of a real-time capacity to a rated capacity, and the compensation device is controlled in groups according to the load rate value;

[0045] (5.5) Predict the load rate by using a deep neural network, and if the predicted load rate exceeds a preset threshold, correct a compensation device control instruction sequence according to the predicted load rate to obtain a corrected control instruction sequence.

[0046] Preferably, the step (6) comprises the following steps:

[0047] (6.1) Collect hexagon diagram characteristic parameters according to real-time monitoring devices, wherein the characteristic parameters comprise voltage phase parameters, current phase parameters and harmonic content parameters, and obtain a characteristic parameter time series data set by extracting sampling points from the characteristic parameters at fixed time intervals;

[0048] (6.2) Obtain a parameter change trend curve by fitting the characteristic parameter time series data set by an exponential smoothing algorithm, calculate a slope of the trend curve to obtain a parameter change rate curve, calculate an acceleration value according to the change rate curve and mark abnormal points;

[0049] (6.3) Receive the parameter change rate and acceleration data by using a long short-term memory network, verify the abnormal points by a parameter change law extraction unit in a network hidden layer to obtain degradation acceleration points;

[0050] (6.4) Calculate a compensation parameter adjustment amount according to a parameter change amplitude of the degradation acceleration point position, correct compensation device operation parameters for the compensation parameter adjustment amount and generate a compensation device operation parameter update instruction;

[0051] (6.5) Obtain the time series data of the historical hexagon diagram feature parameters, calculate the feature parameter change rate at each time point by using the sliding window method, perform trend fitting on the feature parameter change rate, construct a linear regression model of the time series, obtain the trend line slope value, if the trend line slope value exceeds the preset threshold value, mark the time point as a potential degradation point, calculate the fitting degree index, perform second-order difference calculation on the potential degradation point, and determine whether the second-order derivative of the change rate appears a turning point, if yes, determine that the corresponding point is an acceleration point of the passive device degradation.

[0052] Preferably, the step (4.5) comprises the following steps:

[0053] (4.5.1) According to the power factor measurement device, the real-time value of the power factor is collected, and it is judged whether the real-time value of the power factor is lower than the preset power factor threshold value;

[0054] (4.5.2) If the real-time value of the power factor is lower than the preset power factor threshold value, the compensation capacity parameter and the operating state parameter are obtained from the compensation device parameter library;

[0055] (4.5.3) A compensation parameter optimization objective function is constructed by the compensation capacity parameter and the operating state parameter, and a gradient descent method is used to solve the compensation parameter optimization objective function to obtain a reactive power compensation amount adjustment interval;

[0056] (4.5.4) According to the reactive power compensation amount adjustment interval, an adaptive learning rate algorithm is used to optimize the compensation capacity value and the switch control value to obtain a compensation parameter adjustment matrix;

[0057] (4.5.5) The compensation capacity correction value and the switch timing correction value are extracted from the compensation parameter adjustment matrix, and the compensation device control instruction sequence is updated according to the compensation capacity correction value and the switch timing correction value;

[0058] (4.5.6) According to the compensation device control instruction sequence, the reactive power compensation operation is collected, and it is judged whether the power factor measurement value reaches the preset optimization target value.

[0059] Preferably, the step (6.5) comprises the following steps:

[0060] (6.5.1) According to the low-voltage metering cabinet measurement device, the historical hexagon diagram feature parameters are collected, and the feature parameters are sampled by using a fixed length sliding window to obtain a feature parameter change rate data set;

[0061] (6.5.2) The feature parameter change rate data set is fitted by using a linear regression algorithm, the trend line slope value of the fitting curve is calculated, and if the slope value is greater than the preset change rate threshold value, it is determined that the sampling point is a potential abnormal point;

[0062] (6.5.3) Calculate the goodness-of-fit index for the potential abnormal points, if the coefficient of determination, fitting standard deviation, and correlation coefficient simultaneously satisfy the preset goodness condition, record the sampling point to the abnormal point dataset;

[0063] (6.5.4) Perform first-order difference operation on the abnormal point dataset by central difference method to obtain a first-order derivative sequence, perform second-order difference operation on the first-order derivative sequence to obtain a second-order derivative sequence, if the second-order derivative changes from negative to positive, determine that the sampling point is a degradation acceleration point.

[0064] The beneficial effects of the present application are:

[0065] 1. By monitoring the operation data of the AC-DC conversion system in real time and generating a hexagon diagram, the parameter changes of the passive device can be effectively captured, laying a foundation for state evaluation and optimization control.

[0066] 2. The fusion of the hexagon diagram feature parameters and historical data can clearly reveal the causal relationship between the passive device parameter changes and the system performance, and the stress parameters can intuitively reflect the degradation trend after being recorded in matrix form, and the prediction results of the recurrent neural network further quantify the specific influence of these changes on the power factor and the measurement accuracy.

[0067] 3. By adjusting the reactive power compensation capacity, reactive power compensation device parameters and filter parameters, the influence of device degradation on the power factor and the measurement accuracy can be effectively addressed, while conflicts or system oscillations in parameter adjustment are avoided.

[0068] 4. By determining the compensation mode, compensation capacity configuration and switching mode of the reactive power compensation device, and the power factor, the whole process automation from parameter prediction to compensation control is realized, not only reducing the compensation lag, but also avoiding equipment impact through multi-level progressive control, significantly improving the stability and efficiency of the cluster operation.

[0069] 5. By using the particle swarm optimization algorithm to coordinate the reactive power compensation devices of several low-voltage metering cabinets, adjusting the output capacity and operating state of the reactive power compensation devices, the degradation warning is responded, and the risk of high-load operation of the equipment is reduced through algorithm coordination optimization, and the application of sliding time window and characteristic curve further enhances the adaptability of the system to load changes.

