Power supply fault rapid detection system and method based on intelligent diagnosis engine

Abstract

The present invention discloses a power supply fault rapid detection system and method based on an intelligent diagnostic engine, which belongs to the technical field of fault diagnosis of power electronic equipment, and includes: a functional area division module mainly used to divide the power supply into multiple functional areas according to the power supply topology structure; a real-time data acquisition module mainly used to obtain parameter data of each functional area during the power supply operation stage; a current model construction module mainly used to establish a current time series prediction model; a trend prediction module mainly used to generate future current change trends; an abnormality judgment module mainly used to identify abnormal situations by performing similarity matching with real-time data; a fault location module mainly used to match abnormal current waveforms with fault modes, and construct a directed graph to locate the fault position; the advantages of the present invention are: through multimodal data fusion and deep learning models, early warning and precise positioning of power supply faults are achieved, and the operation and maintenance efficiency and reliability of power electronic equipment are significantly improved.

Description

Technical Field

[0001] The present invention relates to fault detection, and in particular to a power supply fault rapid detection system and method based on an intelligent diagnosis engine. Background Art

[0002] With modern society's increasing reliance on electricity supply, the stability and security of power systems are becoming increasingly important. Power equipment failures or grid anomalies can lead to power outages, equipment damage, and even significant economic losses. Therefore, rapidly detecting power failures and promptly identifying and locating the source of the problem are crucial to ensuring reliable power system operation.

[0003] Current power supply fault detection systems and methods rely on fixed fault detection algorithms and simple relay protection devices. These methods typically provide only a rough fault response, making it difficult to achieve detailed monitoring and precise location of the power supply system. Traditional methods are slow to respond to faults and may not accurately identify even subtle changes in the current waveform, resulting in inaccurate fault location, prolonged repair time, and impacted system reliability and availability. Summary of the Invention

[0004] In order to improve the existing power supply fault detection system and method, a power supply fault rapid detection system and method based on an intelligent diagnostic engine is provided. This method accurately determines abnormal conditions of the power supply system through real-time data analysis and current time series prediction. Combined with current waveform matching and directed graph construction, it effectively locates the fault point and realizes rapid fault detection and location.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0006] Power supply fault rapid detection system based on intelligent diagnostic engine, including:

[0007] Functional area division module: The functional area division module is mainly used to divide the power supply into independent functional areas according to the functional relevance of each component in the power supply topology;

[0008] Real-time data acquisition module: The real-time data acquisition module is equipped with a distributed current sensor array, which is mainly used to obtain current, voltage and temperature data of each functional area during the power supply operation stage;

[0009] Current model building module: The current model building module is mainly used to establish a current time series prediction model based on historical operating data. The model includes the current baseline and allowable fluctuation threshold range of each component under standard operating conditions;

[0010] Trend prediction module: The trend prediction module is mainly used to generate the current change trend of each component in the future by inputting real-time monitoring data into the training model;

[0011] Abnormal judgment module: The abnormal judgment module is mainly used to identify abnormal situations based on the current change trend of each component in the future by matching the similarity with the real-time data;

[0012] Fault location module: The fault location module is mainly used to match the current waveform under abnormal conditions with the waveform under other fault types, and locate the fault point by constructing a directed graph;

[0013] Database module: The database module is mainly used to store real-time power parameter data, current model data, and abnormal fault data.

[0014] Preferably, the functional area division module specifically includes:

[0015] Analyze component types, parameters, and connection relationships based on the circuit schematic of the power supply system;

[0016] Analyze the flow of current and voltage signals and mark key nodes, establish the timing logic of component actions, and determine the functional dependency chain;

[0017] Typical functional areas are divided into: input processing area, power conversion area, output voltage regulation area, control and protection area.

[0018] Preferably, the current model building module specifically includes:

[0019] Based on historical power supply operation data, a current time series prediction model is constructed, including:

[0020] Data preprocessing unit: The data preprocessing unit is mainly used to perform wavelet denoising and outlier removal on historical current data;

[0021] Feature extraction unit: The feature extraction unit is mainly used to extract the fundamental component and harmonic distribution characteristics of the current signal using a sliding window Fourier transform;

[0022] Model training unit: The model training unit is trained through an LSTM neural network and constructs a nonlinear regression model of the current of each component changing with the load;

[0023] Model validation unit: The model validation unit is mainly used to ensure that each sub-model has a high goodness of fit through cross-validation.

