A Fault Handling Method and System for Distributed Power Source

Abstract

The present invention discloses a fault processing method and system for distributed power sources, which detects the fault mode by real-time monitoring of the operating data of the distributed power source under extreme operating conditions and combining it with historical fault data; extracts the corresponding fault operating state characteristics from the historical monitoring data to evaluate the severity according to the current fault mode of the distributed power source, and extracts the current countermeasures from a fault countermeasure library according to the evaluation results to control the execution of the distributed power source; collects the second real-time monitoring parameters when the distributed power source executes the countermeasures to evaluate the countermeasure effect, and when the effect does not meet the expected standard, selects and executes a better target countermeasure from the fault countermeasure library based on the recovery rate of the distributed power source under the current countermeasure, which can significantly improve the reliability and safety of the distributed power source under extreme operating conditions, and greatly improves the fault response speed and processing efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a fault processing method and system for distributed power supply. Background Art

[0002] At present, distributed power sources are facing more and more challenges as their operating time increases, especially when encountering changes in external conditions such as extreme weather or sudden load changes. They are often required to work under extreme operating conditions, but this behavior will cause more failures of distributed power sources.

[0003] In the prior art, it is usually adopted to directly monitor the voltage and current signals of distributed power sources under extreme operating conditions, and use this as a basis to timely adjust the working state of the distributed power sources to cope with the occurrence of faults. However, this method is difficult to accurately distinguish between normal fluctuations of the distributed power system and fluctuations caused by faults, and is prone to false alarms or missed alarms, resulting in increasingly low accuracy in fault handling.

[0004] It can be seen that how to solve the current problem of low accuracy in fault handling of distributed power sources has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] The present invention provides a fault processing method and system for a distributed power source, which solves the problem of how to improve the accuracy of fault processing for the distributed power source.

[0006] In order to solve the above technical problems, the first aspect of the present invention provides a fault handling method for a distributed power source, comprising:

[0007] Acquire a first real-time monitoring parameter of a distributed power source under an extreme operating state, and extract a frequency domain feature and a time domain feature of the first real-time monitoring parameter, so as to analyze a correlation between the first real-time monitoring parameter and historical fault data of the distributed power source based on the frequency domain feature and the time domain feature;

[0008] The fault mode of the distributed power source is judged based on the correlation and a pre-built fault mode classification library to obtain a current fault mode, and corresponding fault operation state characteristics are extracted from historical monitoring data of the selected distributed power source according to the current fault mode;

[0009] Inputting the fault operation state characteristics into a pre-built limit state severity assessment model for processing to obtain a severity assessment result corresponding to the current fault mode, so as to extract a current countermeasure from a pre-built fault countermeasure library according to the severity assessment result;

[0010] Collecting a second real-time monitoring parameter of the distributed power source when executing the current countermeasure, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the countermeasure effect of the current countermeasure according to the parameter change trend curve;

[0011] When the response effect does not meet the expected standard, the current response measure is modeled using a gradient boosting tree to obtain a response parameter feature matrix to quantify the recovery rate of the distributed power source under the current response measure, and a target response measure is selected from the fault response measure library based on the recovery rate to control the execution of the distributed power source.

[0012] As one of the preferred solutions, the step of obtaining a first real-time monitoring parameter of a distributed power source under an extreme operation state includes:

[0013] Acquire real-time operating status data of the distributed power source, and determine an operating status ratio parameter of the distributed power source according to the real-time operating status data, so as to normalize the operating status ratio parameter through a convolutional neural network model to obtain a normalized ratio parameter sequence;

[0014] quantifying the standard deviation of the normalized ratio parameter sequence within the time window to determine an abnormality benchmark, and quantifying an abnormality degree score of the normalized ratio parameter sequence according to the abnormality benchmark;

[0015] Quantifying the change rate of the normalized ratio parameter sequence through a sliding time window to obtain a parameter change trend value, and inputting it into a support vector machine with the abnormality degree score to extract features of the real-time operating status data to obtain a feature space clustering result;

[0016] When the center point of the feature space clustering result reaches a preset center threshold standard and the duration of the center point exceeds a preset duration, it is determined that the distributed power source is in the limit operating state, and the operating state ratio parameter is output as the first real-time monitoring parameter.

[0017] As one preferred solution, determining the operating status ratio parameter of the distributed power source according to the real-time operating status data includes:

[0018] Determine the initial operating state ratio parameter of the distributed power source according to the relationship between the real-time operating state data and the corresponding operating state standard data, and standardize the initial operating state ratio parameter through a convolutional neural network model to obtain a standardized initial operating state ratio parameter;

[0019] A filter is used to perform primary correction and smoothing processing on the standardized initial operating state ratio parameter to obtain a smoothed initial operating state ratio parameter, and a jump point in the smoothed initial operating state ratio parameter is secondary corrected to obtain a corrected initial operating state ratio parameter;

[0020] The covariance matrix between the modified initial operating state ratio parameters is quantified to perform correlation calculation, and the modified initial operating state ratio parameters are data fitted based on the correlation calculation results to obtain a ratio parameter fitting curve for linear interpolation completion to obtain the operating state ratio parameters of the distributed power source.

[0021] As one of the preferred solutions, the frequency domain features include the dominant frequency amplitude and spectrum entropy, and the time domain features include the time domain waveform distortion, fluctuation cycle length, rising edge duration and falling edge duration; wherein,

[0022] The extracting the frequency domain features and the time domain features of the first real-time monitoring parameter includes:

[0023] Performing frequency domain conversion on the first real-time monitoring parameter by fast Fourier transform, so as to quantize a power spectrum density curve according to the frequency domain conversion result, and determining the dominant frequency amplitude of the first real-time monitoring parameter according to the power spectrum density curve;

[0024] quantifying the spectrum entropy of the first real-time monitoring parameter according to the energy proportion of different frequency bands in the power spectrum density curve, and performing bandwidth limitation on the first real-time monitoring parameter through a third-order Bessel bandpass filter to obtain a waveform envelope curve and its corresponding time domain waveform distortion;

[0025] quantifying the length of the fluctuation cycle based on the time difference between the local maximum points of the waveform in the waveform envelope curve, and marking the inflection points of the waveform envelope curve using a long short-term memory neural network model to quantify the duration of the rising edge and the duration of the falling edge according to the marked inflection points;

[0026] The rising edge duration and the falling edge duration are decomposed by wavelet transform to obtain wavelet coefficient feature vectors, and the dominant frequency amplitude, spectrum entropy, time domain waveform distortion, fluctuation period length, rising edge duration and falling edge duration are spliced ​​into a multidimensional feature space through the wavelet coefficient feature vectors, and the multidimensional feature space is reduced in principal components to obtain the frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter.

[0027] As one of the preferred solutions, the analyzing the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain characteristics and the time domain characteristics includes:

[0028] Processing the frequency domain features and time domain features of the first real-time monitoring parameter through a convolutional neural network model to obtain a multidimensional feature matrix and segmenting it through a sliding window, quantifying the Pearson correlation coefficient between each segment of feature data and the historical fault data and performing normalization processing to obtain a correlation score;

[0029] Taking the correlation score as a reference value, quantifying the Euclidean distance between the multidimensional feature matrix and the reference value, superimposing the distance values ​​in adjacent time windows and comparing them with a preset distance threshold, and assigning a weight to each segment of feature data based on the comparison result to obtain weighted feature data;

[0030] A five-layer structure autoencoder is used to fuse the weight feature data to obtain a fused feature vector to quantify feature similarity. When the feature similarity and correlation score of the fused feature vector simultaneously reach corresponding thresholds and the duration reaches a preset duration threshold, it is determined that a fault exists in the distributed power supply to realize the correlation analysis between the first real-time monitoring parameter and the historical fault data.

[0031] As one of the preferred solutions, the fault mode of the distributed power source is judged based on the correlation and the pre-built fault mode classification library to obtain the current fault mode, and the corresponding fault operation state characteristics are extracted from the selected historical monitoring data of the distributed power source according to the current fault mode, including:

[0032] When it is determined that the distributed power supply has a fault according to the correlation, the frequency domain features and the time domain features are processed by a four-layer convolutional neural network model to obtain a multi-dimensional fault feature vector;

[0033] quantifying the matching degree between the multidimensional fault feature vector and various types of fault mode feature templates in the fault mode classification library based on cosine similarity, so as to determine the current fault mode of the distributed power source according to the matching degree;

[0034] Determine historical monitoring target data from the selected historical monitoring data of the distributed power source according to the current fault mode, and process the historical monitoring target data through 4-layer wavelet decomposition to determine the fault occurrence time and the fault recovery time;

[0035] Performing mean clustering processing on the historical monitoring period data at a preset time before the fault occurs, and taking the cluster center with the smallest distance from the historical monitoring period data as the stable operation feature before the fault;

[0036] The historical monitoring target data, the historical monitoring data corresponding to the fault recovery time, and the stable operation characteristics before the fault are combined by a support vector machine to obtain a complete fault feature vector as the fault operation state feature output.

