Self-healing operation and maintenance method for state of power equipment

Abstract

The invention relates to a power equipment state self-healing operation and maintenance method, and belongs to the field of power equipment operation and maintenance, and the method comprises the steps: obtaining the real-time operation state data of a power supply in a current multi-source power supply system, collecting the data through a preset sampling frequency, and constructing a power supply state matrix; a conversion time constraint condition is set in advance, a zero-delay switching decision tree is constructed and a power supply switching decision is output in combination with a main power supply fault risk assessment result, and if the main power supply fault risk assessment result is abnormal, the standby power supplies are switched according to the determined switching priorities of the standby power supplies and the power supply switching decision; after the switching is completed, monitoring the power supply continuity index of the power equipment in real time, and if the power supply fluctuation time or the phase jump amplitude exceeds a preset allowable value, smoothing the power supply fluctuation by adopting the rapid power compensation function of the energy storage device; the stability and reliability of the multi-source power supply system are improved, and intelligent power supply switching and self-healing disaster recovery are realized.

Description

Technical Field

[0001] The present invention relates to a self-healing operation and maintenance method for power equipment status, belonging to the field of power equipment operation and maintenance. Background Art

[0002] In a multi-source power supply system, when the main power supply shows hidden fault signs such as high-order harmonic distortion or voltage sag, the probability of the backup power supply being put into use needs to be adaptively adjusted based on the load forecast results and the fault evolution trend. However, this process involves straightening out multiple technical contradictions and sorting out logical relationships.

[0003] First, the fault symptoms of the main power supply are often uncertain and dynamic. How to accurately identify and predict the evolution trend of these faults is a key issue.

[0004] Secondly, load fluctuations have a direct impact on the probability of backup power supply deployment, and the backup power supply deployment strategy needs to be dynamically adjusted under different load conditions.

[0005] Thirdly, the synchronization characteristics of the multi-source power supply system have complex migration patterns under different load fluctuations. How to ensure that the backup power supply can be seamlessly connected with the system when it is put into use and avoid the zero-delay problem during the switching process is a technical difficulty.

[0006] In addition, the impact of the probability adjustment of backup power supply on the continuity of live broadcast power supply also needs in-depth analysis. How to ensure the continuity of power supply while avoiding excessive or unnecessary investment in backup power supply is an issue that needs to be balanced.

[0007] These technical contradictions are intertwined, forming a complex technical problem that requires in-depth research and resolution from multiple dimensions. Summary of the Invention

[0008] According to the problem described in the background, the problem to be solved by the present invention is: to provide a self-healing operation and maintenance method for the status of power equipment to solve the problems mentioned above.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a self-healing operation and maintenance method for power equipment status, comprising the following steps:

[0010] (1) Acquire the real-time operating status data of the power supply in the current multi-source power supply system. The operating status data includes voltage, current and power parameters. The data is collected at a preset sampling frequency to construct a power supply status matrix.

[0011] (2) Obtain the voltage and current data in the power supply status matrix, use fast Fourier transform to calculate the high-order harmonic distortion rate of the power supply, calculate the phase difference between the main power supply voltage and current, and analyze the change trend of the phase difference over time. At the same time, extract the time-frequency characteristics of the voltage and current signals, build a hidden fault symptom recognition model, and output the hidden fault symptom recognition results;

[0012] (3) If the output recognition result is that there is a hidden fault sign, then obtain the harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence data of the main power supply within the preset time window, analyze and obtain the trend characteristics of the fault evolution, analyze the trend characteristics to obtain the fault evolution trend, and perform trend prediction on the load power change of the main power supply to obtain the load fluctuation curve within the preset time;

[0013] (4) Based on the load fluctuation curve, combined with the historical switching failure data of the main power supply and the fault evolution trend, a power supply switching risk assessment model is established, and the main power supply failure risk assessment result is output. The main power supply failure risk assessment result is combined with the historical operation data of the backup power supply to calculate the target input probability of the backup power supply, and a weight distribution algorithm is used to determine the switching priority of each backup power supply;

[0014] (5) Pre-set the switching time constraint conditions, combine the main power supply failure risk assessment results, build a zero-delay switching decision tree, and output the power supply switching decision. If the main power supply failure risk assessment result is abnormal, switch the backup power supply according to the determined switching priority of each backup power supply and the power supply switching decision;

[0015] (6) After the switching is completed, the power supply continuity indicators of the power equipment are monitored in real time, including voltage fluctuation, frequency fluctuation, and phase jump. If the power supply fluctuation time or phase jump amplitude exceeds the preset allowable value, the fast power compensation function of the energy storage device is used to smooth the power supply fluctuation.

[0016] Preferably, the step (1) comprises the following steps:

[0017] (1.1) Synchronously collecting operating data of each power circuit from the multi-source power supply system according to a preset sampling frequency, and obtaining a first set of real-time operating data through a data acquisition device;

[0018] (1.2) Based on the first set of real-time operation data, calculate the minimum sampling frequency value using the Shannon sampling theorem. If the actual sampling frequency is less than the minimum sampling frequency value, increase the sampling frequency and re-collect data until the sampling theorem meets the requirements to obtain a second set of real-time operation data;

[0019] (1.3) For the second set of real-time operation data, a Lagrange interpolation algorithm is used to calculate the missing data points to obtain a third set of real-time operation data;

[0020] (1.4) Establishing a power supply status matrix based on the third set of real-time operation data.

[0021] Preferably, the step (2) comprises the following steps:

[0022] (2.1) obtaining a voltage data sequence and a current data sequence according to the power supply state matrix, performing frequency domain decomposition by fast Fourier transform, and normalizing the frequency domain components obtained by the frequency domain decomposition to obtain a first harmonic distortion rate matrix;

[0023] (2.2) processing the sampling points using a zero phase detector for the fundamental wave components in the voltage data sequence and the current data sequence, and obtaining a first phase difference curve based on the phase increment values ​​between adjacent sampling points;

[0024] (2.3) Decomposing the first harmonic distortion rate matrix and the first phase difference curve using wavelet transform, and obtaining a first eigenvector group by extracting amplitude features, phase features, frequency features, and distortion features;

[0025] (2.4) If the eigenvalues ​​of the first eigenvector group obtained after mapping by the extreme learning algorithm exceed the preset threshold range, the eigenvalues ​​are marked as abnormal data points, and the abnormal data points are classified by a support vector machine to obtain a fault feature space and a fault probability value.