[0024] Furthermore, the power supply fault rapid detection method based on the intelligent diagnosis engine includes:

[0025] Based on circuit topology analysis and functional relevance, the power supply is divided into multiple functional areas;

[0026] Based on historical power supply operation data, a current time series prediction model is built for each functional area. A diagnostic knowledge base containing component-level current baseline models is also built to train the current time series prediction model for each functional area.

[0027] Real-time collection of dynamic current signals from each functional area, input of real-time monitoring data into the training model, and generation of future current change trends for each component;

[0028] Based on the predicted current change trend, the real-time current waveform is matched with it and the similarity index is calculated;

[0029] Based on the similarity index, determine whether the current is abnormal;

[0030] Based on the fault situation, locate the fault source according to the components where the abnormal current is located.

[0031] Preferably, the matching of the real-time current waveform with the predicted current change trend and calculating the similarity index specifically include:

[0032] Based on the predicted current change trend, the predicted current waveform time series data Y=(y1,y2,...,y m ), obtain real-time current waveform time series data X=(x1,x2,...x n );

[0033] Based on the corresponding data points in the time series data, the distance between the two is calculated and a distance matrix is ​​constructed, in which each element stores x i and y j the distance between them;

[0034] Define the cumulative cost matrix, where each element stores the minimum cumulative distance from the starting point to the current point;

[0035] Based on the accumulated cost matrix after filling, backtrack from the lower right corner of the matrix to find the shortest path to the starting point;

[0036] Based on the obtained shortest path, obtain the value in the lower right corner of the cumulative cost matrix and perform normalization to obtain the similarity index. The smaller the value, the more similar the shapes and trends of the two time series data are.

[0037] Preferably, judging whether the current current is abnormal based on the similarity index specifically includes:

[0038] Based on the similarity index obtained in each time period, a function of similarity index change over time is constructed;

[0039] Based on the time-varying function of the similarity index, if the similarity shows a decreasing trend over time, it is determined that the current is abnormal and the power supply is faulty.

[0040] Preferably, locating the fault source based on the fault situation and the components where the abnormal current is located specifically includes:

[0041] Based on the original current waveform within the abnormal time window, time-frequency domain feature extraction is performed to construct a feature vector;

[0042] Based on the acquired fault characteristics, the fault patterns recorded in the diagnostic knowledge base are matched to screen and obtain candidate fault types;

[0043] Construct a fault propagation directed graph, with components within the power supply functional area as nodes and edge weights assigned based on the fault transmission probability of the circuit topology;

[0044] Based on the predicted current waveform time series and the real-time current waveform time series as constraints, the Dijkstra algorithm is used to calculate the shortest propagation path to the abnormal component with the candidate fault source as the starting point;

[0045] Based on the measured abnormal characteristics and propagation paths, the posterior fault probability of each node is updated, and the component with the highest posterior probability is taken as the fault source, and its functional area and location coordinates are output.

[0046] Compared with the prior art, the advantages of the present invention are:

[0047] The functional area division module divides the power supply into multiple independent functional areas based on the functional relevance of each component in the power supply topology, providing a clear structured foundation for subsequent fault analysis. The real-time data acquisition module, using a distributed current sensor array, comprehensively collects current, voltage, and temperature data from each functional area during power supply operation, ensuring comprehensive and real-time data acquisition. The current model construction module leverages historical operating data to establish a current time-series prediction model and accurately defines the current baseline and allowable fluctuation range for each component, providing a scientific basis for subsequent fault prediction. The trend prediction module feeds real-time monitoring data into a trained model to predict future component current trends and promptly identify potential abnormalities. The anomaly judgment module rapidly identifies system anomalies by matching similarity with real-time data. Finally, the fault location module combines current waveforms with fault type waveforms to construct a directed graph to precisely locate the fault point. This system and method achieve accurate and rapid detection and location of power supply faults, significantly improving the reliability and maintenance efficiency of power supply systems while avoiding the delays and misjudgments often associated with traditional fault diagnosis methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 A schematic diagram of the system proposed by the present invention;

[0049] Figure 2 This is a schematic diagram of the current model construction module proposed in the present invention;

[0050] Figure 3 A schematic diagram of the method proposed in the present invention;

[0051] Figure 4 A workflow diagram for obtaining the similarity index proposed in the present invention;

[0052] Figure 5 This is a schematic diagram of the abnormality judgment module proposed by the present invention;