[0037] As one of the preferred solutions, the inputting of the fault operation state characteristics into a pre-built limit state severity assessment model for processing to obtain a severity assessment result corresponding to the current fault mode, and extracting the current countermeasures from a pre-built fault countermeasure library according to the severity assessment result, includes:

[0038] The fault operation state characteristics are processed using the limit state severity assessment model, and the processing result is weighted according to the weight coefficient corresponding to the first real-time monitoring parameter to obtain a state assessment score as a severity assessment result corresponding to the current fault mode;

[0039] The severity level of the distributed power source under the extreme operating state is determined according to the state assessment score to trigger a corresponding target warning signal, and based on the target warning signal, a corresponding response parameter combination is extracted from the fault response measure library as the current response measure.

[0040] As one preferred solution, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the response effect of the current response measure according to the parameter change trend curve, includes:

[0041] The second real-time monitoring parameter is processed using a three-layer convolutional neural network model to obtain a parameter change trend feature, and the mean, standard deviation, first-order difference value and second-order difference value of the parameter change trend feature in a sliding window are quantified to construct a multidimensional parameter change trend feature vector;

[0042] Processing the multidimensional parameter change trend characteristic vector according to wavelet decomposition to obtain high-frequency fluctuation characteristics and low-frequency drift characteristics, quantifying the recovery rates of the high-frequency fluctuation characteristics and the low-frequency drift characteristics based on wavelet changes, and obtaining the parameter change trend curve;

[0043] Predicting the parameter change trend curve through a Kalman filter, and calculating the deviation value between the parameter change trend curve and the normal range boundary based on the prediction result, so as to construct a response effect evaluation index according to the deviation value;

[0044] The response effect evaluation index is weightedly fused. If the processed response effect evaluation index is within the rated standard range and the standard deviation of the second real-time monitoring parameter is less than the preset standard threshold, it is determined that the operating status of the distributed power supply has returned to normal, and it is determined that the response effect of the current response measure has reached the expected standard.

[0045] As one of the preferred solutions, when the response effect does not meet the expected standard, the current response measures are modeled by using a gradient boosting tree to obtain a response parameter feature matrix to quantify the recovery rate of the distributed power source under the current response measures, and select a target response measure from the fault response measure library according to the recovery rate to control the execution of the distributed power source, including:

[0046] When the response effect does not meet the expected standard, a gradient boosting tree is used to model the current response measures and the second real-time monitoring parameters to obtain a response parameter feature matrix, and the recovery rate and parameter recovery amount of the distributed power source in unit time are counted according to the response parameter feature matrix;

[0047] Fitting the parameter recovery amount by a fourth-order polynomial to obtain a parameter recovery fitting curve, and extracting a parameter recovery index from the parameter recovery fitting curve according to the recovery rate and using it as a response effect evaluation index;

[0048] A three-layer neural network model is used to process the response effect evaluation index to obtain a comprehensive recovery effect score, and a weighted calculation is performed on the comprehensive recovery effect score to construct a response measure effect evaluation criterion based on the weighted calculation result;

[0049] According to the countermeasure effect evaluation criteria, from the fault countermeasure library, the fault countermeasure whose recovery rate reaches a preset rate threshold and whose corresponding fluctuation standard deviation is the smallest when the distributed power source is executed is taken as the target countermeasure to control the execution of the distributed power source.

[0050] A second aspect of the present invention provides a fault handling system for a distributed power source, comprising:

[0051] A correlation calculation module, used to obtain a first real-time monitoring parameter of a distributed power source under an extreme operating state, and extract frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter, so as to analyze the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain characteristics and the time domain characteristics;

[0052] A state feature extraction module, used to judge the fault mode of the distributed power source based on the correlation and a pre-built fault mode classification library, obtain the current fault mode, and extract the corresponding fault operation state feature from the selected historical monitoring data of the distributed power source according to the current fault mode;

[0053] A countermeasure execution module is used to input the fault operation state characteristics into a pre-built limit state severity assessment model for processing, obtain a severity assessment result corresponding to the current fault mode, and extract the current countermeasure from a pre-built fault countermeasure library according to the severity assessment result;

[0054] A response effect evaluation module, used for collecting a second real-time monitoring parameter of the distributed power source when executing the current response measure, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the response effect of the current response measure according to the parameter change trend curve;

[0055] The target measure execution module is used to use a gradient boosting tree to model the current countermeasure when the countermeasure effect does not meet the expected standard, obtain a countermeasure parameter feature matrix to quantify the recovery rate of the distributed power supply under the current countermeasure, and select a target countermeasure from the fault countermeasure library according to the recovery rate to control the execution of the distributed power supply.

[0056] Compared with the prior art, the embodiments of the present invention have the following advantages:

[0057] (1) By real-time monitoring of the operating parameters of distributed power sources and extracting their frequency domain and time domain characteristics, the fault mode can be quickly identified, greatly shortening the time for fault discovery. By using the historical fault data of distributed power sources and conducting correlation analysis with the real-time monitoring data, the fault type and cause can be more accurately located. Through the pre-built limit state severity assessment model, the severity of the current fault mode can be quantitatively assessed, providing a basis for the subsequent selection of response measures.

[0058] (2) Dynamically adjust the response measures according to the severity of the fault and the evaluation results of the response effect to ensure the effectiveness and pertinence of the response measures; the entire fault handling process is highly automated, and no excessive human intervention is required from fault identification to the selection and execution of response measures, thereby improving the response speed and processing efficiency, and can significantly improve the reliability and safety of distributed power sources under extreme operating conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the implementation mode will be briefly introduced below. Obviously, the drawings described below are only some implementation modes of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1is a flow chart of a fault handling method for a distributed power source provided by a certain embodiment of the present invention;

[0061] Figure 2 It is a structural diagram of a fault handling system for distributed power sources provided by a certain embodiment of the present invention. DETAILED DESCRIPTION

[0062] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings and embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0063] In the description of this application, the terms "first", "second", "third", etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the feature. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0064] In the description of the present application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two components. The terms "vertical", "horizontal", "left", "right", "upper", "lower" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the system or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0065] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For those of ordinary skill in the art, the specific meanings of the above terms in this application can be understood by specific circumstances.

[0066] In one embodiment, if Figure 1 As shown, the first aspect of the present invention provides a fault handling method for a distributed power source, comprising:

[0067] S1. Obtain a first real-time monitoring parameter of a distributed power source under an extreme operating state, and extract a frequency domain feature and a time domain feature of the first real-time monitoring parameter, so as to analyze a correlation between the first real-time monitoring parameter and historical fault data of the distributed power source based on the frequency domain feature and the time domain feature;

[0068] In one embodiment, obtaining a first real-time monitoring parameter of a distributed power source under a limit operation state includes:

[0069] Acquire real-time operating status data of the distributed power source, and determine an operating status ratio parameter of the distributed power source according to the real-time operating status data, so as to normalize the operating status ratio parameter through a convolutional neural network model to obtain a normalized ratio parameter sequence;

[0070] quantifying the standard deviation of the normalized ratio parameter sequence within the time window to determine an abnormality benchmark, and quantifying an abnormality degree score of the normalized ratio parameter sequence according to the abnormality benchmark;

[0071] Quantifying the change rate of the normalized ratio parameter sequence through a sliding time window to obtain a parameter change trend value, and inputting it into a support vector machine with the abnormality degree score to extract features of the real-time operating status data to obtain a feature space clustering result;

[0072] When the center point of the feature space clustering result reaches a preset center threshold standard and the duration of the center point exceeds a preset duration, it is determined that the distributed power source is in the limit operating state, and the operating state ratio parameter is output as the first real-time monitoring parameter.

[0073] Specifically, the present invention uses a distributed power supply monitoring device to obtain the operating status data at the monitoring point in real time, including voltage data, current data, temperature data and load data. The data collection time interval is set to 10 seconds, and the monitoring data accuracy is set to 3 decimal places. The operating status ratio parameters of the distributed power supply, namely the voltage ratio parameter, the load and power generation capacity ratio parameter and the temperature and rated temperature ratio parameter, are determined, and then these ratio parameters are normalized using a convolutional neural network model. The ratio data sequence within a 30-minute time window is selected as input, and its time series features are extracted through a three-layer convolution structure, and a normalized ratio parameter sequence is output; in the normalization process, selecting a 30-minute time window is equivalent to collecting 180 data points, and the amount of data is sufficient to reflect the operating characteristics of the equipment, and the normalization process eliminates the dimensional differences of different parameters, and unifies the load ratio, voltage ratio, and temperature ratio to the 0-1 interval.