[0070] 6. By using the time series analysis method to perform trend fitting on the historical hexagon diagram feature parameters, the acceleration point of passive device degradation is determined, and the operating parameters of the reactive power compensation device are adjusted according to the passive device degradation acceleration point, realizing accurate monitoring and dynamic optimization of passive device degradation, and through the combination of real-time data and historical trends, not only the diagnosis efficiency is improved, but also the equipment life is prolonged through timely adjustment of operating parameters.

[0071] 7. The application can effectively warn passive device degradation, optimize reactive power compensation strategy, and improve the operation reliability and measurement accuracy of low-voltage metering cabinets. BRIEF DESCRIPTION OF DRAWINGS

[0072] Figure 1 The step flowchart of the application is shown in the figure;

[0073] Figure 2 The flowchart of step (1) is shown in the figure;

[0074] Figure 3 The flowchart of step (2) is shown in the figure;

[0075] Figure 4 The flowchart of step (3) is shown in the figure;

[0076] Figure 5 The flowchart of step (4) is shown in the figure;

[0077] Figure 6 The flowchart of step (5) is shown in the figure;

[0078] Figure 7 The flowchart of step (6) is shown in the figure; DETAILED DESCRIPTION

[0079] The embodiments of the application are further described below in conjunction with the accompanying drawings:

[0080] Embodiment 1

[0081] As shown in the figure, the application provides a real-time monitoring method for passive vector parameters in AC-DC conversion process, including the following steps: Figure 1 (1) Real-time acquisition of operation data of the AC-DC conversion system in the low-voltage metering cabinet, and generation of a hexagon diagram based on the acquired operation data to extract characteristic parameters and analyze the parameters to evaluate the passive device state;

[0082] The characteristic parameters include voltage amplitude, current phase relationship, harmonic components, and other indicators.

[0083] Step (1) includes the following steps:

[0084] (1.1) Sampling the low-voltage metering cabinet AC-DC conversion system according to the preset sampling frequency to obtain AC voltage and current waveform data;

[0085] Sampling is to collect the voltage amplitude and current amplitude of each port of the low-voltage metering cabinet AC-DC conversion system, and to record the AC waveform at fixed sampling intervals under the control of the waveform sampling frequency to obtain the first set of AC voltage and current waveform data.

[0086]

[0087] ​(1.2) Fourier transform processing of the alternating voltage and current waveform data to extract the amplitude and phase characteristics of the fundamental and harmonic signals from the waveform data;

[0088] The Fourier transform processing processes the first set of alternating voltage and current waveform data to calculate the amplitudes of the fundamental and harmonic signals, and extracts the phase difference value between the voltage signal and the current signal from the fundamental signal, i.e. the phase characteristic.

[0089] (1.3) Constructing a hexagon coordinate matrix based on the characteristics of the fundamental and harmonic signals, which contains voltage amplitude, current amplitude, phase difference value, fundamental frequency, total harmonic distortion rate, and grid frequency deviation value;

[0090] A first hexagon coordinate matrix is constructed according to the phase difference value, and the voltage amplitude, current amplitude, phase difference value, fundamental frequency, total harmonic distortion rate, and grid frequency deviation value are labeled in six dimensions respectively. The amplitude and phase information is extracted from each harmonic signal, and the voltage total harmonic distortion rate and current total harmonic distortion rate are calculated to construct a second hexagon coordinate matrix. The numerical characteristics in each dimension are extracted from the two hexagon coordinate matrices, including amplitude characteristics, phase characteristics, and frequency characteristics, to obtain a feature vector group.

[0091] (1.4) Principal component analysis of the hexagon coordinate matrix, and clustering of the feature components obtained by the principal component analysis using a support vector machine to obtain an evaluation index of the monitoring parameters.

[0092] Before principal component analysis, the feature vector group is reduced in dimension, and then the main feature components are extracted by the principal component analysis method to construct a feature classification space. Based on the feature classification space, a support vector machine is used to cluster the feature components to obtain a feature clustering result, forming an evaluation index of the monitoring parameters.

[0093] In step (1.1), the sampling frequency is usually set to 128 times the fundamental frequency, for example, under a 50 Hz power grid, the sampling frequency can reach 6400 Hz to ensure the accuracy of the waveform data; 144 sampling points are collected in each sampling period, with 72 sampling points for voltage and current respectively, and the data is arranged in time sequence to form a waveform sequence;

[0094] In step (1.2), the waveform data is decomposed into fundamental and 2 to 13 harmonic components by Fourier transform, and the high-order harmonics are ignored due to their small content. The phase difference extracted from the fundamental component is calculated by the time difference of the zero-crossing points of the voltage and current waveforms, which reflects the load characteristics, such as the phase difference of resistive load being close to 0 degrees, the inductive load being positive, and the capacitive load being negative.

[0095] In step (1.3), in the hexagon structure construction stage, two hexagon coordinate matrices are constructed according to the extracted feature data, wherein the first matrix reflects the basic running state of the system and contains six dimensions such as voltage effective value, current effective value, phase difference, fundamental frequency, total harmonic distortion rate and power grid frequency deviation value, for example, when the voltage is 220 volts, the current is 5 amperes and the phase difference is 30 degrees, the normalized value is marked on the corresponding coordinate axis, and the second matrix focuses on the harmonic features, and the coordinate axes respectively represent the 2 to 7 harmonic contents, for example, the 2 harmonic content is 2.5% and the 3 harmonic content is 4.2%, and the hexagon is drawn in the polar coordinate mode with the origin as the center.

[0096] In step (1.4), the feature vector groups of the two matrices are processed by dimension reduction, the first three principal components are extracted by principal component analysis, the cumulative contribution rate is more than 85%, the feature classification space is constructed, the feature components are clustered by support vector machine with radial basis kernel function, the parameters are optimized according to cross validation, and finally the system state is divided into three categories of normal, sub-health and abnormal.

[0097] In step (1) of the application, the generation and analysis process of the hexagon fully embodies the dynamic characteristics of the system operation, and the sampling frequency and harmonic analysis range can be adjusted according to the actual scene.