[0026] Preferably, the step (3) comprises the following steps:

[0027] (3.1) Obtain the harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence of the main power supply, and pre-process them through a wavelet denoiser to obtain the first trend sequence;

[0028] (3.2) constructing a time domain feature matrix based on the first trend sequence, and extracting trend components from the time domain feature matrix using singular value decomposition to obtain a second trend sequence;

[0029] (3.3) extracting time series features using a long short-term memory neural network for the second trend sequence, and establishing a feature vector group for the time series features to obtain an evolution matrix;

[0030] (3.4) The evolution matrix is ​​decomposed in time series, and a correlation characteristic diagram is constructed according to the change trend of the harmonic distortion rate, the change trend of the phase difference, and the change trend of the load power. The load fluctuation curve within a preset time is generated by combining the correlation characteristic diagram.

[0031] Preferably, the step (4) comprises the following steps:

[0032] (4.1) Obtain the load fluctuation curve, fault records, and evolution trend data of the main power supply, and perform denoising processing through wavelet decomposition to obtain the first risk feature matrix;

[0033] (4.2) training a deep neural network model based on the first risk feature matrix and using a probability density estimation method to determine the risk level of the main power supply;

[0034] (4.3) extracting the startup time, load transfer time, and voltage stability parameters from the backup power supply operation database based on the risk level, and calculating a second probability matrix using a sliding time window;

[0035] (4.4) Based on the risk level and the second probability matrix, the weighted response probability of each backup power supply is calculated by Bayesian probability to obtain the backup power supply switching priority ranking result.

[0036] Preferably, the step (5) comprises the following steps:

[0037] (5.1) Decompose the switching response time, standby startup time, and load transfer time constraints using a time axis splitter to obtain a time constraint sequence;

[0038] (5.2) Based on the time constraint sequence, a random forest algorithm is used to construct a zero-delay handover decision tree, and judgment is made according to the three dimensions of risk level, time constraint, and standby status to obtain a decision sequence;

[0039] (5.3) Based on the decision sequence, a state detector is used to perform a pre-start detection on the backup power supply, obtain voltage parameters, current parameters, and power parameters, and obtain a detection result;

[0040] (5.4) If the risk level of the main power supply exceeds the preset threshold and the backup power supply status meets the switching conditions, the synchronous controller will send a switching instruction to the backup power supply.

[0041] (5.5) During the switching process, the voltage amplitude and phase difference information between each power supply is obtained in real time, and the phase difference and voltage amplitude difference between each power supply are compared with the corresponding preset range. If the phase difference and voltage amplitude difference exceed the corresponding preset range, a control signal is generated based on the calculated adjustment amount through a dynamic adjustment algorithm to correct the synchronization characteristics of each power supply, which include frequency, voltage amplitude and phase.

[0042] Preferably, the step (6) comprises the following steps:

[0043] (6.1) Real-time sampling of voltage fluctuation data, frequency fluctuation data, and phase jump data of the power supply circuit is performed using a data collector to obtain a first fluctuation sequence;

[0044] (6.2) comparing the fluctuation amplitude of the first fluctuation sequence with a preset allowable value, marking the fluctuation parameters that exceed the preset allowable value to obtain a first abnormal sequence;

[0045] (6.3) An adaptive neural network is used to train the first abnormal sequence, and a fuzzy control rule base is set according to the fluctuation feature classification result to calculate the compensation amount to obtain a first compensation sequence;

[0046] (6.4) executing the first compensation sequence through the power controller to adjust the output power waveform of the energy storage device and perform real-time compensation for voltage fluctuations, frequency fluctuations, and phase jumps to obtain a correction sequence;

[0047] (6.5) Based on the power supply system operation data after switching, update the power supply switching risk assessment model parameters, use the reinforcement learning algorithm to optimize the calculation strategy of the backup power supply input probability, establish a power supply continuity evaluation model, and evaluate the effect of the self-healing disaster recovery method on the improvement of power supply stability by analyzing the voltage sag time and frequency changes during the switching process, and output an operation performance report.

[0048] Preferably, the step (5.5) comprises the following steps:

[0049] (5.5.1) Collect voltage amplitude and phase angle sampling data from the power supply circuit, and process the sampling data through a digital filter to obtain voltage amplitude difference and phase difference;

[0050] (5.5.2) Calculating the deviation excess ratio using a linear mapping function based on the voltage amplitude difference and phase difference, and setting a voltage amplitude deviation threshold and a phase difference deviation threshold based on the deviation excess ratio to obtain an adjustment target value;

[0051] (5.5.3) Use a neural network predictor to process the adjustment target value and optimize the prediction result through the gradient descent method to obtain the correction instruction;

[0052] (5.5.4) The correction instructions are executed by the synchronous controller to adjust the power supply circuit, and the adjustment process is monitored using closed-loop feedback to obtain correction data.

[0053] Preferably, the step (6.5) comprises the following steps:

[0054] (6.5.1) Acquire power supply switching data from the power supply circuit, the power supply switching data including voltage sag data, frequency change data, and power supply switching timing data, and perform time segmentation on the power supply switching data using a timing sampler to obtain a first operation sequence;

[0055] (6.5.2) Establishing a deep reinforcement learning state space based on the first operation sequence, wherein the state space includes power supply parameter states, switching timing states, and load response states, and mapping the state space using an action selector to obtain a first state map;

[0056] (6.5.3) performing time-frequency decomposition on the voltage sag data and the frequency variation data in the first state map using wavelet analysis to obtain a first characteristic matrix;

[0057] (6.5.4) Constructing disaster recovery evaluation indicators based on the first characteristic matrix, wherein the disaster recovery evaluation indicators include power supply stability level, disaster recovery response time, and self-healing repair time. An evaluation matrix is ​​obtained by calculating the weights of each dimension through hierarchical analysis;

[0058] (6.5.5) Based on the evaluation matrix, generate an operational performance report based on the three modules of evaluation indicators, disaster recovery effect, and optimization suggestions.