[0074] Then, based on the normalized data, the mean and standard deviation of each ratio parameter in the time window are calculated, and the standard deviation threshold is set to three times the standard deviation as the abnormality judgment benchmark, so as to calculate the proportion of data points with normalized load ratio parameters exceeding 0.9, the proportion of data points with voltage ratio parameters exceeding 1.1, and the proportion of data points with temperature ratio parameters exceeding 0.95 in four time windows, and set weight coefficients for each ratio parameter, with the load ratio weight being 0.4, the voltage ratio weight being 0.3, and the temperature ratio weight being 0.3, so as to calculate the abnormality degree score of the weighted normalized ratio parameter sequence. Taking photovoltaic power generation equipment as an example, under standard operating conditions, the load ratio fluctuates in the range of 0.6-0.8, the voltage ratio fluctuates in the range of 0.95-1.05, and the temperature ratio fluctuates in the range of 0.7-0.9. Data points exceeding three times the standard deviation are marked as outliers. The parameter abnormality score reflects the comprehensive performance of the equipment's operating status. The data points with a load ratio exceeding 0.9 in the four time windows account for 25%, indicating that the equipment is in a high-load operating state, and the data points with a voltage ratio exceeding 1.1 account for 15%, indicating that the power supply voltage is too high. The data points with a temperature ratio exceeding 0.95 account for 20%, reflecting that the equipment temperature is close to the limit. The weight setting takes into account the degree of influence of each ratio parameter on the safe operation of the equipment. The highest weight of the load ratio reflects the dominant role of the load level in the equipment's operating status.

[0075] Finally, the change rates of the three ratio parameters are calculated according to the sliding time window to reflect the dynamic change characteristics of the equipment operation status. The window length is set to 5 minutes. The deviation value of the load ratio change rate and the threshold value of 0.05, the deviation value of the voltage ratio change rate and the threshold value of 0.03, and the deviation value of the temperature ratio change rate and the threshold value of 0.02 in the window are calculated. The parameter change trend is evaluated based on the deviation value. The support vector machine is used to extract the features of the calculated abnormal degree score and the parameter change trend evaluation value. A classifier is constructed based on the radial basis kernel function, and the samples are mapped to the feature space for clustering. The distance from the center point of the clustering result to the preset threshold value of 0.8 is less than the preset distance threshold, indicating that it meets the preset center threshold standard. When its duration exceeds 15 minutes, it is determined that the distributed power supply enters the limit operation state, and the operation state ratio parameter is output as the first real-time monitoring parameter.

[0076] The present invention utilizes a convolutional neural network model to normalize the operating status ratio parameters, which can eliminate the dimensional differences between different parameters and improve the accuracy and consistency of data processing; by quantifying the standard deviation of the normalized ratio parameter sequence within a time window, the abnormal benchmark is determined to accurately identify abnormal fluctuations in the parameter sequence, providing a scientific basis for anomaly detection; the whole process from data acquisition, processing, analysis to judgment output is automated, reducing manual intervention and improving work efficiency; and the introduction of machine learning algorithms such as convolutional neural networks and support vector machines makes the solution have a certain level of intelligence, can adaptively process and analyze data, and improve the flexibility and scalability of the system.

[0077] In one embodiment, determining the operating status ratio parameter of the distributed power source according to the real-time operating status data includes:

[0078] Determine the initial operating state ratio parameter of the distributed power source according to the relationship between the real-time operating state data and the corresponding operating state standard data, and standardize the initial operating state ratio parameter through a convolutional neural network model to obtain a standardized initial operating state ratio parameter;

[0079] A filter is used to perform primary correction and smoothing processing on the standardized initial operating state ratio parameter to obtain a smoothed initial operating state ratio parameter, and a jump point in the smoothed initial operating state ratio parameter is secondary corrected to obtain a corrected initial operating state ratio parameter;

[0080] The covariance matrix between the modified initial operating state ratio parameters is quantified to perform correlation calculation, and the modified initial operating state ratio parameters are data fitted based on the correlation calculation results to obtain a ratio parameter fitting curve for linear interpolation completion to obtain the operating state ratio parameters of the distributed power source.

[0081] Specifically, the present invention obtains a voltage ratio parameter by dividing the voltage value 220 volts specified in the standard power supply specification by the voltage data of the monitoring point, obtains a load and power generation capacity ratio parameter by dividing the load data of the monitoring point by the rated power generation capacity data indicated on the nameplate of the corresponding distributed power source, and obtains a temperature and rated temperature ratio parameter by dividing the temperature data of the monitoring point by the rated temperature of 85 degrees Celsius on the equipment nameplate, which can also be referred to as load ratio, voltage ratio, and temperature ratio, and uses these initial operating state ratio parameters as input, and performs data normalization processing on them through a convolutional neural network model to eliminate the dimensional differences of different parameters.

[0082] The standardized initial operating state ratio parameter obeys the normal distribution, with a range of plus or minus 3 times the standard deviation. Those exceeding the range are judged as outliers. The median filter with a window length of 5 is used to perform the initial correction on the abnormal data points, and then a Gaussian filter with a window length of 3 is used to smooth the data after the initial correction; the difference of the smoothed data between adjacent sampling points is calculated, and the sampling points with a difference exceeding 0.1 are judged as jump points. The jump point data are corrected by the arithmetic mean of the ratio data of the previous and next sampling points after smoothing, and the corrected initial operating state ratio parameters are obtained.

[0083] The naive Bayes classifier is used to calculate the covariance matrix between the modified initial operating state ratio parameters, and the correlation coefficient is calculated according to the covariance matrix. The correlation coefficient with an absolute value exceeding 0.8 is judged as a strong correlation. For the strongly correlated data, a third-order polynomial function is used to fit the data, and the missing data points in the fitting curve are linearly interpolated to complete the missing data points, so as to obtain the operating state ratio parameters of the distributed power supply.

[0084] The present invention provides the possibility of taking timely countermeasures by quickly responding to changes in power supply operation, thereby ensuring the stability and safety of the power supply system; uses a convolutional neural network model for normalization processing, improves the accuracy of data processing, and provides a reliable basis for subsequent analysis; determines the abnormality benchmark by quantifying the standard deviation of the normalized ratio parameter sequence in the time window, and then determines the abnormality degree score, which can accurately identify abnormal fluctuations in the parameter sequence, providing a scientific basis for abnormality detection. At the same time, the severity of the abnormality is further quantified according to the abnormal benchmark quantification, providing an important reference for subsequent abnormality processing; effective trend analysis and prediction help to discover potential problems in advance and take preventive measures; through in-depth feature extraction and clustering analysis, the operating status of the power supply can be better understood, and corresponding management strategies can be formulated; the central point of the feature space clustering result and its duration judgment standard are adopted, so that the scheme can accurately identify whether the distributed power supply is in an extreme operating state, improves the accuracy and reliability of the judgment, and helps to take measures in time to avoid the power supply system from shutting down due to overload or failure.

[0085] In one embodiment, the frequency domain features include the dominant frequency amplitude and spectrum entropy, and the time domain features include the time domain waveform distortion, fluctuation cycle length, rising edge duration and falling edge duration; wherein,

[0086] The extracting the frequency domain features and the time domain features of the first real-time monitoring parameter includes:

[0087] Performing frequency domain conversion on the first real-time monitoring parameter by fast Fourier transform, so as to quantize a power spectrum density curve according to the frequency domain conversion result, and determining the dominant frequency amplitude of the first real-time monitoring parameter according to the power spectrum density curve;

[0088] quantifying the spectrum entropy of the first real-time monitoring parameter according to the energy proportion of different frequency bands in the power spectrum density curve, and performing bandwidth limitation on the first real-time monitoring parameter through a third-order Bessel bandpass filter to obtain a waveform envelope curve and its corresponding time domain waveform distortion;

[0089] quantifying the length of the fluctuation cycle based on the time difference between the local maximum points of the waveform in the waveform envelope curve, and marking the inflection points of the waveform envelope curve using a long short-term memory neural network model to quantify the duration of the rising edge and the duration of the falling edge according to the marked inflection points;

[0090] The rising edge duration and the falling edge duration are decomposed by wavelet transform to obtain wavelet coefficient feature vectors, and the dominant frequency amplitude, spectrum entropy, time domain waveform distortion, fluctuation period length, rising edge duration and falling edge duration are spliced ​​into a multidimensional feature space through the wavelet coefficient feature vectors, and the multidimensional feature space is reduced in principal components to obtain the frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter.

[0091] Specifically, the present invention uses fast Fourier transform to perform frequency domain conversion on the first real-time monitoring parameter, that is, the operating status ratio parameter, and sets the sampling frequency to 100 Hz. The time window length is 1000 sampling points, which is sufficient to capture the periodic changes in the operating status of the distributed power supply, and calculates the power spectrum density curve based on the Fourier transform coefficient, determines the dominant frequency amplitude of the spectrum distribution in the range of 0 to 50 Hz, calculates the spectrum entropy value according to the energy proportion of different frequency bands in the spectrum distribution curve, and multiplies the energy proportion of each frequency band by the logarithmic value corresponding to the frequency band to obtain the normalized spectrum entropy.