[0098] Through the above steps, the running data of the AC-DC conversion system is monitored in real time and the hexagon is generated, which can effectively capture the parameter change of the passive device and lay a foundation for state evaluation and optimization control.

[0099] (2) The historical running parameters of the passive device are combined with the hexagon feature parameters, data fusion and machine learning algorithms are used for processing, a mapping relationship with the AC-DC conversion system is established, and the failure mode of the passive device is obtained, and whether the parameter change of the passive device affects the AC-DC conversion system is judged according to the preset performance threshold;

[0100] The step (2) comprises the following steps:

[0101] (2.1) Obtain the historical running record of the passive device in the memory, and the historical running record contains device running time, temperature parameter, voltage stress parameter and current stress parameter;

[0102] After reading each parameter, the current hexagon feature parameter set and the power factor and measurement accuracy parameters in the metering cabinet are collected, the data is marked according to time sequence, the abnormal value and missing value are removed to obtain the first data set;

[0103] (2.2) Construct a device stress parameter matrix according to the historical running record, and the stress parameter matrix obtains the failure mode classification result by Gaussian mixture clustering;

[0104] According to the historical operation record, the device stress parameter matrix is constructed, specifically for the passive device parameters in the first data set, the temperature change curve, the voltage stress curve, and the current stress curve are extracted, the device stress parameter matrix is constructed, and the corresponding relationship between the device stress parameter and the operation time is obtained; then, according to the device failure record in the passive device historical operation data record, the device failure data is classified by using Gaussian mixture clustering, the failure mode classification result is obtained, and the corresponding relationship between the failure mode and the stress parameter matrix is established;

[0105] (2.3) The mapping relationship between the stress parameter matrix and the AC / DC conversion system operation parameter is established by using a recurrent neural network, and the operation parameter includes a power factor value and a measurement accuracy value;

[0106] Specific operation parameters include input parameters and output parameters, the input parameters include temperature change, voltage stress value, and current stress value, and the output parameters include power factor and measurement accuracy value;

[0107] (2.4) The influence degree of the passive device on the performance of the AC / DC conversion system is judged, specifically including: constructing a device performance degradation index according to the failure mode classification result;

[0108] According to the failure mode classification result, the failure probability distribution curve of each type is calculated, and the device performance degradation index is constructed in combination with the stress parameter matrix, including temperature degradation index, voltage degradation index, and current degradation index;

[0109] (2.5) The temperature threshold parameter, the voltage threshold parameter, and the current threshold parameter are extracted according to the performance degradation index;

[0110] The specific extraction method is to construct a degradation curve according to the device performance degradation index, and to extract the state parameters corresponding to the preset performance threshold from the degradation curve, including the temperature threshold parameter, the voltage threshold parameter, and the current threshold parameter;

[0111] (2.6) The power factor deviation value and the measurement accuracy deviation value are calculated through the threshold parameter;

[0112] The power factor deviation value and the measurement accuracy deviation value of the AC / DC conversion system are calculated according to the state parameter, and the influence degree of the passive device parameter change on the performance of the AC / DC conversion system is judged through the deviation value.

[0113] In step (2.1), the collection of historical operation data can be based on a fixed time interval, for example, recording temperature and stress value once every hour, to ensure the continuity and representativeness of the data;

[0114] In step (2.2), the correspondence between the device stress parameters and the running time is that each row of the matrix corresponds to the parameter value at a certain time point, and each column represents a different stress dimension, such as temperature deviation, voltage ratio, and current peak ratio, which reflects the stress state of the device at different running stages;

[0115] In the failure mode analysis stage, the Gaussian mixture clustering algorithm is used to classify the device failure records in the historical running data, and the distribution characteristics of multiple failure modes are obtained, for example, for electrolytic capacitors, the failure types such as electrolyte leakage, internal short circuit, and open circuit can be identified, each type of failure mode is associated with a specific combination of stress parameters, then the probability distribution curves of each type of failure mode are calculated according to the classification results, and the performance degradation indicators are constructed in combination with the device stress parameter matrix, these indicators include temperature degradation factor, voltage degradation factor, and current degradation factor, which are used to quantify the degradation degree of the device over time; in practical application, the Gaussian mixture clustering divides the failure data into multiple clusters through iterative optimization of the expectation maximization algorithm, the center point of each cluster represents a typical failure mode, and the probability distribution curve reveals the occurrence trend of different failure modes with the increase of running time.

[0116] In step (2.3), in the mapping relationship construction stage, the recursive neural network is used to train the device stress parameter matrix and the running parameters of the AC / DC conversion system, and the nonlinear mapping relationship between the two is established, wherein the input layer receives stress parameters such as temperature change, voltage stress value, and current stress value, the output layer generates system performance indicators such as power factor value and measurement accuracy value, through the processing of the hidden layer of the network, the long short-term memory unit is used to capture the time dependence, ensuring that the model can reflect the dynamic influence of the evolution of the device state over time, after training, the mapping relationship is used to predict the potential influence of passive device parameter changes on system performance, in this step, the structure of the recursive neural network can contain 128 hidden layer neurons, through multiple iterations to optimize the weights, so that the prediction error is controlled within an acceptable range.

[0117] In step (2.6), the calculation of the deviation value can be obtained by multiple simulation running data to ensure the reliability of the judgment result.

[0118] In step (2) of the present application, the fusion of the hexagon diagram characteristic parameters and the historical data through the above steps can clearly reveal the causal relationship between the passive device parameter changes and the system performance. After the stress parameters are recorded in the form of a matrix, the degradation trend can be intuitively reflected, and the prediction results of the recursive neural network further quantify the specific influence of these changes on the power factor and the measurement accuracy. The algorithm implementation details of this step can be adjusted by technicians according to actual scenarios to adapt to the needs of different types of passive devices.