[0053] Figure 6 This is a schematic diagram of the fault location module proposed in the present invention. DETAILED DESCRIPTION

[0054] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0055] See Figure 1 As shown in the figure, the power failure rapid detection system based on the intelligent diagnosis engine includes:

[0056] Functional area division module: The functional area division module is mainly used to divide the power supply into independent functional areas according to the functional relevance of each component in the power supply topology;

[0057] Real-time data acquisition module: The real-time data acquisition module is equipped with a distributed current sensor array, which is mainly used to obtain current, voltage and temperature data of each functional area during the power supply operation stage;

[0058] Current model building module: The current model building module is mainly used to establish a current time series prediction model based on historical operating data. The model includes the current baseline and allowable fluctuation threshold range of each component under standard operating conditions;

[0059] Trend prediction module: The trend prediction module is mainly used to generate the current change trend of each component in the future by inputting real-time monitoring data into the training model;

[0060] Abnormal judgment module: The abnormal judgment module is mainly used to identify abnormal situations based on the current change trend of each component in the future by matching the similarity with the real-time data;

[0061] Fault location module: The fault location module is mainly used to match the current waveform under abnormal conditions with the waveform under other fault types, and locate the fault point by constructing a directed graph;

[0062] Database module: The database module is mainly used to store real-time power parameter data, current model data, and abnormal fault data.

[0063] See Figure 1 As shown, the functional area division module specifically includes:

[0064] Analyze component types, parameters, and connection relationships based on the circuit schematic of the power supply system;

[0065] Analyze the flow of current and voltage signals and mark key nodes, establish the timing logic of component actions, and determine the functional dependency chain;

[0066] Typical functional areas are divided into: input processing area, power conversion area, output voltage regulation area, control and protection area.

[0067] Specifically, the timing logic of the component action includes: startup phase, steady-state phase, and protection triggering. In the startup phase, the input voltage is filtered, the control IC is powered on → the soft-start capacitor is charged → the PWM duty cycle gradually increases from 0. In the steady-state phase, the feedback node voltage is compared with the voltage → the error amplifier adjusts the PWM duty cycle → the voltage across the output filter capacitor is maintained stable. The protection triggers the overcurrent detection resistor voltage drop exceeding the threshold → the PWM is turned off → locked or hiccup mode;

[0068] The input processing area includes: EMI filter, rectifier bridge, surge protection circuit; the power conversion area includes: PFC inductor, switch tube, transformer; the output voltage regulation area includes: output filter capacitor, linear voltage regulator, load adjustment circuit; the control and protection area includes: control IC, feedback network, overvoltage and overcurrent protection circuit.

[0069] See Figure 2 As shown in Figure 1, the current model building module specifically includes:

[0070] Based on historical power supply operation data, a current time series prediction model is constructed, including:

[0071] Data preprocessing unit: The data preprocessing unit is mainly used to perform wavelet denoising and outlier removal on historical current data;

[0072] Feature extraction unit: The feature extraction unit is mainly used to extract the fundamental component and harmonic distribution characteristics of the current signal using a sliding window Fourier transform;

[0073] Model training unit: The model training unit is trained through an LSTM neural network and constructs a nonlinear regression model of the current of each component changing with the load;

[0074] Model validation unit: The model validation unit is mainly used to ensure that each sub-model has a high goodness of fit through cross-validation.

[0075] Specifically, the data preprocessing unit uses a threshold denoising algorithm based on wavelet basis to reduce the noise of the historical current data, and calculates the signal mean and standard deviation in the sliding window. The window length is:

[0076] ;

[0077] ;

[0078] in, is the mean of the signal in the window, reflecting the average level of the data, N is the length of the sliding window (that is, the number of data points included), t is the time, is the pre-processed current signal value at position i, is the standard deviation of the signal within the window, which measures the dispersion of the data around the mean;

[0079] Define outliers based on Z-score and replace them with linear interpolation or median filling;

[0080] The feature extraction unit uses sliding window Fourier transform (window length 10ms, overlap rate 50%) to extract the fundamental amplitude, 3-15 harmonic content ratio and phase offset features from the preprocessed data. The preprocessed signal is first divided into sliding windows (window length T, step size Δt), and discrete Fourier transform is performed on each window data. The formula is:

[0081] ;