[0092] For the normalized time domain series such as load ratio, voltage ratio, and temperature ratio, bandwidth limitation is performed through a third-order Bessel bandpass filter to filter out frequency components below 0.1 Hz and above 45 Hz. The waveform envelope curve is extracted from the preprocessed time domain sequence, and the root mean square error between the actual waveform and the ideal sinusoidal waveform is defined as the waveform distortion.

[0093] Then, the length of the fluctuation cycle is calculated based on the time difference between the local maximum points of the waveform, and a recursive neural network model based on long short-term memory units is used to mark the waveform inflection points in the voltage ratio time domain series. The input layer scale is set to 48 neurons, and the hidden layer adopts a double-layer structure with 64 neurons in each layer. The network outputs the starting and ending points of the rising segment and the falling segment of the waveform, and the rising edge duration and the falling edge duration are calculated based on the marked points.

[0094] The first real-time monitoring parameter sequence is segmented according to 50 sampling points in a fixed time window. The segmented parameter sequence is decomposed into four layers through wavelet transform. Discrete wavelet basis functions are used for transformation to obtain wavelet coefficients at four scales. The mean and standard deviation of the coefficient sequence at each scale are calculated as feature quantities. The dominant frequency amplitude, spectral entropy, time domain waveform distortion, fluctuation period length, rising edge duration and falling edge duration are concatenated to form a multidimensional feature space. The most representative feature components are extracted based on principal component dimensionality reduction as the frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter.

[0095] The present invention effectively reveals the spectral characteristics and spectral complexity of the signal by performing frequency domain conversion on the first real-time monitoring parameter through fast Fourier transform (FFT), and further reveals the envelope characteristics and time domain distortion of the signal through waveform envelope analysis of the data signal; splices the time domain characteristics and frequency domain characteristics into a multi-dimensional feature space through wavelet transform, and effectively reduces the dimension of the feature space through principal component dimensionality reduction (PCA), while retaining key information and improving the efficiency and accuracy of feature extraction; the scheme combines frequency domain and time domain feature extraction methods at the same time, which can comprehensively reflect the dynamic change characteristics of the first real-time monitoring parameter, and provides strong support for subsequent fault diagnosis, status monitoring and performance evaluation.

[0096] In one embodiment, analyzing the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain feature and the time domain feature includes:

[0097] Processing the frequency domain features and time domain features of the first real-time monitoring parameter through a convolutional neural network model to obtain a multidimensional feature matrix and segmenting it through a sliding window, quantifying the Pearson correlation coefficient between each segment of feature data and the historical fault data and performing normalization processing to obtain a correlation score;

[0098] Taking the correlation score as a reference value, quantifying the Euclidean distance between the multidimensional feature matrix and the reference value, superimposing the distance values ​​in adjacent time windows and comparing them with a preset distance threshold, and assigning a weight to each segment of feature data based on the comparison result to obtain weighted feature data;

[0099] A five-layer structure autoencoder is used to fuse the weight feature data to obtain a fused feature vector to quantify feature similarity. When the feature similarity and correlation score of the fused feature vector simultaneously reach corresponding thresholds and the duration reaches a preset duration threshold, it is determined that a fault exists in the distributed power supply to realize the correlation analysis between the first real-time monitoring parameter and the historical fault data.

[0100] Specifically, the present invention uses a neural network model with a three-layer convolution structure to process the frequency domain features and time domain features of the ratio parameters. The convolution kernel size of the model is set to 3, and the number of convolution layer channels is 16, 32, and 64, respectively, to construct a multidimensional feature matrix with a dimension of 300×6, and the multidimensional feature matrix is ​​segmented by a sliding window length of 500 sampling points and a step size of 100 sampling points to calculate the Pearson correlation coefficient between each segment of feature data and the historical fault data, and the correlation coefficient is normalized to obtain a correlation score.

[0101] Taking the correlation score as the benchmark value, the Euclidean distance between the multidimensional feature matrix and the benchmark value is calculated, the distance values ​​are superimposed in four adjacent time windows and compared with the preset distance threshold of 0.8, and the feature data are weighted based on the comparison results, where the load ratio frequency domain feature weight is 0.3 and the time domain feature weight is 0.2, the voltage ratio frequency domain feature weight is 0.2 and the time domain feature weight is 0.1, and the temperature ratio frequency domain feature weight is 0.1 and the time domain feature weight is 0.1, and the weighted feature data is obtained.

[0102] Finally, the weight feature data is fused through a five-layer autoencoder, in which the number of nodes in the encoding layer is 256, 128, and 64, respectively, and the number of nodes in the decoding layer is 128 and 256, respectively. The input feature dimension is 300, the intermediate hidden layer dimension is 64, the learning rate is set to 0.01, and the number of training rounds is 1000. The weight feature data is reconstructed by dimensionality reduction to obtain a fused feature vector. The reconstruction error is calculated for feature selection, and then the correlation score threshold is set to 0.8 and the feature similarity threshold is set to 0.75. When the two thresholds are triggered at the same time and the duration exceeds 60 seconds, it is determined that the distributed power supply has a fault, and the fault is identified through a random forest. The random forest contains 200 decision trees, the maximum depth of the tree is set to 8, and the minimum number of leaf node samples is set to 5. The contribution of each feature to fault classification is calculated based on the training data.

[0103] The present invention uses a convolutional neural network model to efficiently process and analyze the correlation of multi-dimensional features to ensure the validity and accuracy of the features; uses a sliding window to segment the multi-dimensional feature matrix, and quantifies the Pearson correlation coefficient between each segment of feature data and historical fault data, which not only considers the overall trend of the feature data, but also pays attention to local changes, thereby improving the sensitivity of fault detection; uses the correlation score as a benchmark value, dynamically allocates weights and uses a five-layer structure autoencoder for feature fusion, so that key features receive more attention in subsequent processing, thereby improving the accuracy of fault detection, and also considers the overall similarity and local correlation of feature data, thereby improving the reliability and accuracy of fault detection; comprehensively considers multiple factors when making fault judgments, thereby ensuring the accuracy and reliability of fault judgments, and providing strong support for the rapid processing of faults and the stable operation of distributed power supply systems.

[0104] S2. judging the fault mode of the distributed power source based on the correlation and the pre-built fault mode classification library to obtain the current fault mode, and extracting the corresponding fault operation state characteristics from the selected historical monitoring data of the distributed power source according to the current fault mode;

[0105] In one embodiment, step S2 includes:

[0106] When it is determined that the distributed power supply has a fault according to the correlation, the frequency domain features and the time domain features are processed by a four-layer convolutional neural network model to obtain a multi-dimensional fault feature vector;

[0107] quantifying the matching degree between the multidimensional fault feature vector and various types of fault mode feature templates in the fault mode classification library based on cosine similarity, so as to determine the current fault mode of the distributed power source according to the matching degree;

[0108] Determine historical monitoring target data from the selected historical monitoring data of the distributed power source according to the current fault mode, and process the historical monitoring target data through 4-layer wavelet decomposition to determine the fault occurrence time and the fault recovery time;

[0109] Performing mean clustering processing on the historical monitoring period data at a preset time before the fault occurs, and taking the cluster center with the smallest distance from the historical monitoring period data as the stable operation feature before the fault;

[0110] The historical monitoring target data, the historical monitoring data corresponding to the fault recovery time, and the stable operation characteristics before the fault are combined by a support vector machine to obtain a complete fault feature vector as the fault operation state feature output.

[0111] Specifically, when determining that a distributed power supply has a fault based on the calculated correlation, the present invention uses a convolutional neural network with a four-layer structure to process the frequency domain characteristics and time domain characteristics of the load ratio, voltage ratio, and temperature ratio. The convolution kernel size is set to 3, and the number of channels is 16, 32, 64, and 128, respectively. The mean drift, mutation amplitude, and fluctuation frequency of these ratio parameters are extracted to form a multidimensional fault feature vector. Then, based on the cosine similarity, the matching degree between the multidimensional fault feature vector and various fault mode feature templates pre-stored in the fault mode classification library is calculated. If the matching degree exceeds 0.85, it is determined to be the fault mode, and the current fault mode of the distributed power supply is obtained. There are several fault mode feature templates stored in the fault mode classification library, which include fault modes, fault feature vectors, etc.

[0112] According to the current fault mode determined, the historical monitoring target data corresponding to the current fault mode is processed by 4-layer wavelet decomposition. The db4 wavelet basis function is selected to extract the energy distribution and time domain characteristics of the four frequency bands, and the change rate of these characteristic parameters in 100 adjacent sampling points is calculated. If the change rate exceeds 0.3 and lasts for more than 300 seconds, the position where the change rate begins to exceed the limit is marked as the time of fault occurrence, and the position where the change rate returns to normal is marked as the time of fault recovery.