[0059] The beneficial effects of the present invention are:

[0060] 1. Collecting the real-time operating status data of each power supply and constructing a power supply status matrix can truly reflect the operating status of the multi-source power supply system, which is conducive to the subsequent accurate identification of faults and prediction of fault evolution trends.

[0061] 2. Construct a model for identifying latent fault signs and ultimately output the fault symptom identification results. Fast Fourier transform is used to analyze the signal's harmonic content from a frequency domain perspective. Phase difference analysis reveals the impact of load dynamics on the system. The combination of wavelet transform and extreme learning algorithm can efficiently extract multidimensional features and achieve rapid mapping. Support vector machine classification further improves the accuracy of fault identification. The above processing can effectively output latent fault types and their occurrence probabilities, providing reliable technical support for self-healing operation and maintenance of power equipment status.

[0062] 3. When hidden fault signs are detected, key operating data of the main power supply are immediately extracted from the preset time window, including the harmonic distortion rate sequence, the voltage and current phase difference sequence, and the load power sequence. These data reflect the dynamic changes in the power supply operating status and provide a basis for fault evolution analysis. The advantage of the multidimensional data analysis method is that it comprehensively considers the mutual influence between the harmonic distortion rate, phase difference, and load power.

[0063] 4. The load fluctuation curve reflects the dynamic changes in electricity demand, the historical switching failure records reveal the weak links in the system, and the fault evolution trend indicates the potential deterioration risk. Deep neural networks can integrate these characteristics through multi-layer nonlinear mapping to output accurate risk assessment results. The advantage of the multi-dimensional assessment method is that it not only considers the real-time status and historical experience of the main power supply, but also combines the performance differences of the backup power supply to ensure that the switching priority of each backup power supply determined by the weight distribution algorithm is consistent with actual performance.

[0064] 5. The zero-delay handover decision tree ensures accurate handover timing through multi-dimensional judgment, while real-time synchronous adjustment avoids the impact of sudden parameter changes on the load. Compared with traditional methods, this coordinated adjustment method improves handover smoothness while significantly enhancing system stability.

[0065] 6. During the switching process, the synchronization characteristics of each power supply are dynamically adjusted. After the switching, the power supply continuity indicators are monitored and the fault-tolerant compensation mechanism is triggered. This fast compensation mechanism effectively reduces the risk of unstable power supply after the switching and ensures the normal operation of the load equipment.

[0066] 7. The present invention optimizes the backup power supply input strategy through reinforcement learning, establishes a power supply continuity evaluation model, improves the stability and reliability of the multi-source power supply system, and realizes intelligent power switching and self-healing disaster recovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 Schematic diagram of the steps of the present invention;

[0068] Figure 2 Schematic diagram of the process of step (1);

[0069] Figure 3 Schematic diagram of the process of step (2);

[0070] Figure 4 Schematic diagram of the process of step (3);

[0071] Figure 5 Schematic diagram of the process of step (4);

[0072] Figure 6 is a schematic flow chart of step (5);

[0073] Figure 7 is a schematic flow chart of step (6); DETAILED DESCRIPTION

[0074] The embodiments of the present invention are further described below with reference to the accompanying drawings:

[0075] Example 1

[0076] like Figure 1As shown, the present invention provides a self-healing operation and maintenance method for power equipment status, comprising the following steps:

[0077] (1) Acquire the real-time operating status data of the power supply in the current multi-source power supply system. The operating status data includes voltage, current and power parameters. The data is collected at a preset sampling frequency to construct a power supply status matrix.

[0078] like Figure 2 As shown, the step (1) includes the following steps:

[0079] (1.1) Synchronously collecting operating data of each power circuit from the multi-source power supply system according to a preset sampling frequency, and obtaining a first set of real-time operating data through a data acquisition device;

[0080] Operation data includes voltage parameters, current parameters and power parameters between multiple power supply points;

[0081] (1.2) Based on the first set of real-time operation data, calculate the minimum sampling frequency value using the Shannon sampling theorem. If the actual sampling frequency is less than the minimum sampling frequency value, increase the sampling frequency and re-collect data until the sampling theorem meets the requirements to obtain a second set of real-time operation data;

[0082] (1.3) Merging the second set of real-time operation data sets, and using the Lagrange interpolation algorithm to supplement the missing data points to obtain the third set of real-time operation data sets;

[0083] (1.4) Establishing a power supply status matrix based on the third set of real-time operation data.

[0084] The third set of real-time operation data sets is used to establish a three-dimensional matrix structure, where the horizontal axis represents the sampling time series, the vertical axis represents the power point number, and the depth axis represents the voltage parameters, current parameters, and power parameters. The first matrix is ​​obtained through normalization processing;

[0085] Performing correlation analysis on the voltage parameters, current parameters, and power parameters in the first matrix, calculating the influence degree between the parameters by Pearson correlation coefficient, and optimizing the first matrix according to the influence degree value to obtain the power supply state matrix;

[0086] The voltage parameter threshold interval, current parameter threshold interval and power parameter threshold interval are set based on the power supply status matrix, the out-of-limit data are marked by the threshold comparator, and the marked data are corrected by the least squares method to obtain the corrected power supply status matrix.

[0087] In step (1.1), a first set of real-time operating data is acquired through a data acquisition device and the data is preliminarily sorted. During the acquisition process, a preset sampling frequency is set according to the system operating characteristics to ensure data integrity.

[0088] In step (1.4), a three-dimensional power supply status matrix is ​​constructed for the third set of real-time operation data, in which the horizontal axis is the sampling time series, the vertical axis is the power supply circuit number, and the depth axis includes voltage, current and power parameters. The parameter values ​​are mapped to the 0-1 interval through normalization processing, and the Pearson correlation coefficient is used to analyze the correlation between parameters and optimize the matrix structure.