[0082] in ,f s is the sampling frequency, X(f) is the frequency domain complex signal, which represents the amplitude and phase corresponding to the frequency component f. is the preprocessed current signal value at position k, T is the total number of sampling points in the sliding window, j is the imaginary unit, which is used to construct a complex exponential function to decompose the orthogonal frequency components of the signal. is a complex exponential basis function used to project the time domain signal onto an orthogonal basis of different frequencies;

[0083] Construct a multi-dimensional feature set including time domain (peak-to-peak value of waveform, rise time), frequency domain (harmonic energy distribution) and statistical domain (skewness, variance);

[0084] The model training unit uses the XGBoost algorithm, takes the ambient temperature and input voltage as input parameters, and the current value of each component as the output target to establish a nonlinear regression prediction model;

[0085] The model validation unit evaluates the model performance through cross-validation and uses the coefficient of determination R 2The model accuracy is verified by an index of ≥0.95, and the prediction residual is tested for white noise to ensure the model stability and generalization ability.

[0086] See Figure 3 As shown, the power supply fault rapid detection method based on the intelligent diagnosis engine includes:

[0087] Based on circuit topology analysis and functional relevance, the power supply is divided into multiple functional areas;

[0088] Based on historical power supply operation data, a current time series prediction model is built for each functional area. A diagnostic knowledge base containing component-level current baseline models is also built to train the current time series prediction model for each functional area.

[0089] Real-time collection of dynamic current signals from each functional area, input of real-time monitoring data into the training model, and generation of future current change trends for each component;

[0090] Based on the predicted current change trend, the real-time current waveform is matched with it and the similarity index is calculated;

[0091] Based on the similarity index, determine whether the current is abnormal;

[0092] Based on the fault situation, locate the fault source according to the components where the abnormal current is located.

[0093] Specifically, during the current waveform similarity calculation and anomaly determination process, the DTW distance between the real-time waveform and the predicted waveform is calculated through dynamic time warping. The formula is:

[0094] ;

[0095] Among them, D DTW is the dynamic time-warping distance between the real-time current waveform and the predicted waveform, For all possible regular paths In the example, we select the path that minimizes the cumulative squared difference. The measured current value of the real-time current waveform at the i-th time point, is the predicted current value of the current waveform at the jth time point;

[0096] And normalize the similarity, the formula is:

[0097] ;

[0098] Among them, S DTW is the normalized similarity;

[0099] Set the threshold S th (such as 0.8), if , it is determined to be abnormal, and the root mean square error between the real-time waveform and the baseline can be calculated by combining residual analysis:

[0100] ;

[0101] Where RMSE is the root mean square error between the real-time waveform and the baseline, T is the time window length, and I real (t+k) is the measured current value of the real-time current waveform at the t+kth time point, I base (t+k) is the baseline current value of the current waveform at the t+kth time point;

[0102] Based on regional correlation analysis, for example: if the anomaly occurs in the power conversion area, the related components include MOSFET, freewheeling diode, and high-frequency inductor.

[0103] See Figure 4 As shown in the figure, based on the predicted current change trend, the real-time current waveform is matched with it, and the similarity index is calculated, including:

[0104] Based on the predicted current change trend, the predicted current waveform time series data Y=(y1,y2,...,y m ), obtain real-time current waveform time series data X=(x1,x2,...x n ), y m is the mth predicted current waveform data, x n is the nth real-time current waveform data;

[0105] Based on the corresponding data points in the time series data, the distance between the two is calculated and a distance matrix is ​​constructed, in which each element stores x i and y j the distance between them;

[0106] Define the cumulative cost matrix, where each element stores the minimum cumulative distance from the starting point to the current point;

[0107] Based on the accumulated cost matrix after filling, backtrack from the lower right corner of the matrix to find the shortest path to the starting point;

[0108] Based on the obtained shortest path, obtain the value in the lower right corner of the cumulative cost matrix and perform normalization to obtain the similarity index. The smaller the value, the more similar the shapes and trends of the two time series data are.