[0113] Then, we traced back 1800 seconds of operation data from the time of the fault, and used k-means clustering to process the historical monitoring period data before the fault, that is, the historical operation status ratio parameters before the fault. The number of clusters was set to 3, and the Euclidean distance between each cluster center and the characteristic parameter sequence was calculated, and the cluster center with the smallest distance was selected as the stable operation feature before the fault. Finally, we used support vector machine to combine the historical monitoring target data, the historical monitoring data corresponding to the fault recovery time, and the stable operation features before the fault, and constructed a complete fault feature vector as the fault operation status feature output.

[0114] The present invention utilizes a four-layer convolutional neural network model for efficient multi-dimensional feature extraction, which can more accurately capture the complexity and subtle differences of fault data, thereby improving the accuracy of fault judgment; by quantifying the matching degree between the multi-dimensional fault feature vector and various fault mode feature templates in the fault mode classification library through cosine similarity, the current fault mode of the distributed power supply can be accurately determined, which not only improves the speed of fault identification, but also ensures the accuracy of the fault mode, providing strong support for subsequent fault processing; 4-layer wavelet decomposition is used for accurate fault time positioning, so as to quickly respond to faults and reduce the fault impact time; a complete fault feature vector is output through a support vector machine, which not only contains the whole process information of fault occurrence, development and recovery, but also reflects the stable operating state before the fault, providing a comprehensive and accurate basis for subsequent fault analysis and processing; the whole scheme combines a variety of advanced algorithms and technologies such as deep learning, wavelet decomposition, mean clustering and support vector machine, realizes the intelligent judgment and processing of distributed power supply faults, which not only improves the efficiency and accuracy of fault processing, but also reduces the cost and risk of manual intervention.

[0115] S3, inputting the fault operation state characteristics into a pre-built limit state severity assessment model for processing, obtaining a severity assessment result corresponding to the current fault mode, and extracting a current countermeasure from a pre-built fault countermeasure library according to the severity assessment result;

[0116] In one embodiment, step S3 includes:

[0117] The fault operation state characteristics are processed using the limit state severity assessment model, and the processing result is weighted according to the weight coefficient corresponding to the first real-time monitoring parameter to obtain a state assessment score as a severity assessment result corresponding to the current fault mode;

[0118] The severity level of the distributed power source under the extreme operating state is determined according to the state assessment score to trigger a corresponding target warning signal, and based on the target warning signal, a corresponding response parameter combination is extracted from the fault response measure library as the current response measure.

[0119] Specifically, the present invention uses a state evaluator based on a gradient boosting tree as a limit state severity assessment model to process the fault operation state characteristics. The model is a three-layer tree structure, each layer contains 50 decision trees, the maximum depth of the tree is set to 6, the minimum number of leaf node samples is 10, and the processing results are weighted according to the load ratio weight 0.4, the voltage ratio weight 0.3, and the temperature ratio weight 0.3 to obtain a state assessment score between 0 and 1 as the severity assessment result corresponding to the current fault mode. Taking photovoltaic power generation equipment as an example, when the load ratio parameter drops from 0.8 to 0.65, the corresponding weight 0.4 is calculated to obtain a contribution score of 0.32, the voltage ratio drops from 1.0 to 0.9, the weight 0.3 contributes to 0.27, the temperature ratio rises from 0.7 to 0.85, and the weight 0.3 contributes to 0.31. The weighted calculation of the three parameters obtains a state assessment score of 0.78.

[0120] A four-level warning mechanism is established in advance based on the status assessment score. The mild warning threshold of 0.6 corresponds to a parameter exceeding the rated value by 20%, the moderate warning threshold of 0.7 corresponds to a parameter exceeding the rated value by 35%, the severe warning threshold of 0.8 corresponds to a parameter exceeding the rated value by 50%, and the critical warning threshold of 0.9 corresponds to a parameter exceeding the rated value by 70%. When the status assessment score exceeds the corresponding warning threshold and lasts for more than 180 seconds, the target warning signal of the corresponding level is triggered, and the response parameter combination of the corresponding warning level is extracted from the pre-built fault response measure library. For example, a smooth response parameter combination is used for mild warnings, and the response parameter adjustment range is limited to the rated value. The value is within the range of plus or minus 15%, the moderate warning adopts a quick response response parameter combination, and the adjustment interval is expanded to plus or minus 30% of the rated value. The severe warning adopts a deep adjustment response parameter combination, and the adjustment interval is expanded to plus or minus 45% of the rated value. The critical warning adopts an extreme response parameter combination, and the adjustment interval is expanded to plus or minus 60% of the rated value. The control step is set based on the parameter deviation value to use it as the current response measure and control the execution of distributed power sources; among them, the fault response measure library contains the response parameters corresponding to various fault modes, the fault status assessment score range, and the system fluctuation data and recovery rate after the corresponding fault measures are executed.

[0121] Among them, closed-loop control is performed on the corresponding parameter combination through a proportional-integral regulator, the proportional coefficient is set to 0.6 and the integral coefficient is set to 0.4, the sampling period is set to 1 second, the output correction amount is calculated according to the parameter deviation value, and the status assessment score is recalculated based on the corrected operating parameters. If the score drops below the corresponding warning threshold and lasts for 300 seconds, the warning signal of this level is lifted and it is determined that the operating status has returned to a safe range. The safety range is defined as the deviation of each parameter from the rated value not exceeding 10%.

[0122] The present invention considers multiple dimensions of fault characteristics through a network model to perform accurate severity assessment, which not only improves the accuracy of the assessment, but also helps to better understand the nature and impact of the fault; the reasonable application of weight coefficients can fully consider the importance of different monitoring parameters in the assessment, ensure the objectivity and fairness of the assessment results, and avoid misjudgment due to abnormal individual parameters; the severity level of the distributed power supply under the extreme operating state is determined according to the state assessment score, which provides a clear basis for the triggering of subsequent warning signals and the selection of response measures; timely warning signal triggering wins precious time for rapid response and processing of faults; based on the target warning signal, the corresponding response parameter combination is extracted from the fault response measure library, which can ensure the pertinence and effectiveness of the current response measures, avoid blind response and ineffective processing, and improve the efficiency and accuracy of fault processing.

[0123] S4, collecting a second real-time monitoring parameter of the distributed power source when executing the current countermeasure, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the countermeasure effect of the current countermeasure according to the parameter change trend curve;

[0124] Specifically, during the execution of the current countermeasures, the second real-time operating status data is continuously acquired and each ratio parameter, namely, the second real-time monitoring data, is calculated, and then the change trend of each ratio parameter is identified and analyzed online to dynamically evaluate the execution effect of the countermeasures.

[0125] In one embodiment, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the response effect of the current response measure according to the parameter change trend curve, includes:

[0126] The second real-time monitoring parameter is processed using a three-layer convolutional neural network model to obtain a parameter change trend feature, and the mean, standard deviation, first-order difference value and second-order difference value of the parameter change trend feature in a sliding window are quantified to construct a multidimensional parameter change trend feature vector;

[0127] Processing the multidimensional parameter change trend characteristic vector according to wavelet decomposition to obtain high-frequency fluctuation characteristics and low-frequency drift characteristics, quantifying the recovery rates of the high-frequency fluctuation characteristics and the low-frequency drift characteristics based on wavelet changes, and obtaining the parameter change trend curve;

[0128] Predicting the parameter change trend curve through a Kalman filter, and calculating the deviation value between the parameter change trend curve and the normal range boundary based on the prediction result, so as to construct a response effect evaluation index according to the deviation value;

[0129] The response effect evaluation index is weightedly fused. If the processed response effect evaluation index is within the rated standard range and the standard deviation of the second real-time monitoring parameter is less than the preset standard threshold, it is determined that the operating status of the distributed power supply has returned to normal, and it is determined that the response effect of the current response measure has reached the expected standard.

[0130] Specifically, the present invention adopts a convolutional neural network model with a three-layer structure to process the second real-time monitoring parameter in real time. The convolution kernel size of each layer of the model is set to 3, the number of channels is 32, 64, and 128 respectively, and the sampling period is set to 0.1 second. The parameter change trend characteristics are obtained, and sliding window processing is performed on them. The window length is 1000 sampling points to calculate the parameter mean, standard deviation, first-order difference value, and second-order difference value of the parameter change trend characteristics in the window, and then construct a multi-dimensional parameter change trend feature vector.

[0131] The parameter change trend feature vector is decomposed into four layers of wavelets using the db4 wavelet basis function. The mean, standard deviation and kurtosis of the wavelet coefficients of each layer are calculated to extract the high-frequency fluctuation characteristics and low-frequency drift characteristics of the feature vector. The recovery rate of the parameter change trend feature vector in different frequency bands, i.e., corresponding to the high-frequency fluctuation characteristics and low-frequency drift characteristics, is calculated based on the wavelet coefficients. The recovery rate is compared with the preset threshold of 0.1. When the threshold is exceeded, the inflection point of the parameter change is recorded to generate a parameter change trend curve.