[0119] (3) If it is determined that the passive device parameter change has an impact on the AC-DC conversion system, a passive device degradation early warning signal is generated, and the reactive power compensation capacity, the reactive power compensation device parameter and the filter parameter are adjusted to determine the adjustment amount;

[0120] The step (3) comprises the following steps:

[0121] (3.1) receiving a warning data matrix in which at least one of the device temperature change, the voltage change and the current change exceeds the preset threshold interval, and classifying the warning data matrix by a multilayer perception machine to obtain a warning level value;

[0122] According to the passive device degradation early warning threshold, the device parameter change is determined. When any one of the temperature change, the voltage change and the current change exceeds the preset threshold interval, a warning data matrix containing the change value is generated, and a warning level value is obtained by classifying the warning data matrix by a multilayer perception machine;

[0123] (3.2) constructing a compensation parameter calculation matrix for the warning level value, extracting a reactive power compensation capacity reference value, a compensation device reference parameter and a compensation operating point reference value from a compensation parameter database according to the warning level value, and calculating a compensation capacity adjustment value by a reactive power compensation algorithm;

[0124] (3.3) extracting a frequency range reference value, a bandwidth reference value and a gain reference value from a filter parameter database according to the warning level value and the compensation capacity adjustment value, and obtaining a filter parameter adjustment value by least square method operation;

[0125] (3.4) constructing a correction parameter matrix according to the compensation capacity adjustment value and the filter parameter adjustment value, and obtaining a final adjustment value of the reactive power compensation capacity, a final adjustment value of the compensation device parameter and a final adjustment value of the filter parameter by optimization operation of the correction parameter matrix by a random forest regression method.

[0126] The correction parameter matrix is normalized by a data standardization method to obtain a normalized parameter matrix, a compensation parameter optimization vector is constructed according to the normalized parameter matrix, and then the optimization vector is solved by a random forest regression method to obtain the final adjustment value of the reactive power compensation capacity, the final adjustment value of the compensation device parameter and the final adjustment value of the filter parameter.

[0127] In step (3.1), the multilayer perception machine adopts a three-layer structure, the input layer receives the change value data, the hidden layer processes the nonlinear relationship by an activation function such as ReLU, and the output layer gives a warning level value from 1 to 4, indicating the degradation degree from light to heavy;

[0128] In step (3.2), in the parameter adjustment stage, the reactive power compensation algorithm determines the specific value of the compensation capacity to be increased based on the difference between the power factor target value and the current value through iterative optimization. The core is to calculate the compensation demand through real-time power factor deviation and calibrate it combined with historical adjustment effect to ensure that the adjustment amount is neither excessive compensation nor insufficient;

[0129] In step (3.3), in the filter optimization stage, the filter parameter adjustment value is obtained through least square method operation to adapt to the harmonic change brought by the increase of compensation capacity. The application of least square method ensures that the adjusted filter parameter is highly matched with the system operation state by minimizing the sum of squares of errors. In actual scenarios, the adjustment of filter parameters is closely related to reactive compensation, aiming to suppress the additional harmonic interference caused by device degradation and improve system stability.

[0130] In step (3.4), in the optimization calculation stage, random forest regression can effectively avoid the overfitting problem of single model through multiple sampling and voting mechanism to ensure the robustness of the adjustment result.

[0131] In step (3.4) of the present application, the above steps are implemented to realize the closed-loop control from the generation of early warning signals to parameter adjustment. Taking electrolytic capacitor as an example, when the temperature change exceeds 15 degrees Celsius and triggers the early warning, the system not only generates the corresponding signal, but also quickly classifies the degradation level through multi-layer perception, combines the database reference value and algorithm operation, and dynamically adjusts the reactive compensation and filter parameters. This adjustment can effectively cope with the influence of device degradation on power factor and measurement accuracy, while avoiding conflicts or system shocks in parameter adjustment. The implementation of this step does not require manual intervention, and all calculation processes are automatically completed, ensuring the efficiency and consistency of the adjustment. The specific parameter adjustment range can be flexibly set by technical personnel according to the actual device characteristics.

[0132] (4) Adopting machine learning algorithm, taking the adjustment amount and historical parameter data of passive device as input, predicting the trend of passive device parameter change, combining the prediction result to determine the compensation mode, compensation capacity configuration and switching mode of reactive power compensation device, and dynamically compensating and coordinating the power factor of the whole cluster;

[0133] The step (4) includes the following steps:

[0134] (4.1) Construct a training data matrix according to the historical parameter data of the passive device, process the training data matrix through a long short-term memory network, and obtain a passive device parameter change trend prediction curve;

[0135] The training data matrix contains time series values of temperature parameters, voltage parameters and current parameters;

[0136] (4.2) extracting a passive device parameter variation time distribution sequence according to the predicted curve, constructing a parameter variation matrix, the parameter variation matrix containing temperature variation, voltage variation and current variation, and calculating a compensation mode selection sequence number according to the parameter variation matrix;

[0137] (4.3) extracting a compensation capacity initial configuration value corresponding to the compensation mode selection sequence number from a compensation device parameter database, optimizing the compensation capacity initial configuration value through a Kalman filter, calculating an optimal estimation value according to a measurement noise variance and a process noise variance, and obtaining a compensation capacity optimized configuration value;

[0138] The compensation capacity initial configuration value contains reactive power compensation, harmonic compensation and unbalance compensation.

[0139] (4.4) generating a compensation device switching time sequence table according to the compensation capacity optimized configuration value, the compensation device switching time sequence table containing a switching sequence number, a switching time interval and a switching holding time, adjusting the compensation device switching time sequence table in real time through an adaptive feedback compensation algorithm, and calculating a compensation deviation according to a real-time power factor measurement value.

[0140] Adjusting the compensation device switching parameters to obtain the final compensation control instruction.