[0109] Specifically, construct the distance matrix D, define the cumulative cost matrix C, and calculate the Euclidean distance for each predicted data point and real-time data point:

[0110] ;

[0111] Among them, d(i,j) is the Euclidean distance between the predicted data point and the real-time data point, p i For the predicted data point, r j For real-time data points;

[0112] The distance matrix D is:

[0113] ;

[0114] Where D is the distance matrix, N is the number of rows in the matrix, M is the number of columns in the matrix, and d(N,M) is the Euclidean distance between the predicted data point and the real-time data point;

[0115] Initialize the first element of the cumulative cost matrix C(1,1)=D(1,1). For the other elements C(i,j) in the matrix, calculate the minimum path from (i-1,j-1), (i-1,j), (i,j-1) and accumulate the current distance. The specific calculation method is:

[0116] C(i,j)=D(i,j)+min((i-1,j-1), (i-1,j), (i,j-1));

[0117] Among them, D(i,j) is the data in the i-th row and j-th column of the distance matrix D, C(i-1,j-1) is the previous point in the diagonal direction, C(i-1,j) is the previous point in the vertical direction, and C(i,j-1) is the previous point in the horizontal direction. Continue to fill the cumulative cost matrix until the lower right corner of the matrix;

[0118] Starting from the end point (N, M), work backwards to the starting point (1, 1)(1, 1), selecting the adjacent point that minimizes the cumulative cost at each step, and recording the sequence of aligned point pairs. , where K is the path length.

[0119] See Figure 5 As shown, based on the similarity index, determining whether the current current is abnormal specifically includes:

[0120] Based on the similarity index obtained in each time period, a function of similarity index change over time is constructed;

[0121] Based on the time-varying function of the similarity index, if the similarity shows a decreasing trend over time, it is determined that the current is abnormal and the power supply is faulty.

[0122] Specifically, under normal circumstances, the similarity should remain relatively stable. If the similarity shows a continuous downward trend, it may indicate a power failure. By setting a threshold and using a sliding window method to monitor changes in the similarity, when the similarity drops repeatedly and exceeds the threshold, the system will determine that there is an abnormality in the power supply and trigger an alarm.

[0123] Divide the time axis into fixed windows (e.g., every 10 minutes is a window), each window contains m similarity values, S k (k=1,2,...,m), and calculate the mean of the window, the formula is:

[0124] ;

[0125] Among them, t is the window center timestamp, S k is the similarity value, is the mean similarity, m is the number of similarity values;

[0126] The trend is quantified by the linear regression slope, and the time series Perform a linear fit:

[0127] ;

[0128] Among them, S(t) is the time series observation value at time t, and the slope Indicates the trend direction. If If the trend is statistically significant (e.g., p value < 0.05), it is considered to be a downward trend. is the intercept term, which represents the reference value when t=0, is the random error, representing fluctuations not captured by the model;

[0129] Compute the moving average difference of adjacent windows: If K times in a row , which was judged to be a continuous decline.

[0130] See Figure 6 As shown in the figure, based on the fault situation and the components where the abnormal current is located, locating the fault source specifically includes:

[0131] Based on the original current waveform within the abnormal time window, time-frequency domain feature extraction is performed to construct a feature vector;

[0132] Based on the acquired fault characteristics, the fault patterns recorded in the diagnostic knowledge base are matched to screen and obtain candidate fault types;

[0133] Construct a fault propagation directed graph, with components within the power supply functional area as nodes and edge weights assigned based on the fault transmission probability of the circuit topology;

[0134] Based on the predicted current waveform time series and the real-time current waveform time series as constraints, the Dijkstra algorithm is used to calculate the shortest propagation path to the abnormal component with the candidate fault source as the starting point;

[0135] Based on the measured abnormal characteristics and propagation paths, the posterior fault probability of each node is updated, and the component with the highest posterior probability is taken as the fault source, and its functional area and location coordinates are output.

[0136] Specifically, each component (such as MOSFET, capacitor, inductor) is a graph node , the directed graph is represented as:

[0137] G=(V,E), Faults may be transmitted to ;

[0138] Where G represents a graph, V is the set of vertices in the graph G, and E is the set of edges in the graph G;

[0139] In the process of updating the posterior fault probability and locating the fault source, the candidate fault source The prior probability of , the path weight reflects the possibility of fault transmission, and the likelihood function is defined as:

[0140] ;

[0141] in, is the probability of fault transfer, v k is the candidate fault source node, W i→j is the fault transfer weight of edge i→j, reflecting the possibility of fault propagation along this edge, Represents the weight W of all edges on path P i→j Perform multiplication;

[0142] Understandably, the fault patterns in the fault diagnosis knowledge base are based on historical data and experience. This may result in incomplete knowledge base coverage and variability in fault patterns, making it difficult to accurately match certain new or complex fault types. Regularly updating and expanding the fault pattern knowledge base, including more fault data samples, and using machine learning algorithms combined with real-time data for pattern learning and identification, can improve the accuracy of fault matching.