[0089] The specific settings of the sampling frequency and threshold interval in this step (1) can be adjusted by technical personnel according to the actual scenario, and the collected power supply status matrix can truly reflect the operating status of the multi-source power supply system.

[0090] (2) Obtain the voltage and current data in the power supply status matrix, use fast Fourier transform to calculate the high-order harmonic distortion rate of the power supply, calculate the phase difference between the main power supply voltage and current, and analyze the change trend of the phase difference over time. At the same time, extract the time-frequency characteristics of the voltage and current signals, build a hidden fault symptom recognition model, and output the hidden fault symptom recognition results;

[0091] like Figure 3 As shown, step (2) includes the following steps:

[0092] (2.1) obtaining a voltage data sequence and a current data sequence according to the power supply state matrix, performing frequency domain decomposition by fast Fourier transform, and normalizing the frequency domain components obtained by the frequency domain decomposition to obtain a first harmonic distortion rate matrix;

[0093] Specifically, the first harmonic distortion rate matrix is ​​obtained by calculating the ratio of the higher harmonic content to the fundamental content;

[0094] (2.2) processing the sampling points using a zero phase detector for the fundamental wave components in the voltage data sequence and the current data sequence, and obtaining a first phase difference curve based on the phase increment values ​​between adjacent sampling points;

[0095] Specifically, the phase increment values ​​between adjacent sampling points are calculated, and the first phase difference curve is obtained according to the phase increment accumulation;

[0096] (2.3) Decomposing the first harmonic distortion rate matrix and the first phase difference curve using wavelet transform, and obtaining a first eigenvector group by extracting amplitude features, phase features, frequency features, and distortion features;

[0097] The extreme learning algorithm is used to perform nonlinear mapping on the first eigenvector group, and the second eigenvector group is obtained through activation function processing. The time-frequency feature matrix is ​​constructed according to the mapping rule.

[0098] (2.4) If the eigenvalues ​​obtained after mapping the first eigenvector group by the extreme learning algorithm exceed a preset threshold range, the eigenvalues ​​are marked as abnormal data points, and the abnormal data points are classified using a support vector machine to obtain a fault feature space and a fault probability value;

[0099] For the fault feature space, the temporal correlation of abnormal data points is extracted, the probability of fault occurrence is calculated, and the type of hidden fault and the probability value of occurrence are output according to the probability distribution law.

[0100] In step (2.2), after generating the first phase difference curve, its changing trend over time is analyzed to reflect the load characteristics. By analyzing the phase difference curve, the stability of the system operation state can be preliminarily determined;

[0101] In step (2.3), the wavelet transform can effectively decompose the time-frequency characteristics of the signal. The extreme learning algorithm uses the activation function to quickly map the feature vector by randomly initializing the input weights and biases.

[0102] In step (2.4), during the classification process of the support vector machine, the feature space is mapped to a high dimension through the kernel function, and the hyperplane is optimized based on the training samples to identify three typical fault characteristics: voltage sag type, phase imbalance type, and harmonic pollution type;

[0103] In step (2) of the present invention, the application of fast Fourier transform is intended to analyze the harmonic content of the signal from the frequency domain perspective, and the phase difference analysis reveals the impact of load dynamics on the system; the combination of wavelet transform and extreme learning algorithm can efficiently extract multidimensional features and achieve rapid mapping, and the classification of support vector machine further improves the accuracy of fault identification.

[0104] Through the above processing, the types of hidden faults and their occurrence probabilities can be effectively output, providing reliable technical support for the self-healing operation and maintenance of power equipment status.

[0105] (3) If the output recognition result is that there is a hidden fault sign, then obtain the harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence data of the main power supply within the preset time window, analyze and obtain the trend characteristics of the fault evolution, analyze the trend characteristics to obtain the fault evolution trend, and perform trend prediction on the load power change of the main power supply to obtain the load fluctuation curve within the preset time;

[0106] The harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence data obtained from the preset time window reflect the dynamic changes of the power supply operating status and provide a basis for fault evolution analysis.

[0107] like Figure 4 As shown, the step (3) includes the following steps:

[0108] (3.1) Obtain the harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence of the main power supply within a preset time window, perform preprocessing using a wavelet denoiser, and normalize the processed data using a data standardization method to obtain a first trend sequence;

[0109] (3.2) constructing a time domain feature matrix based on the first trend sequence, and extracting trend components from the time domain feature matrix using singular value decomposition to obtain a second trend sequence;

[0110] (3.3) extracting time series features from the second trend sequence using a long short-term memory neural network, establishing a feature vector group for the time series features, and calculating the change trend of each dimension of data within a preset time window to obtain a first evolution matrix;

[0111] (3.4) Performing time series decomposition on the first evolution matrix, constructing a correlation characteristic diagram based on the change trend of the harmonic distortion rate, the phase difference, and the load power, and generating a load fluctuation curve within a preset time by combining the correlation characteristic diagram.

[0112] The second evolution matrix is ​​obtained by calculating the correlation coefficient between features. Data clustering is performed based on the second evolution matrix. The probability of each type of fault development is calculated through the cluster center. The risk level threshold is set according to the probability distribution, and the trend change inflection point within the preset time window is marked as an early warning.

[0113] A recursive neural network is used to establish a prediction model for the load power data of the main power supply. The weight parameters are obtained through historical data training. The load power change trend is predicted within a preset time interval to obtain the load power fluctuation curve.

[0114] The predicted load power fluctuation curve is verified, the deviation between the predicted curve and the actual curve is calculated, the prediction parameters are corrected according to the deviation, and the corrected load power prediction curve is output.