[0132] The parameter change trend curve is predicted through the Kalman filter, and the state space equation is established based on the latest 5000 data points. The prediction window length is set to 100 data points. The deviation value between the parameter change trend curve and the normal range boundary is calculated according to the prediction result. When the standard deviation of the deviation value is less than 0.05, the parameter is judged to enter the stable recovery stage, and then the response effect evaluation index including prediction error, recovery rate and stability is established.

[0133] Finally, the response effect evaluation indicators are weighted and fused, and the load ratio parameter weight is set to 0.4, the voltage ratio parameter weight is set to 0.3, and the temperature ratio parameter weight is set to 0.3. The parameter recovery status is judged based on the fused evaluation indicators. When the three parameters operate stably within the range of plus or minus 10% of the rated value, and the standard deviation is less than 0.03 within 300 seconds of continuous monitoring, it is determined that the equipment operation status has returned to normal, and it is determined that the response effect of the current response measures has met the expected standards.

[0134] The present invention uses a three-layer convolutional neural network model to perform multidimensional feature extraction on the second real-time monitoring parameter to construct a multidimensional parameter change trend feature vector, thereby improving the accuracy of feature extraction and enhancing the comprehensiveness and representativeness of the feature vector; using wavelet decomposition to perform fine frequency band feature analysis on the multidimensional parameter change trend feature vector, it is helpful to deeply understand the inherent law of parameter change and the influence of different frequency bands on the operating state, and provides an important basis for the subsequent recovery rate quantification and response effect evaluation; weighted fusion processing is performed on the response effect evaluation index, and combined with the comparison result of the standard deviation of the second real-time monitoring parameter with the preset standard threshold, it can scientifically determine whether the operating state of the distributed power supply has returned to normal and whether the response effect of the current response measures has reached the expected standard, which takes into account the overall change trend of the parameters and the fluctuation of the parameters, and ensures the accuracy and reliability of the determination result; the whole scheme realizes the real-time processing, feature extraction, frequency band analysis, prediction evaluation and response effect determination of the real-time monitoring parameters of the distributed power supply, can monitor the operating state in real time and evaluate the effect of the response measures, which is helpful to timely discover and deal with potential problems and improve the stability and reliability of the distributed power system.

[0135] S5. When the response effect does not meet the expected standard, the current response measure is modeled by using a gradient boosting tree to obtain a response parameter feature matrix to quantify the recovery rate of the distributed power source under the current response measure, and a target response measure is selected from the fault response measure library according to the recovery rate to control the execution of the distributed power source;

[0136] In one embodiment, step S5 includes:

[0137] When the response effect does not meet the expected standard, a gradient boosting tree is used to model the current response measures and the second real-time monitoring parameters to obtain a response parameter feature matrix, and the recovery rate and parameter recovery amount of the distributed power source in unit time are counted according to the response parameter feature matrix;

[0138] Fitting the parameter recovery amount by a fourth-order polynomial to obtain a parameter recovery fitting curve, and extracting a parameter recovery index from the parameter recovery fitting curve according to the recovery rate and using it as a response effect evaluation index;

[0139] A three-layer neural network model is used to process the response effect evaluation index to obtain a comprehensive recovery effect score, and a weighted calculation is performed on the comprehensive recovery effect score to construct a response measure effect evaluation criterion based on the weighted calculation result;

[0140] According to the countermeasure effect evaluation criteria, from the fault countermeasure library, the fault countermeasure whose recovery rate reaches a preset rate threshold and whose corresponding fluctuation standard deviation is the smallest when the distributed power source is executed is taken as the target countermeasure to control the execution of the distributed power source.

[0141] Specifically, when the response effect does not meet the expected standard, the present invention uses a gradient boosting tree to model the current response measures and the second real-time monitoring parameters during their execution, sets the maximum depth of the tree to 8, and the minimum number of leaf node samples to 10, to construct a response parameter feature matrix including response measure type, parameter setting value, execution time, and parameter interaction intensity, and based on the response parameter feature matrix, uses a 60-second time window to calculate the recovery rate and recovery change of distributed power sources per unit time.

[0142] The parameter recovery amount is fitted with a fourth-order polynomial to obtain a parameter recovery fitting curve, in order to extract the parameter recovery indicators of the curve, such as the mean recovery rate, the standard deviation of the recovery process, the maximum fluctuation range, and the degree of crosstalk between parameters as response effect evaluation indicators. The parameter recovery standard deviation threshold is set to 0.05, the maximum fluctuation range threshold is set to 0.1, and the crosstalk degree threshold is set to 0.15. The effectiveness of the response measures is quantitatively evaluated based on the response effect evaluation indicators.

[0143] Finally, a three-layer neural network structure model is used to process the response effect evaluation indicators. The input layer is set with 12 nodes corresponding to the current response measures and parameter recovery indicators, the number of hidden layer nodes is 24, the activation function uses a linear rectification function, the number of training rounds is set to 1000, the learning rate is set to 0.01, and the output layer corresponds to the comprehensive recovery effect score of the three ratio parameters. The response effect evaluation criteria are established by weighted calculation.

[0144] According to the countermeasure effect evaluation criteria, the target countermeasure is retrieved from the fault countermeasure library, and the load ratio parameter unit time recovery rate threshold is set to 0.02, the voltage ratio parameter unit time recovery rate threshold is set to 0.015, and the temperature ratio parameter unit time recovery rate threshold is set to 0.01. The fault countermeasure whose recovery rate meets the threshold requirements and has the smallest fluctuation standard deviation, that is, the countermeasure parameter combination, is selected as the target countermeasure. After verifying the reliability of the selected countermeasure based on historical execution data, the distributed power generation is controlled to execute.

[0145] The present invention constructs a response parameter feature matrix through a gradient boosting tree, which can comprehensively capture the complex relationship between response measures and monitoring parameters; according to the response parameter feature matrix, the recovery rate and parameter recovery amount of distributed power sources in unit time are counted, and the influence of response measures on fault recovery can be quantitatively evaluated, which not only considers the time factor, but also takes into account the actual situation of parameter changes, ensuring the accuracy and reliability of the evaluation results; high-precision parameter recovery fitting is performed through a fourth-order polynomial to accurately describe the changing trend of parameter recovery, which provides a basis for the subsequent extraction of parameter recovery indicators from the fitting curve; a three-layer neural network model is used to process the parameter recovery indicator to obtain a comprehensive recovery effect score, which comprehensively considers the influence of multiple evaluation indicators and constructs a more comprehensive and objective response measure effect evaluation criterion; based on the evaluation criterion, the target response measure is selected from the fault response measure library, which not only considers the requirements of the recovery rate, but also takes into account the stability during the execution process, ensuring the pertinence and effectiveness of the response measures.

[0146] In one embodiment, when the distributed power source enters the extreme operating state again, a dynamic monitor based on wavelet transform is used to perform time-frequency analysis on the load ratio parameters, voltage ratio parameters, and temperature ratio parameters of the distributed power source that executes the target response measures. The sampling frequency is set to 100 Hz, and the db4 wavelet basis function is selected for 4-layer decomposition. The dominant frequency amplitude, spectrum entropy value, and frequency domain energy distribution characteristics are extracted through a 500-point sliding window, and the mean, variance, and kurtosis values ​​of the coefficients of each frequency band are calculated to construct a multi-dimensional monitoring feature matrix.

[0147] According to the constructed monitoring feature matrix, a convolutional neural network is used to extract the fluctuation period and the duration characteristics of the rising and falling edges. The network structure contains three convolutional layers, the convolution kernel size is set to 3, the number of channels is 32, 64, and 128 respectively, and the input data format is a 36×500 feature sequence. The periodic change law and the rising and falling edge change trend in the fluctuation feature sequence are extracted, and the mapping relationship matrix between the frequency domain features and the time domain features is established based on the extracted features.

[0148] The characteristic change trend is dynamically tracked through the proportional-integral regulator. The frequency domain characteristic proportional coefficient is set to 0.6 and the integral coefficient is set to 0.4, and the time domain characteristic proportional coefficient is set to 0.5 and the integral coefficient is set to 0.5. The deviation between the characteristic change amount and the preset threshold is calculated. When the frequency domain characteristic change rate exceeds 0.1 or the time domain characteristic change rate exceeds 0.15, the corresponding parameters are corrected online based on the mapping relationship matrix, and the correction step is set to 0.2 times the current deviation value.

[0149] According to the corrected response parameters, a new characteristic change trend curve is calculated, and the weight coefficients of the three characteristics of dominant frequency amplitude, spectral entropy value and fluctuation period are set to 0.4, 0.3 and 0.3 respectively. Weighted constraints are imposed on the correction amount of the response parameters. Through recursive calculation, the stability index of the characteristic change trend is guaranteed to be less than 0.1, and the coupling index between parameters is less than 0.15, so as to ensure the effectiveness of the correction of the response parameters.