[0141] (4.5) if the overall power factor is lower than a preset threshold, determining a reactive power compensation adjustment range, calculating a reactive power compensation device operating parameter adjustment value according to the adjustment range, updating the operating strategy of the reactive power compensation device through the parameter adjustment value, iteratively updating the overall power factor of the cluster, judging whether the updated overall power factor reaches a preset optimization target value, if the optimization target value is not reached, updating the operating strategy again until the overall power factor reaches the optimization target value.

[0142] The step (4.5) includes the following steps:

[0143] (4.5.1) acquiring a real-time power factor value according to a power factor measurement device, and judging whether the real-time power factor value is lower than a preset power factor threshold;

[0144] (4.5.2) if the real-time power factor value is lower than the preset power factor threshold, obtaining compensation capacity parameters and operating state parameters from a compensation device parameter library;

[0145] (4.5.3) constructing a compensation parameter optimization objective function through the compensation capacity parameters and the operating state parameters, and solving the compensation parameter optimization objective function to obtain a reactive power compensation adjustment interval by using a gradient descent method.

[0146] (4.5.4) According to the reactive compensation amount adjustment interval, an adaptive learning rate algorithm is used to optimize the compensation capacity value and switch control value to obtain a compensation parameter adjustment matrix; the input parameters include the compensation capacity value, the switch control value, and the response time.

[0147] (4.5.5) The compensation capacity correction value and the switch timing correction value are extracted from the compensation parameter adjustment matrix, and the compensation device control instruction sequence is updated according to the compensation capacity correction value and the switch timing correction value;

[0148] (4.5.6) According to the compensation device control instruction sequence, the power factor measurement value is collected in the reactive power compensation operation, and it is judged whether the power factor measurement value reaches the preset optimization target value.

[0149] The measured value is compared with the preset optimization target value. If the measured value does not reach the target value and does not exceed the maximum iteration number, return to step 2 to recalculate the compensation amount adjustment interval until the termination condition is met.

[0150] In step (4.1), in the parameter trend prediction stage, the historical parameter data of the passive device is obtained from the storage, a multi-dimensional training data matrix is constructed, and then the long short-term memory network is used for time series modeling of the matrix. The network adopts a three-layer structure, the input layer receives the time series of daily parameters, the hidden layer is equipped with 64 neurons to capture long-term dependencies, and the output layer generates a 7-day parameter change trend prediction curve. Then the time distribution sequence of the temperature, voltage and current change amount is extracted from the prediction curve to construct a parameter change matrix. In this step, the long short-term memory network can effectively retain the key patterns in the historical data through the forgetting gate and update gate mechanism, thereby improving the prediction accuracy;

[0151] In step (4.2), in the compensation strategy determination stage, the compensation mode selection number is calculated according to the parameter change matrix. The fast compensation mode is selected for rapid changes, and the conventional compensation mode is selected for slow changes;

[0152] In step (4.3), the Kalman filter is introduced to smooth the measurement error and model uncertainty, and to ensure that the compensation capacity is closer to the actual demand of the system;

[0153] In step (4.4), the adaptive feedback compensation algorithm is used to adjust the compensation device switching timing table in real time. The compensation deviation is calculated according to the real-time power factor measurement value to form the final compensation control instruction for smooth adjustment.

[0154] In step (4) of the present application, the full flow automation from parameter prediction to compensation control is realized through the above steps, not only reducing the compensation lag, but also avoiding the equipment impact through multi-level progressive control, significantly improving the stability and efficiency of the cluster operation, and the specific parameter configuration can be further adjusted according to the actual load characteristics.

[0155] (5) According to the compensation mode, compensation capacity configuration and switching mode of the reactive power compensation device, and the overall power factor of the cluster and the passive device degradation early warning signal, the particle swarm optimization algorithm is used to coordinate the reactive power compensation devices of several low-voltage metering cabinets, adjust the output capacity and operating state of the reactive power compensation device, and optimize the output capacity and operating state to improve the system performance.

[0156] The step (5) comprises the following steps:

[0157] (5.1) Obtain the low-voltage metering cabinet compensation device operating data to construct an operating state matrix, which contains compensation mode attribute values and capacity configuration values and switching mode serial numbers.

[0158] (5.2) According to the operating state matrix, build a fitness function, which contains compensation effect constraint terms and device operating constraint terms.

[0159] The fitness function of the particle swarm optimization algorithm is built by collecting the cluster power factor measurement value and the passive device degradation early warning signal.

[0160] (5.3) Search the parameters of the fitness function through the particle swarm optimization algorithm, the particle position represents the compensation capacity distribution ratio, the particle velocity represents the capacity adjustment step, and the search range is limited within the rated capacity range of the compensation device.

[0161] (5.4) Calculate the device load rate according to the compensation capacity distribution ratio, the load rate is the ratio of real-time capacity to rated capacity, and the compensation device is controlled according to the load rate value.

[0162] (5.5) Use a deep neural network to predict the load rate, and if the predicted load rate exceeds the preset threshold, correct the compensation device control instruction sequence according to the predicted load rate to obtain the corrected control instruction sequence.

[0163] The compensation device control instruction sequence is generated based on the particle swarm optimization result, and contains switching sequence, capacity allocation value and response time, and the operation state of each compensation device is controlled in stages.

[0164] In step (5.1), the compensation mode attribute value includes on-site compensation or centralized compensation.

[0165] In step (5.2), the compensation effect constraint term and the device operation constraint term aim to balance the compensation efficiency and the equipment safety, in this step, the design of the fitness function integrates multiple objectives through a weighting method, ensuring that the optimization result meets the power factor demand and avoids overload;

[0166] In step (5.4), the grouping control realizes hierarchical management, and the particle swarm optimization algorithm can quickly converge to a reasonable allocation scheme through simulating group cooperation, thereby improving the calculation efficiency.