[0115] In step (3.1), wavelet denoising filters out high-frequency noise through multi-resolution analysis. This preprocessing ensures the smoothness of the data;

[0116] In step (3.2), the singular value decomposition can separate the periodic and mutation components of the data;

[0117] In step (3.3), the long short-term memory neural network captures the temporal regularity within a 24-hour period by setting 96 time steps and a sampling point every 15 minutes;

[0118] In step (3.4), the correlation feature graph clearly shows the multidimensional characteristics of fault evolution by visualizing the relationship between these trends;

[0119] In step (3) of the present invention, the advantage of the multidimensional data analysis method is that it comprehensively considers the mutual influence between harmonic distortion rate, phase difference and load power; the correlation characteristic diagram reveals how load mutation causes harmonic pollution and phase imbalance, and the long short-term memory neural network captures the temporal dependence of these changes; the prediction function of the recursive neural network further provides a forward-looking basis for operation and maintenance decisions, enabling the system to respond to potential risks in advance.

[0120] (4) Based on the load fluctuation curve, combined with the historical switching failure data of the main power supply and the fault evolution trend, a power supply switching risk assessment model is established, and the main power supply failure risk assessment result is output. The main power supply failure risk assessment result is combined with the historical operation data of the backup power supply to calculate the target input probability of the backup power supply, and a weight distribution algorithm is used to determine the switching priority of each backup power supply;

[0121] like Figure 5 As shown, the step (4) includes the following steps:

[0122] (4.1) Obtain the load fluctuation curve of the main power supply, historical switching fault records, and evolution trend data, perform denoising through wavelet decomposition, extract three statistical features: fluctuation amplitude, fault frequency, and trend slope, and construct the first risk feature matrix;

[0123] (4.2) training a deep neural network model based on the first risk feature matrix and using a probability density estimation method to determine the risk level of the main power supply;

[0124] The training of the deep neural network model specifically includes performing data standardization on the first risk feature matrix, setting the number of nodes in the input layer of the deep neural network to the number of fluctuation features, the number of nodes in the hidden layer to twice the number of features, and the number of nodes in the output layer to the number of preset risk levels, and training to obtain the first probability matrix;

[0125] (4.3) extracting the startup time, load transfer time, and voltage stability parameters from the backup power supply operation database based on the risk level, and calculating a second probability matrix using a sliding time window;

[0126] (4.4) Calculating the weighted response probability of each backup power source using Bayesian probability according to the risk level and the second probability matrix to obtain a backup power source switching priority ranking result;

[0127] Bayesian probability is used to calculate the probability of each backup power source being deployed at different risk levels, and a backup power source response probability vector is constructed. For this backup power source response probability vector, the startup time weight coefficient, load response weight coefficient, and power supply stability weight coefficient are set, and the weighted response probability is calculated to obtain the priority ranking result.

[0128] The priority sorting results are verified, the actual success rate of each backup power supply is calculated through historical switching records, the weight coefficient is calibrated according to the success rate, and the final backup power supply switching priority is output.

[0129] In step (4.1), wavelet decomposition filters out noise and retains trend information through multi-scale analysis. The extracted features include fluctuation amplitude, failure frequency and trend slope, which provide data support for risk assessment;

[0130] In step (4.2), the risk level is obtained by calculating the risk probability threshold using the probability density estimation method for the risk probability distribution in the first probability matrix, and comparing the risk probability with the threshold;

[0131] In step (4) of the present invention, the load fluctuation curve reflects the dynamic changes in power demand, the historical switching failure records reveal the weak links in the system, and the fault evolution trend indicates the potential deterioration risk. The deep neural network can integrate these features through multi-layer nonlinear mapping to output accurate risk assessment results;

[0132] Moreover, the advantage of the multi-dimensional evaluation method is that it not only considers the real-time status and historical experience of the main power supply, but also incorporates the performance differences of the backup power supply.

[0133] (5) Pre-set the switching time constraint conditions, combine the main power supply failure risk assessment results, build a zero-delay switching decision tree, and output the power supply switching decision. If the main power supply failure risk assessment result is abnormal, switch the backup power supply according to the determined switching priority of each backup power supply and the power supply switching decision;

[0134] like Figure 6 As shown, the step (5) includes the following steps:

[0135] (5.1) Decompose the switching response time, standby startup time, and load transfer time constraints using a time axis splitter to obtain a time constraint sequence;

[0136] The constraints include the switching response time being less than a preset response threshold, the backup power supply startup time being less than a preset startup threshold, and the load transfer time being less than a preset transfer threshold;

[0137] (5.2) Based on the time constraint sequence, a random forest algorithm is used to construct a zero-delay handover decision tree, and judgment is made according to the three dimensions of risk level, time constraint, and standby status to obtain a decision sequence;

[0138] (5.3) Based on the decision sequence, a state detector is used to perform a pre-start detection on the backup power supply, obtain voltage parameters, current parameters, and power parameters, and obtain a detection result;

[0139] Specifically, the switching time in the decision sequence is segmented and statistics are performed, and the switching delay value in each time period is calculated. If the delay value exceeds a preset threshold range, the switching condition parameters in the time period are adjusted to obtain a second decision sequence;

[0140] According to the second decision sequence, a state detector is used to perform a pre-startup test on the backup power supply, obtain voltage parameters, current parameters, and power parameters of the backup power supply, and determine whether the startup conditions of the backup power supply are met to obtain a first test result;

[0141] A switching condition judgment matrix is ​​established based on the first detection result, and risk level thresholds, voltage parameter thresholds, current parameter thresholds, and power parameter thresholds are set. A comprehensive assessment of the status of the main power supply and backup power supply is performed to obtain a second detection result.

[0142] (5.4) If the risk level of the main power supply exceeds the preset threshold and the backup power supply status meets the switching conditions, the synchronous controller sends a switching instruction to the backup power supply;

[0143] Specifically, a synchronous controller is used to read the second detection result. If the risk level of the main power supply exceeds a preset threshold and the backup power supply status meets the switching conditions, a switching instruction is sent to the corresponding backup power supply in order of priority;

[0144] The backup power supply is started according to the switching instruction, and the load transfer process is monitored in real time through the load response detector. The voltage fluctuation value, current mutation value and power change value during the switching process are recorded to obtain the switching process data;

[0145] (5.5) During the switching process, the voltage amplitude and phase difference information between each power supply is obtained in real time, and the phase difference and voltage amplitude difference between each power supply are compared with the corresponding preset range. If the phase difference and voltage amplitude difference exceed the corresponding preset range, a control signal is generated based on the calculated adjustment amount through a dynamic adjustment algorithm to correct the synchronization characteristics of each power supply, which include frequency, voltage amplitude and phase.