[0150] In the embodiment of the present application, based on the problem of how to improve the accuracy of fault handling of distributed power sources, a fault handling method for distributed power sources is designed, which realizes obtaining a first real-time monitoring parameter of the distributed power source under an extreme operating state, and extracting the frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter, so as to analyze the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain characteristics and the time domain characteristics; based on the correlation and a pre-constructed fault mode classification library, the fault mode of the distributed power source is judged to obtain the current fault mode, and according to the current fault mode, the corresponding fault operation state characteristics are extracted from the selected historical monitoring data of the distributed power source; the fault operation state characteristics are input into a pre-constructed extreme state severity assessment model for processing to obtain the corresponding fault state characteristics of the current fault mode. severity assessment results, to extract current countermeasures from a pre-built fault countermeasure library according to the severity assessment results; collect a second real-time monitoring parameter of the distributed power source when executing the current countermeasure, to generate a parameter change trend curve based on the second real-time monitoring parameter, and evaluate the countermeasure effect of the current countermeasure according to the parameter change trend curve; when the countermeasure effect does not meet the expected standard, use a gradient boosting tree to model the current countermeasure, obtain a countermeasure parameter feature matrix to quantify the recovery rate of the distributed power source under the current countermeasure, and select a target countermeasure from the fault countermeasure library according to the recovery rate to control the execution of the distributed power source. The technical solution can significantly improve the reliability and safety of distributed power sources under extreme operating conditions, and greatly improve the fault response speed and processing efficiency.

[0151] It should be noted that although the steps in the above flowchart are shown in sequence according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders.

[0152] In another embodiment, if Figure 2 As shown, the second aspect of the present invention provides a fault handling system for a distributed power source, comprising:

[0153] A correlation calculation module 10 is used to obtain a first real-time monitoring parameter of a distributed power source under an extreme operating state, and extract a frequency domain feature and a time domain feature of the first real-time monitoring parameter, so as to analyze the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain feature and the time domain feature;

[0154] A state feature extraction module 20 is used to judge the fault mode of the distributed power source based on the correlation and the pre-built fault mode classification library to obtain the current fault mode, and extract the corresponding fault operation state feature from the selected historical monitoring data of the distributed power source according to the current fault mode;

[0155] A countermeasure execution module 30 is used to input the fault operation state characteristics into a pre-built limit state severity assessment model for processing, obtain a severity assessment result corresponding to the current fault mode, and extract a current countermeasure from a pre-built fault countermeasure library according to the severity assessment result;

[0156] A response effect evaluation module 40, configured to collect a second real-time monitoring parameter of the distributed power source when executing the current response measure, to generate a parameter change trend curve based on the second real-time monitoring parameter, and to evaluate the response effect of the current response measure according to the parameter change trend curve;

[0157] The target measure execution module 50 is used to use a gradient boosting tree to model the current countermeasure when the countermeasure effect does not meet the expected standard, obtain a countermeasure parameter characteristic matrix to quantify the recovery rate of the distributed power source under the current countermeasure, and select a target countermeasure from the fault countermeasure library according to the recovery rate to control the execution of the distributed power source.

[0158] It should be noted that each module in the above-mentioned fault handling system for distributed power sources can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules. For the specific definition of a fault handling system for distributed power sources, please refer to the definition of a fault handling method for distributed power sources in the above text. The two have the same functions and effects, which will not be repeated here.

[0159] In summary, the present invention relates to the field of data processing technology, and discloses a fault handling method and system for distributed power sources, which monitors the operating data of distributed power sources under extreme operating conditions in real time and combines it with historical fault data to judge the fault mode; according to the current fault mode of the distributed power source, the corresponding fault operating state characteristics are extracted from the historical monitoring data to evaluate the severity, and according to the evaluation result, the second real-time monitoring parameters at that time are extracted from the fault response measures library to evaluate the response effect; when the effect does not meet the expected standard, based on the recovery rate of the distributed power source under the current response measures, a better target response measure is selected from the fault response measures library and executed, which can significantly improve the reliability and safety of the distributed power source under extreme operating conditions, and greatly improve the fault response speed and processing efficiency.

[0160] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0161] The above-mentioned embodiments only express several preferred implementation modes of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in the technical field, several improvements and substitutions can be made without departing from the technical principles of the present invention, and these improvements and substitutions should also be regarded as the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be based on the protection scope of the claims.

Claims

1. A fault handling method for distributed power supply, characterized in that: include: Acquire a first real-time monitoring parameter of a distributed power source under an extreme operating state, and extract a frequency domain feature and a time domain feature of the first real-time monitoring parameter, so as to analyze a correlation between the first real-time monitoring parameter and historical fault data of the distributed power source based on the frequency domain feature and the time domain feature; The fault mode of the distributed power source is judged based on the correlation and a pre-built fault mode classification library to obtain a current fault mode, and corresponding fault operation state characteristics are extracted from historical monitoring data of the selected distributed power source according to the current fault mode; Inputting the fault operation state characteristics into a pre-built limit state severity assessment model for processing, obtaining a severity assessment result corresponding to the current fault mode, and extracting a current countermeasure from a pre-built fault countermeasure library according to the severity assessment result; Collecting a second real-time monitoring parameter of the distributed power source when executing the current countermeasure, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the countermeasure effect of the current countermeasure according to the parameter change trend curve; When the response effect does not meet the expected standard, the current response measure is modeled by using a gradient boosting tree to obtain a response parameter feature matrix to quantify the recovery rate of the distributed power source under the current response measure, and a target response measure is selected from the fault response measure library according to the recovery rate to control the execution of the distributed power source; The obtaining of a first real-time monitoring parameter of a distributed power source under a limit operation state includes: Acquire real-time operating status data of the distributed power source, and determine an operating status ratio parameter of the distributed power source according to the real-time operating status data, so as to normalize the operating status ratio parameter through a convolutional neural network model to obtain a normalized ratio parameter sequence; quantifying the standard deviation of the normalized ratio parameter sequence within the time window to determine an abnormality benchmark, and quantifying an abnormality degree score of the normalized ratio parameter sequence according to the abnormality benchmark; Quantifying the change rate of the normalized ratio parameter sequence through a sliding time window to obtain a parameter change trend value, and inputting it into a support vector machine with the abnormality degree score to extract features of the real-time operating status data to obtain a feature space clustering result; When the center point of the feature space clustering result reaches a preset center threshold standard and the duration of the center point exceeds a preset duration, it is determined that the distributed power source is in the limit operation state, and the operation state ratio parameter is output as the first real-time monitoring parameter; The step of determining the operating status ratio parameter of the distributed power source according to the real-time operating status data comprises: Determine the initial operating state ratio parameter of the distributed power source according to the relationship between the real-time operating state data and the corresponding operating state standard data, and standardize the initial operating state ratio parameter through a convolutional neural network model to obtain a standardized initial operating state ratio parameter; A filter is used to perform primary correction and smoothing processing on the standardized initial operating state ratio parameter to obtain a smoothed initial operating state ratio parameter, and a jump point in the smoothed initial operating state ratio parameter is secondary corrected to obtain a corrected initial operating state ratio parameter; The covariance matrix between the modified initial operating state ratio parameters is quantified to perform correlation calculation, and the modified initial operating state ratio parameters are data fitted based on the correlation calculation results to obtain a ratio parameter fitting curve for linear interpolation completion to obtain the operating state ratio parameters of the distributed power source.

2. A fault handling method for distributed power supply according to claim 1, characterized in that: The frequency domain features include the dominant frequency amplitude and spectrum entropy, and the time domain features include the time domain waveform distortion, fluctuation cycle length, rising edge duration and falling edge duration; wherein, The extracting the frequency domain features and the time domain features of the first real-time monitoring parameter includes: Performing frequency domain conversion on the first real-time monitoring parameter by fast Fourier transform, so as to quantize a power spectrum density curve according to the frequency domain conversion result, and determining the dominant frequency amplitude of the first real-time monitoring parameter according to the power spectrum density curve; quantifying the spectrum entropy of the first real-time monitoring parameter according to the energy proportion of different frequency bands in the power spectrum density curve, and performing bandwidth limitation on the first real-time monitoring parameter through a third-order Bessel bandpass filter to obtain a waveform envelope curve and its corresponding time domain waveform distortion; quantifying the length of the fluctuation cycle based on the time difference between the local maximum points of the waveform in the waveform envelope curve, and marking the inflection points of the waveform envelope curve using a long short-term memory neural network model to quantify the duration of the rising edge and the duration of the falling edge according to the marked inflection points; The rising edge duration and the falling edge duration are decomposed by wavelet transform to obtain wavelet coefficient feature vectors, and the dominant frequency amplitude, spectrum entropy, time domain waveform distortion, fluctuation period length, rising edge duration and falling edge duration are spliced ​​into a multidimensional feature space through the wavelet coefficient feature vectors, and the multidimensional feature space is reduced in principal components to obtain the frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter.