[0167] In step (5.5), the deep neural network training set is constructed based on historical operation data, the input features include 24-hour load rate, capacity allocation value and operation time, the network adopts a three-layer structure with 64 nodes in the hidden layer, and the future 4-hour load rate is predicted through nonlinear mapping, and then the control instruction sequence is corrected according to the prediction result to reduce the operation pressure, and other device parameters remain stable, in this step, the deep neural network can identify load abnormalities in advance through learning historical patterns, thereby ensuring the timeliness and accuracy of adjustment.

[0168] In step (5.5), the deep neural network training set is constructed based on historical operation data, the input features include 24-hour load rate, capacity allocation value and operation time, the network adopts a three-layer structure with 64 nodes in the hidden layer, and the future 4-hour load rate is predicted through nonlinear mapping, and then the control instruction sequence is corrected according to the prediction result to reduce the operation pressure, and other device parameters remain stable, in this step, the deep neural network can identify load abnormalities in advance through learning historical patterns, thereby ensuring the timeliness and accuracy of adjustment.

[0169] (6) Real-time monitoring of the change of the characteristic parameters of the hexagon diagram, adopting a time series analysis method to perform trend fitting on the historical characteristic parameters of the hexagon diagram, judging the acceleration point of the passive device degradation, and adjusting the operation parameters of the reactive power compensation device according to the acceleration point of the passive device degradation to maintain the system stability.

[0170] The step (6) comprises the following steps:

[0171] (6.1) According to the real-time monitoring device, the characteristic parameters of the hexagon diagram are collected, and the characteristic parameters include voltage phase parameters, current phase parameters and harmonic content parameters; time series data sets of the characteristic parameters are obtained by extracting sampling points from the characteristic parameters at fixed time intervals;

[0172] (6.2) The trend curve of the parameter change is obtained by fitting the time series data set of the characteristic parameters by the exponential smoothing algorithm, the slope of the trend curve is calculated to obtain the parameter change rate curve, the acceleration value is calculated according to the change rate curve, and the abnormal point is marked;

[0173] When the acceleration exceeds the preset threshold, it is marked as an abnormal point, and the initial degradation acceleration point set is obtained by clustering the abnormal points;

[0174] (6.3) The long short-term memory network is used to receive the parameter change rate and acceleration data, and the abnormal points are verified by the parameter change rule extraction unit in the network hidden layer to obtain the degradation acceleration point;

[0175] The long short-term memory network is used to train the historical sequence of the characteristic parameters, the input layer receives the parameter change rate and acceleration data, the hidden layer includes the parameter change rule extraction unit, and the output layer gives the parameter change prediction value. According to the network prediction result, the initial degradation acceleration point set is verified, the abnormal points that do not conform to the change rule are removed, and the final position of the degradation acceleration point is obtained;

[0176] (6.4) The compensation parameter adjustment amount is calculated according to the parameter change amplitude of the degradation acceleration point position, the compensation device operating parameters are corrected according to the compensation parameter adjustment amount, and the compensation device operating parameter update instruction is generated;

[0177] The compensation parameter reference value is read from the reactive power compensation device parameter library, the compensation parameter adjustment amount is calculated according to the parameter change amplitude of the degradation acceleration point position, the compensation device operating parameters are corrected based on the compensation parameter adjustment amount, including the compensation capacity increase and decrease value, the switching delay adjustment value and the response rate change value, and the compensation device operating parameter update instruction is generated;

[0178] (6.5) The time series data of the historical hexagon diagram characteristic parameters are obtained, the characteristic parameter change rate of each time point is calculated by using the sliding window method, the trend fitting of the characteristic parameter change rate is performed, the linear regression model of the time series is constructed, the trend line slope value is obtained, and if the trend line slope value exceeds the preset threshold, the time point is marked as a potential degradation point. Calculate the fitting index, and calculate the second-order difference of the potential degradation point to determine whether the second-order derivative of the change rate appears a turning point, if so, the corresponding point is determined as the acceleration point of the passive device degradation.

[0179] The step (6.5) comprises the following steps:

[0180] (6.5.1) According to the low-voltage metering cabinet measuring device, the historical hexagon diagram characteristic parameters are collected, the characteristic parameters are sampled through a fixed length sliding window, and a characteristic parameter change rate data set is obtained;

[0181] The hexagon diagram characteristic parameters include voltage harmonic parameters, current harmonic parameters and phase parameters, the window length is set as a multiple of the sampling period, the parameter change rate in the window is calculated, and a characteristic parameter change rate data set is constructed;

[0182] (6.5.2) The linear regression algorithm is used to fit the characteristic parameter change rate data set, the trend line slope value of the fitting curve is calculated, and if the slope value is greater than a preset change rate threshold, the sampling point is determined as a potential abnormal point;

[0183] (6.5.3) The goodness-of-fit index is calculated for the potential abnormal point, and if the determination coefficient, fitting standard deviation and correlation coefficient simultaneously satisfy the preset goodness condition, the sampling point is recorded to an abnormal point data set;

[0184] (6.5.4) The first-order difference operation is performed on the abnormal point data set through the central difference method, a first-order derivative sequence is obtained, the second-order difference operation is performed on the first-order derivative sequence, a second-order derivative sequence is obtained, and if the second-order derivative changes from negative to positive, the sampling point is determined as a degradation acceleration point;

[0185] Before the first-order difference operation is performed on the smooth sequence through the central difference method, the abnormal point data set is subjected to spline interpolation processing, a continuous change rate curve is generated, and a change rate smooth sequence is constructed.

[0186] In the acceleration point verification and adjustment stage, the long short-term memory network is used to deeply analyze the parameter change rate and acceleration, then the reference value is extracted from the reactive power compensation device parameter library, the adjustment amount is calculated according to the parameter change amplitude at the acceleration point, and finally the update instruction is generated to adjust the operating parameters, so that the system can respond to the degradation influence in time, the combination of network verification and parameter adjustment improves the accuracy and practicality of the acceleration point judgment;

[0187] In step (6) of the present application, the above steps are used to realize accurate monitoring and dynamic optimization of passive device degradation, this method combines real-time data and historical trends, not only improves the diagnosis efficiency, but also prolongs the service life of the equipment through timely adjustment of operating parameters, and the specific threshold value and window length can be optimized according to actual requirements.