[0146] The step (5.5) comprises the following steps:

[0147] (5.5.1) Collect voltage amplitude and phase angle sampling data from the power supply circuit, and process the sampling data through a digital filter to obtain voltage amplitude difference and phase difference;

[0148] The processing adopts denoising processing, and the voltage amplitude difference and phase difference between each power supply are calculated based on the processed data to obtain the first parameter group;

[0149] (5.5.2) Calculating the deviation excess ratio using a linear mapping function based on the voltage amplitude difference and phase difference, and setting a voltage amplitude deviation threshold and a phase difference deviation threshold based on the deviation excess ratio to obtain an adjustment target value;

[0150] Specifically, the voltage amplitude difference value in the first parameter group is compared with the preset voltage amplitude deviation range, the phase difference value is compared with the preset phase difference deviation range, and the deviation exceeding limit ratio is calculated using a linear mapping function to obtain the first deviation group;

[0151] For the first deviation group, the voltage amplitude deviation threshold, the phase difference deviation threshold, and the frequency deviation threshold are set respectively, and the adjustment target value is calculated according to the deviation value of each parameter to obtain the first adjustment target group;

[0152] (5.5.3) Use a neural network predictor to process the adjustment target value and optimize the prediction result through the gradient descent method to obtain the correction instruction;

[0153] Specifically, a neural network predictor is used to process the first adjustment target group. The input layer is set with three nodes: voltage deviation, phase deviation, and frequency deviation. The number of hidden layer nodes is set to twice that of the input layer. The output layer corresponds to the three adjustment parameters to obtain the second adjustment target group.

[0154] The second adjustment target group is optimized using a gradient descent method, and optimal adjustment parameters are obtained through iterative calculation. Amplitude correction instructions, phase correction instructions, and frequency correction instructions are generated according to the adjustment parameters to obtain a first control instruction group.

[0155] (5.5.4) Execute correction instructions through the synchronous controller to adjust the power supply circuit, and use closed-loop feedback to monitor the adjustment process to obtain correction data;

[0156] Specifically, the synchronization controller executes a first control instruction set to adjust the frequency, voltage amplitude, and phase parameters of each power supply in real time. Closed-loop feedback is used to monitor the adjustment process and generate a first correction data set. Based on the first correction data set, the corrected voltage amplitude and phase differences are calculated. The parameter changes before and after correction are compared, and the correction effect data is recorded to obtain synchronization characteristic evaluation results.

[0157] In step (5) of the present invention, the zero-delay switching decision tree ensures the accuracy of the switching timing through multi-dimensional judgment, and the real-time synchronous adjustment avoids the impact of parameter mutation on the load; compared with the traditional method, this collaborative adjustment method significantly improves the system stability while improving the switching smoothness.

[0158] (6) After the switching is completed, the power supply continuity indicators of the power equipment are monitored in real time, including voltage fluctuation, frequency fluctuation, and phase jump. If the power supply fluctuation time or phase jump amplitude exceeds the preset allowable value, the fast power compensation function of the energy storage device is used to smooth the power supply fluctuation.

[0159] like Figure 7As shown, step (6) includes the following steps:

[0160] (6.1) Real-time sampling of voltage fluctuation data, frequency fluctuation data, and phase jump data of the power supply circuit is performed using a data collector to obtain a first fluctuation sequence;

[0161] The first fluctuation sequence is obtained by calculating the fluctuation amplitude and duration of each parameter by setting a sampling time window;

[0162] (6.2) comparing the fluctuation amplitude of the first fluctuation sequence with a preset allowable value, marking the fluctuation parameters that exceed the preset allowable value to obtain a first abnormal sequence;

[0163] The preset allowable values ​​include the upper limit of voltage fluctuation amplitude, the upper limit of frequency fluctuation amplitude and the upper limit of phase jump amplitude;

[0164] (6.3) An adaptive neural network is used to train the first abnormal sequence, and a fuzzy control rule base is set according to the fluctuation feature classification result to calculate the compensation amount to obtain a first compensation sequence;

[0165] The input layer nodes of the adaptive neural network correspond to the fluctuation parameters, the number of hidden layer nodes is set to twice that of the input layer, and the output layer corresponds to the fluctuation judgment result. The fluctuation feature classification result is obtained through network training;

[0166] According to the classification results of the fluctuation characteristics, a fuzzy control rule base is set up, which includes voltage compensation rules, frequency compensation rules and phase compensation rules. The compensation amount of each parameter is calculated using fuzzy reasoning to obtain the first compensation sequence;

[0167] (6.4) executing the first compensation sequence through the power controller to adjust the output power waveform of the energy storage device and perform real-time compensation for voltage fluctuations, frequency fluctuations, and phase jumps to obtain a correction sequence;

[0168] Specifically, the energy storage device is parameterized to set a power output upper limit, a response time threshold, and a continuous compensation duration, and a power compensation instruction is generated according to the first compensation sequence to obtain a second compensation sequence;

[0169] The power controller executes a second compensation sequence to adjust the output power waveform of the energy storage device and compensates for voltage fluctuations, frequency fluctuations, and phase jumps in real time to obtain a first correction sequence;

[0170] The first correction sequence is processed by a waveform smoother, the deviation between the corrected fluctuation parameter and the preset allowable value is calculated, and the compensation effect data is recorded to obtain the monitoring and evaluation results.

[0171] (6.5) Based on the power supply system operation data after switching, update the power supply switching risk assessment model parameters, use the reinforcement learning algorithm to optimize the calculation strategy of the backup power supply input probability, establish a power supply continuity evaluation model, and evaluate the effect of the self-healing disaster recovery method on the improvement of power supply stability by analyzing the voltage sag time and frequency changes during the switching process, and output an operation performance report.