3. A fault handling method for distributed power supply according to claim 1, characterized in that: The analyzing the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain feature and the time domain feature includes: Processing the frequency domain features and time domain features of the first real-time monitoring parameter through a convolutional neural network model to obtain a multidimensional feature matrix and segmenting it through a sliding window, quantifying the Pearson correlation coefficient between each segment of feature data and the historical fault data and performing normalization processing to obtain a correlation score; Taking the correlation score as a reference value, quantifying the Euclidean distance between the multidimensional feature matrix and the reference value, superimposing the distance values ​​in adjacent time windows and comparing them with a preset distance threshold, and assigning a weight to each segment of feature data based on the comparison result to obtain weighted feature data; A five-layer structure autoencoder is used to fuse the weight feature data to obtain a fused feature vector to quantify feature similarity. When the feature similarity and correlation score of the fused feature vector simultaneously reach corresponding thresholds and the duration reaches a preset duration threshold, it is determined that a fault exists in the distributed power supply to realize the correlation analysis between the first real-time monitoring parameter and the historical fault data.

4. A method for handling faults in a distributed power source according to claim 3, characterized in that: The fault mode of the distributed power source is judged based on the correlation and the pre-built fault mode classification library to obtain the current fault mode, and the corresponding fault operation state characteristics are extracted from the selected historical monitoring data of the distributed power source according to the current fault mode, including: When it is determined that the distributed power supply has a fault according to the correlation, the frequency domain features and the time domain features are processed by a four-layer convolutional neural network model to obtain a multi-dimensional fault feature vector; quantifying the matching degree between the multidimensional fault feature vector and various types of fault mode feature templates in the fault mode classification library based on cosine similarity, so as to determine the current fault mode of the distributed power source according to the matching degree; Determine historical monitoring target data from the selected historical monitoring data of the distributed power source according to the current fault mode, and process the historical monitoring target data through 4-layer wavelet decomposition to determine the fault occurrence time and the fault recovery time; Performing mean clustering processing on the historical monitoring period data at a preset time before the fault occurs, and taking the cluster center with the smallest distance from the historical monitoring period data as the stable operation feature before the fault; The historical monitoring target data, the historical monitoring data corresponding to the fault recovery time, and the stable operation characteristics before the fault are combined by a support vector machine to obtain a complete fault feature vector as the fault operation state feature output.

5. A fault handling method for distributed power supply according to claim 1, characterized in that: The inputting of the fault operation state characteristics into a pre-built limit state severity assessment model for processing to obtain a severity assessment result corresponding to the current fault mode, and extracting current countermeasures from a pre-built fault countermeasure library according to the severity assessment result, includes: The fault operation state characteristics are processed using the limit state severity assessment model, and the processing result is weighted according to the weight coefficient corresponding to the first real-time monitoring parameter to obtain a state assessment score as a severity assessment result corresponding to the current fault mode; The severity level of the distributed power source under the extreme operating state is determined according to the state assessment score to trigger a corresponding target warning signal, and based on the target warning signal, a corresponding response parameter combination is extracted from the fault response measure library as the current response measure.

6. A fault handling method for distributed power supply according to claim 1, characterized in that: The generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the response effect of the current response measure according to the parameter change trend curve, includes: The second real-time monitoring parameter is processed using a three-layer convolutional neural network model to obtain a parameter change trend feature, and the mean, standard deviation, first-order difference value and second-order difference value of the parameter change trend feature in a sliding window are quantified to construct a multidimensional parameter change trend feature vector; Processing the multidimensional parameter change trend characteristic vector according to wavelet decomposition to obtain high-frequency fluctuation characteristics and low-frequency drift characteristics, quantifying the recovery rates of the high-frequency fluctuation characteristics and the low-frequency drift characteristics based on wavelet changes, and obtaining the parameter change trend curve; Predicting the parameter change trend curve through a Kalman filter, and calculating the deviation value between the parameter change trend curve and the normal range boundary based on the prediction result, so as to construct a response effect evaluation index according to the deviation value; The response effect evaluation index is weightedly fused. If the processed response effect evaluation index is within the rated standard range and the standard deviation of the second real-time monitoring parameter is less than the preset standard threshold, it is determined that the operating status of the distributed power supply has returned to normal, and it is determined that the response effect of the current response measure has reached the expected standard.

7. A fault handling method for distributed power supply according to claim 1, characterized in that: When the response effect does not meet the expected standard, the current response measures are modeled by using a gradient boosting tree to obtain a response parameter feature matrix to quantify the recovery rate of the distributed power source under the current response measures, and a target response measure is selected from the fault response measure library according to the recovery rate to control the execution of the distributed power source, including: When the response effect does not meet the expected standard, a gradient boosting tree is used to model the current response measures and the second real-time monitoring parameters to obtain a response parameter feature matrix, and the recovery rate and parameter recovery amount of the distributed power source in unit time are counted according to the response parameter feature matrix; Fitting the parameter recovery amount by a fourth-order polynomial to obtain a parameter recovery fitting curve, and extracting a parameter recovery index from the parameter recovery fitting curve according to the recovery rate and using it as a response effect evaluation index; A three-layer neural network model is used to process the response effect evaluation index to obtain a comprehensive recovery effect score, and a weighted calculation is performed on the comprehensive recovery effect score to construct a response measure effect evaluation criterion based on the weighted calculation result; According to the countermeasure effect evaluation criteria, from the fault countermeasure library, the fault countermeasure whose recovery rate reaches a preset rate threshold and whose corresponding fluctuation standard deviation is the smallest when the distributed power source is executed is taken as the target countermeasure to control the execution of the distributed power source.

8. A fault handling system for distributed power sources, characterized in that: include: A correlation calculation module, used to obtain a first real-time monitoring parameter of a distributed power source under an extreme operating state, and extract frequency domain characteristics and time domain characteristics of the first real-time monitoring parameter, so as to analyze the correlation between the first real-time monitoring parameter and the historical fault data of the distributed power source based on the frequency domain characteristics and the time domain characteristics; A state feature extraction module, used to judge the fault mode of the distributed power source based on the correlation and a pre-built fault mode classification library, obtain the current fault mode, and extract the corresponding fault operation state feature from the selected historical monitoring data of the distributed power source according to the current fault mode; A countermeasure execution module is used to input the fault operation state characteristics into a pre-built limit state severity assessment model for processing, obtain a severity assessment result corresponding to the current fault mode, and extract the current countermeasure from a pre-built fault countermeasure library according to the severity assessment result; A response effect evaluation module, used for collecting a second real-time monitoring parameter of the distributed power source when executing the current response measure, generating a parameter change trend curve based on the second real-time monitoring parameter, and evaluating the response effect of the current response measure according to the parameter change trend curve; A target measure execution module is used to use a gradient boosting tree to model the current countermeasure when the countermeasure effect does not meet the expected standard, obtain a countermeasure parameter feature matrix to quantify the recovery rate of the distributed power source under the current countermeasure, and select a target countermeasure from the fault countermeasure library according to the recovery rate to control the execution of the distributed power source; The obtaining of a first real-time monitoring parameter of a distributed power source under a limit operation state includes: Acquire real-time operating status data of the distributed power source, and determine an operating status ratio parameter of the distributed power source according to the real-time operating status data, so as to normalize the operating status ratio parameter through a convolutional neural network model to obtain a normalized ratio parameter sequence; quantifying the standard deviation of the normalized ratio parameter sequence within the time window to determine an abnormality benchmark, and quantifying an abnormality degree score of the normalized ratio parameter sequence according to the abnormality benchmark; Quantifying the change rate of the normalized ratio parameter sequence through a sliding time window to obtain a parameter change trend value, and inputting it into a support vector machine with the abnormality degree score to extract features of the real-time operating status data to obtain a feature space clustering result; When the center point of the feature space clustering result reaches a preset center threshold standard and the duration of the center point exceeds a preset duration, it is determined that the distributed power source is in the limit operation state, and the operation state ratio parameter is output as the first real-time monitoring parameter; The step of determining the operating status ratio parameter of the distributed power source according to the real-time operating status data comprises: Determine the initial operating state ratio parameter of the distributed power source according to the relationship between the real-time operating state data and the corresponding operating state standard data, and standardize the initial operating state ratio parameter through a convolutional neural network model to obtain a standardized initial operating state ratio parameter; A filter is used to perform primary correction and smoothing processing on the standardized initial operating state ratio parameter to obtain a smoothed initial operating state ratio parameter, and a jump point in the smoothed initial operating state ratio parameter is secondary corrected to obtain a corrected initial operating state ratio parameter; The covariance matrix between the modified initial operating state ratio parameters is quantified to perform correlation calculation, and the modified initial operating state ratio parameters are data fitted based on the correlation calculation results to obtain a ratio parameter fitting curve for linear interpolation completion to obtain the operating state ratio parameters of the distributed power source.

Images