Claims

1. A passive vector parameter real-time monitoring method in an AC-DC conversion process, characterized in that, The method comprises the following steps: (1) collecting the operation data of the AC-DC conversion system in the low-voltage metering cabinet, generating a hexagon diagram to extract characteristic parameters, and analyzing the parameters to evaluate the state of the passive device; (2) combining the historical operation parameters of the passive device with the characteristic parameters of the hexagon diagram, establishing a mapping relationship with the AC-DC conversion system, obtaining the failure mode of the passive device, and judging whether the parameter change of the passive device affects the AC-DC conversion system according to the preset performance threshold; (3) if the parameter change of the passive device affects the AC-DC conversion system, a passive device degradation warning signal is generated, and the reactive power compensation capacity, the reactive power compensation device parameters and the filter parameters are adjusted to determine the adjustment amount; (4) using a machine learning algorithm, the adjustment amount and the historical parameter data of the passive device are used as input to predict the trend of the passive device parameter change, and the compensation mode, compensation capacity configuration and switching mode of the reactive power compensation device are determined according to the prediction result, and the power factor of the cluster is dynamically compensated and coordinated; (5) according to the compensation mode, compensation capacity configuration and switching mode of the reactive power compensation device, the overall power factor of the cluster and the passive device degradation warning signal, the particle swarm optimization algorithm is used to coordinate the reactive power compensation devices of several low-voltage metering cabinets, and the output capacity and operating state of the reactive power compensation device are adjusted; (6) real-time monitoring of the change of the characteristic parameters of the hexagon diagram, using time series analysis method to trend fitting the historical hexagon diagram characteristic parameters, judging the acceleration point of passive device degradation, and adjusting the operating parameters of the reactive power compensation device according to the acceleration point of passive device degradation; The step (4) comprises the following steps: (4.1) constructing a training data matrix according to the historical parameter data of the passive device, processing the training data matrix through a long short-term memory network to obtain a passive device parameter change trend prediction curve; (4.2) extracting a parameter change amount time distribution sequence according to the prediction curve, constructing a parameter change matrix, the parameter change matrix includes temperature change amount, voltage change amount and current change amount, and calculating a compensation mode selection serial number according to the parameter change matrix; (4.3) extracting a compensation capacity initial configuration value corresponding to the compensation mode selection serial number from a reactive power compensation device parameter database, optimizing the compensation capacity initial configuration value through a Kalman filter, calculating an optimal estimation value according to a measurement noise variance and a process noise variance, and obtaining a compensation capacity optimized configuration value; (4.4) generating a reactive power compensation device switching time sequence table according to the compensation capacity optimized configuration value, the reactive power compensation device switching time sequence table includes a switching sequence number, a switching time and an interval switching holding time, and adjusting the reactive power compensation device switching time sequence table in real time through an adaptive feedback compensation algorithm, and calculating a compensation deviation according to a real-time power factor measurement value. (4.5) If the overall power factor is lower than a preset threshold, determine a reactive power compensation amount adjustment range, calculate a reactive power compensation device operation parameter adjustment value according to the adjustment range, update the operation strategy of the reactive power compensation device through the parameter adjustment value, iteratively update the overall power factor of the cluster, judge whether the updated overall power factor reaches a preset optimization target value, if not, update the operation strategy again until the overall power factor reaches the optimization target value; The step (5) comprises the following steps: (5.1) Obtain the reactive power compensation device operation data of the low-voltage metering cabinet to construct an operation state matrix, wherein the operation state matrix comprises compensation mode attribute values, compensation capacity configuration values and switching mode serial numbers; (5.2) Construct a fitness function according to the operation state matrix, wherein the fitness function comprises a compensation effect constraint term and a device operation constraint term; (5.3) Perform parameter search on the fitness function through a particle swarm optimization algorithm, wherein a particle position represents a compensation capacity distribution ratio and a particle speed represents a capacity adjustment step length; (5.4) Calculate a reactive power compensation device load rate according to the compensation capacity distribution ratio, wherein the load rate is a ratio of a real-time capacity to a rated capacity, and the reactive power compensation device is controlled in groups according to the load rate value; (5.5) Predict the load rate by using a deep neural network, and if the predicted load rate exceeds a preset threshold, correct a reactive power compensation device control instruction sequence according to the predicted load rate to obtain a corrected control instruction sequence.

2. The method of claim 1, wherein the passive vector parameter real-time monitoring method in AC-DC conversion process is characterized in that, The step (1) comprises the following steps: (1.1) Sample the AC-DC conversion system of the low-voltage metering cabinet according to a preset sampling frequency to obtain AC voltage and current waveform data; (1.2) Process the AC voltage and current waveform data by Fourier transform to extract amplitude and phase characteristics of fundamental wave signals and harmonic signals from the waveform data; (1.3) Construct a hexagon diagram coordinate matrix according to the characteristics of the fundamental wave signals and the harmonic signals, wherein the hexagon diagram coordinate matrix comprises voltage amplitude, current amplitude, phase difference value, fundamental wave frequency, total harmonic distortion rate and power grid frequency deviation value; (1.4) Perform principal component analysis on the hexagon diagram coordinate matrix, cluster feature components obtained by the principal component analysis by using a support vector machine, and obtain an evaluation index of monitoring parameters.