[0172] The step (6.5) comprises the following steps:

[0173] (6.5.1) Acquire power supply switching data from the power supply circuit, the power supply switching data including voltage sag data, frequency change data, and power supply switching timing data, and perform time segmentation on the power supply switching data using a timing sampler to obtain a first operation sequence;

[0174] (6.5.2) Establishing a deep reinforcement learning state space based on the first operation sequence, wherein the state space includes a power supply parameter state, a switching timing state, and a load response state, and mapping the state space using an action selector to obtain a first state map;

[0175] The action space includes three dimensions: input time selection, switching process control, and power supply recovery adjustment to obtain the first state mapping;

[0176] (6.5.3) performing time-frequency decomposition on the voltage sag data and the frequency variation data in the first state map using wavelet analysis to obtain a first characteristic matrix;

[0177] Specifically, three evaluation indicators (weights) are set based on the first state mapping: voltage recovery time, frequency stability, and load responsiveness. The reward value of each action is calculated through a scoring mechanism to obtain the first optimization sequence. Wavelet analysis is used to perform time-frequency decomposition on the voltage sag and frequency variation data in the first optimization sequence, extracting the three characteristic parameters of fluctuation duration, fluctuation amplitude, and recovery speed to obtain the first characteristic matrix.

[0178] (6.5.4) Constructing disaster recovery evaluation indicators based on the first characteristic matrix, wherein the disaster recovery evaluation indicators include power supply stability level, disaster recovery response time, and self-healing repair time. An evaluation matrix is ​​obtained by calculating the weights of each dimension through hierarchical analysis;

[0179] (6.5.5) Based on the evaluation matrix, generate an operational performance report based on the three modules of evaluation indicators, disaster recovery effect, and optimization suggestions;

[0180] Specifically, a comprehensive scoring is performed on the first evaluation matrix, and a scoring standard is established including three levels: basic power supply indicators, disaster recovery processing indicators, and self-healing effect indicators. The scores of each level are calculated to obtain a first scoring sequence;

[0181] The evaluation results are analyzed by the feedback updater, the switching risk assessment parameters are corrected according to the power supply stability improvement effect, and the backup power supply input strategy is updated to obtain the first update sequence;

[0182] A report generator is used to integrate various evaluation data, and an operation performance report is generated according to the three modules of evaluation indicators, disaster recovery effect, and optimization suggestions. Detailed evaluation data is recorded to obtain the final evaluation results.

[0183] In step (6.3), the adaptive neural network uses 15 input nodes corresponding to the fluctuation amplitude and duration, 30 hidden nodes to extract features, and 3 output nodes to classify them into persistent deviation, instantaneous mutation and periodic disturbance.

Claims

1. A self-healing operation and maintenance method for power equipment status, characterized in that: The following steps are involved: (1) Acquire the real-time operating status data of the power supply in the current multi-source power supply system. The operating status data includes voltage, current and power parameters. The data is collected at a preset sampling frequency to construct a power supply status matrix. (2) Obtain the voltage and current data in the power supply status matrix, use fast Fourier transform to calculate the high-order harmonic distortion rate of the power supply, calculate the phase difference between the main power supply voltage and current, and analyze the change trend of the phase difference over time. At the same time, extract the time-frequency characteristics of the voltage and current signals, build a hidden fault symptom recognition model, and output the hidden fault symptom recognition results; (3) If the output recognition result is that there is a hidden fault sign, then obtain the harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence data of the main power supply within the preset time window, analyze and obtain the trend characteristics of the fault evolution, analyze the trend characteristics to obtain the fault evolution trend, and perform trend prediction on the load power change of the main power supply to obtain the load fluctuation curve within the preset time; (4) Based on the load fluctuation curve, combined with the historical switching failure data of the main power supply and the fault evolution trend, a power supply switching risk assessment model is established, and the main power supply failure risk assessment result is output. The main power supply failure risk assessment result is combined with the historical operation data of the backup power supply to calculate the target input probability of the backup power supply, and a weight distribution algorithm is used to determine the switching priority of each backup power supply; (5) Pre-set the switching time constraint conditions, combine the main power supply failure risk assessment results, build a zero-delay switching decision tree, and output the power supply switching decision. If the main power supply failure risk assessment result is abnormal, switch the backup power supply according to the determined switching priority of each backup power supply and the power supply switching decision; (6) After the switching is completed, the power supply continuity indicators of the power equipment are monitored in real time, including voltage fluctuation, frequency fluctuation, and phase jump. If the power supply fluctuation time or phase jump amplitude exceeds the preset allowable value, the fast power compensation function of the energy storage device is used to smooth the power supply fluctuation.

2. A self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (1) comprises the following steps: (1.1) Synchronously collecting operating data of each power circuit from the multi-source power supply system according to a preset sampling frequency, and obtaining a first set of real-time operating data through a data acquisition device; (1.2) Based on the first set of real-time operation data, calculate the minimum sampling frequency value using the Shannon sampling theorem. If the actual sampling frequency is less than the minimum sampling frequency value, increase the sampling frequency and re-collect data until the sampling theorem meets the requirements to obtain a second set of real-time operation data; (1.3) For the second set of real-time operation data, a Lagrange interpolation algorithm is used to calculate the missing data points to obtain a third set of real-time operation data; (1.4) Establishing a power supply status matrix based on the third set of real-time operation data.

3. The self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (2) comprises the following steps: (2.1) obtaining a voltage data sequence and a current data sequence according to the power supply state matrix, performing frequency domain decomposition by fast Fourier transform, and normalizing the frequency domain components obtained by the frequency domain decomposition to obtain a first harmonic distortion rate matrix; (2.2) processing the sampling points using a zero phase detector for the fundamental wave components in the voltage data sequence and the current data sequence, and obtaining a first phase difference curve based on the phase increment values ​​between adjacent sampling points; (2.3) Decomposing the first harmonic distortion rate matrix and the first phase difference curve using wavelet transform, and obtaining a first eigenvector group by extracting amplitude features, phase features, frequency features, and distortion features; (2.4) If the eigenvalues ​​of the first eigenvector group obtained after mapping by the extreme learning algorithm exceed the preset threshold range, the eigenvalues ​​are marked as abnormal data points, and the abnormal data points are classified by a support vector machine to obtain a fault feature space and a fault probability value.