3. The method of claim 1, wherein the passive vector parameter real-time monitoring method in AC-DC conversion process is characterized in that, The step (2) comprises the following steps: (2.1) Obtain a passive device historical operation record in a storage, wherein the historical operation record comprises device operation time, temperature parameter, voltage stress parameter and current stress parameter; (2.2) Construct a device stress parameter matrix according to the historical operation record, wherein the stress parameter matrix obtains a failure mode classification result by Gaussian mixture clustering; (2.3) Establish a mapping relationship between the stress parameter matrix and AC-DC conversion system operation parameters by using a recurrent neural network, wherein the operation parameters comprise power factor value and metering accuracy value; (2.4) Judge the influence degree of the passive device on the performance of the AC-DC conversion system, specifically comprising: constructing a device performance degradation index according to the failure mode classification result; (2.5) extracting temperature threshold parameters, voltage threshold parameters and current threshold parameters according to the performance degradation index; (2.6) calculating power factor deviation values and measurement accuracy deviation values through temperature threshold parameters, voltage threshold parameters and current threshold parameters.

4. The method of claim 1, wherein the passive vector parameter real-time monitoring method in AC-DC conversion process is characterized in that, The step (3) comprises the following steps: (3.1) receiving a pre-warning data matrix in which at least one of the device temperature change amount, the voltage change amount and the current change amount exceeds the preset threshold interval, and classifying the pre-warning data matrix through a multi-layer perception machine to obtain a pre-warning level value; (3.2) extracting a reactive power compensation capacity reference value, a reactive power compensation device reference parameter and a compensation operating point reference value from a compensation parameter database according to the pre-warning level value, and calculating a compensation capacity adjustment value by using a reactive power compensation algorithm; (3.3) extracting a frequency range reference value, a bandwidth reference value and a gain reference value from a filter parameter database according to the pre-warning level value and the compensation capacity adjustment value, and obtaining a filter parameter adjustment value by least square method operation; (3.4) constructing a correction parameter matrix according to the compensation capacity adjustment value and the filter parameter adjustment value, and obtaining a final reactive power compensation capacity adjustment value, a final reactive power compensation device parameter adjustment value and a final filter parameter adjustment value by optimizing operation of the correction parameter matrix through a random forest regression method.

5. The method of claim 1, wherein the method is characterized by: The step (6) comprises the following steps: (6.1) collecting hexagon diagram characteristic parameters according to a real-time monitoring device, the characteristic parameters including voltage phase parameters, current phase parameters and harmonic content parameters, and extracting sampling points from the characteristic parameters at fixed time intervals to obtain a characteristic parameter time series data set; (6.2) fitting the characteristic parameter time series data set through an exponential smoothing algorithm to obtain a parameter change trend curve, calculating the slope of the trend curve to obtain a parameter change rate curve, calculating an acceleration value according to the change rate curve and marking an abnormal point; (6.3) receiving the parameter change rate and acceleration data through a long short-term memory network, verifying the abnormal point through a parameter change law extraction unit in a network hidden layer, and obtaining a degradation acceleration point; (6.4) calculating a compensation parameter adjustment amount according to the parameter change amplitude of the degradation acceleration point position, correcting the operation parameters of the reactive power compensation device according to the compensation parameter adjustment amount, and generating an operation parameter update instruction of the reactive power compensation device; (6.5) obtaining time series data of historical hexagon diagram characteristic parameters, calculating the characteristic parameter change rate at each time point through a sliding window method, fitting the characteristic parameter change rate, constructing a linear regression model of the time series, obtaining a trend line slope value, and marking the time point as a potential degradation point if the trend line slope value exceeds a preset threshold value, calculating a fitting degree index, and judging whether the second derivative of the change rate appears a turning point through second-order difference calculation, and determining the corresponding point as an acceleration point of passive device degradation if yes.

6. The method of claim 1, wherein the passive vector parameter real-time monitoring method in AC-DC conversion process is characterized in that, The step (4.5) comprises the following steps: (4.5.1) collecting a real-time power factor value according to a power factor measuring device, and judging whether the real-time power factor value is lower than a preset power factor threshold value; (4.5.2) if the real-time power factor value is lower than a preset power factor threshold, obtaining compensation capacity parameters and operating state parameters from a reactive power compensation device parameter library; (4.5.3) constructing a compensation parameter optimization objective function through the compensation capacity parameters and the operating state parameters, and solving the compensation parameter optimization objective function by using a gradient descent method to obtain a reactive power compensation amount adjustment interval; (4.5.4) according to the reactive power compensation amount adjustment interval, using an adaptive learning rate algorithm to optimize compensation capacity values and switch control amounts to obtain a compensation parameter adjustment matrix; (4.5.5) extracting compensation capacity correction values and switch timing correction values from the compensation parameter adjustment matrix, and updating a reactive power compensation device control instruction sequence according to the compensation capacity correction values and the switch timing correction values; (4.5.6) collecting a power factor measurement value according to the reactive power compensation device control instruction sequence, and judging whether the power factor measurement value reaches a preset optimization target value.

7. The method of claim 5, wherein the passive vector parameter real-time monitoring method in AC-DC conversion process is characterized by, The step (6.5) comprises the following steps: (6.5.1) collecting historical hexagon diagram feature parameters according to a real-time monitoring device of a low-voltage metering cabinet, and obtaining a feature parameter change rate data set by sampling the feature parameters through a fixed length sliding window; (6.5.2) fitting the feature parameter change rate data set by using a linear regression algorithm, calculating a trend line slope value of a fitting curve, and determining that the sampling point is a potential abnormal point if the slope value is greater than a preset change rate threshold; (6.5.3) calculating a goodness of fit index for the potential abnormal point, and recording the sampling point to an abnormal point data set if the determination coefficient, the fitting standard deviation and the correlation coefficient simultaneously satisfy a preset goodness condition; (6.5.4) performing first-order difference operation on the abnormal point data set by using a central difference method to obtain a first-order derivative sequence, performing second-order difference operation on the first-order derivative sequence to obtain a second-order derivative sequence, and determining that the sampling point is a degradation acceleration point if the second-order derivative changes from negative to positive.

Images