4. A self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (3) comprises the following steps: (3.1) Obtain the harmonic distortion rate sequence, voltage and current phase difference sequence, and load power sequence of the main power supply, and pre-process them through a wavelet denoiser to obtain the first trend sequence; (3.2) constructing a time domain feature matrix based on the first trend sequence, and extracting trend components from the time domain feature matrix using singular value decomposition to obtain a second trend sequence; (3.3) extracting time series features using a long short-term memory neural network for the second trend sequence, and establishing a feature vector group for the time series features to obtain an evolution matrix; (3.4) The evolution matrix is ​​decomposed in time series, and a correlation characteristic diagram is constructed according to the change trend of the harmonic distortion rate, the change trend of the phase difference, and the change trend of the load power. The load fluctuation curve within a preset time is generated by combining the correlation characteristic diagram.

5. The self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (4) comprises the following steps: (4.1) Obtain the load fluctuation curve, fault records, and evolution trend data of the main power supply, and perform denoising processing through wavelet decomposition to obtain the first risk feature matrix; (4.2) training a deep neural network model based on the first risk feature matrix and using a probability density estimation method to determine the risk level of the main power supply; (4.3) extracting the startup time, load transfer time, and voltage stability parameters from the backup power supply operation database based on the risk level, and calculating a second probability matrix using a sliding time window; (4.4) Based on the risk level and the second probability matrix, the weighted response probability of each backup power supply is calculated by Bayesian probability to obtain the backup power supply switching priority ranking result.

6. A self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (5) comprises the following steps: (5.1) Decompose the switching response time, standby startup time, and load transfer time constraints using a time axis splitter to obtain a time constraint sequence; (5.2) Based on the time constraint sequence, a random forest algorithm is used to construct a zero-delay handover decision tree, and judgment is made according to the three dimensions of risk level, time constraint, and standby status to obtain a decision sequence; (5.3) Based on the decision sequence, a state detector is used to perform a pre-start detection on the backup power supply, obtain voltage parameters, current parameters, and power parameters, and obtain a detection result; (5.4) If the risk level of the main power supply exceeds the preset threshold and the backup power supply status meets the switching conditions, the synchronous controller sends a switching instruction to the backup power supply; (5.5) During the switching process, the voltage amplitude and phase difference information between each power supply is obtained in real time, and the phase difference and voltage amplitude difference between each power supply are compared with the corresponding preset range. If the phase difference and voltage amplitude difference exceed the corresponding preset range, a control signal is generated based on the calculated adjustment amount through a dynamic adjustment algorithm to correct the synchronization characteristics of each power supply, which include frequency, voltage amplitude and phase.

7. The self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (6) comprises the following steps: (6.1) Real-time sampling of voltage fluctuation data, frequency fluctuation data, and phase jump data of the power supply circuit is performed using a data collector to obtain a first fluctuation sequence; (6.2) comparing the fluctuation amplitude of the first fluctuation sequence with a preset allowable value, marking the fluctuation parameters that exceed the preset allowable value to obtain a first abnormal sequence; (6.3) An adaptive neural network is used to train the first abnormal sequence, and a fuzzy control rule base is set according to the fluctuation feature classification result to calculate the compensation amount to obtain a first compensation sequence; (6.4) executing the first compensation sequence through the power controller to adjust the output power waveform of the energy storage device and perform real-time compensation for voltage fluctuations, frequency fluctuations, and phase jumps to obtain a correction sequence; (6.5) Based on the power supply system operation data after switching, update the power supply switching risk assessment model parameters, use the reinforcement learning algorithm to optimize the calculation strategy of the backup power supply input probability, establish a power supply continuity evaluation model, and evaluate the effect of the self-healing disaster recovery method on the improvement of power supply stability by analyzing the voltage sag time and frequency changes during the switching process, and output an operation performance report.

8. The self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (5.5) comprises the following steps: (5.5.1) Collect voltage amplitude and phase angle sampling data from the power supply circuit, and process the sampling data through a digital filter to obtain voltage amplitude difference and phase difference; (5.5.2) Calculating the deviation excess ratio using a linear mapping function based on the voltage amplitude difference and phase difference, and setting a voltage amplitude deviation threshold and a phase difference deviation threshold based on the deviation excess ratio to obtain an adjustment target value; (5.5.3) Use a neural network predictor to process the adjustment target value and optimize the prediction result through the gradient descent method to obtain the correction instruction; (5.5.4) The correction instructions are executed by the synchronous controller to adjust the power supply circuit, and the adjustment process is monitored using closed-loop feedback to obtain correction data.

9. The self-healing operation and maintenance method for power equipment status according to claim 1, characterized in that: The step (6.5) comprises the following steps: (6.5.1) Acquire power supply switching data from the power supply circuit, the power supply switching data including voltage sag data, frequency change data, and power supply switching timing data, and perform time segmentation on the power supply switching data using a timing sampler to obtain a first operation sequence; (6.5.2) Establishing a deep reinforcement learning state space based on the first operation sequence, wherein the state space includes power supply parameter states, switching timing states, and load response states, and mapping the state space using an action selector to obtain a first state map; (6.5.3) performing time-frequency decomposition on the voltage sag data and the frequency variation data in the first state map using wavelet analysis to obtain a first characteristic matrix; (6.5.4) Constructing disaster recovery evaluation indicators based on the first characteristic matrix, wherein the disaster recovery evaluation indicators include power supply stability level, disaster recovery response time, and self-healing repair time. An evaluation matrix is ​​obtained by calculating the weights of each dimension through hierarchical analysis; (6.5.5) Based on the evaluation matrix, generate an operational performance report based on the three modules of evaluation indicators, disaster recovery effect, and optimization suggestions.

Images