[0143] During the posterior fault probability update process, the probability distribution of the fault source may be inaccurate due to inaccurate prior data or uncertainty in the propagation path. Bayesian inference is used to update the posterior probability and incorporate more prior information to correct the probability estimate. To improve accuracy, multimodal data can be used to refine the posterior probability estimate.

[0144] It should be noted that the order in which the embodiments of the present invention are described above is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0145] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for rapid detection of power failures based on an intelligent diagnostic engine, characterized in that: include: Based on circuit topology analysis and functional relevance, the power supply is divided into multiple functional areas; Based on historical power supply operation data, a current time series prediction model is built for each functional area. A diagnostic knowledge base containing component-level current baseline models is also built to train the current time series prediction model for each functional area. Real-time collection of dynamic current signals from each functional area, input of real-time monitoring data into the training model, and generation of future current change trends for each component; Based on the predicted current change trend, the real-time current waveform is matched with it and the similarity index is calculated; Based on the similarity index, determine whether the current is abnormal; Based on the original current waveform within the abnormal time window, time-frequency domain feature extraction is performed to construct a feature vector; Based on the acquired fault characteristics, the fault patterns recorded in the diagnostic knowledge base are matched to screen and obtain candidate fault types; Construct a fault propagation directed graph, with components within the power supply functional area as nodes and edge weights assigned based on the fault transmission probability of the circuit topology; Based on the predicted current waveform time series and the real-time current waveform time series as constraints, the Dijkstra algorithm is used to calculate the shortest propagation path to the abnormal component with the candidate fault source as the starting point; Based on the measured abnormal characteristics and propagation paths, the posterior fault probability of each node is updated, and the component with the highest posterior probability is taken as the fault source, and its functional area and location coordinates are output.

2. The method for rapid detection of power failure based on intelligent diagnosis engine according to claim 1, characterized in that: The method of matching the real-time current waveform with the predicted current change trend and calculating the similarity index specifically includes: Based on the predicted current change trend, the predicted current waveform time series data is obtained. The sequence data Y=(y1,y2,...,y m ), obtain real-time current waveform time series data X=(x1,x2,...,x n ); Based on the corresponding data points in the time series data, the distance between the two is calculated and a distance matrix is ​​constructed, in which each element stores x i and y j the distance between them; Define the cumulative cost matrix, where each element stores the minimum cumulative distance from the starting point to the current point; Based on the accumulated cost matrix after filling, backtrack from the lower right corner of the matrix to find the shortest path to the starting point; Based on the obtained shortest path, obtain the value in the lower right corner of the cumulative cost matrix and perform normalization to obtain the similarity index. The smaller the value, the more similar the shapes and trends of the two time series data are.

3. The method for rapid detection of power failure based on intelligent diagnosis engine according to claim 1, characterized in that: The determining whether the current current is abnormal based on the similarity index specifically includes: Based on the similarity index obtained in each time period, a function of similarity index change over time is constructed; Based on the time-varying function of the similarity index, if the similarity shows a decreasing trend over time, it is determined that the current is abnormal and the power supply is faulty.

4. A power supply fault rapid detection system based on an intelligent diagnostic engine, for implementing a power supply fault rapid detection method based on an intelligent diagnostic engine as claimed in any one of claims 1 to 3, characterized in that: include: Functional area division module: The functional area division module is mainly used to divide the power supply into independent functional areas according to the functional relevance of each component in the power supply topology; Real-time data acquisition module: The real-time data acquisition module is equipped with a distributed current sensor array, which is mainly used to obtain current, voltage and temperature data of each functional area during the power supply operation stage; Current model building module: The current model building module is mainly used to establish a current time series prediction model based on historical operating data. The model includes the current baseline and allowable fluctuation threshold range of each component under standard operating conditions; Trend prediction module: The trend prediction module is mainly used to generate the current change trend of each component in the future by inputting real-time monitoring data into the training model; Abnormal judgment module: The abnormal judgment module is mainly used to identify abnormal situations based on the current change trend of each component in the future by matching the similarity with the real-time data; Fault location module: The fault location module is mainly used to match the current waveform under abnormal conditions with the waveform under other fault types, and locate the fault point by constructing a directed graph; Database module: The database module is mainly used to store real-time power parameter data, current model data, and abnormal fault data.

5. The power failure rapid detection system based on intelligent diagnosis engine according to claim 4 is characterized in that: The functional area division module specifically includes: Analyze component types, parameters, and connection relationships based on the circuit schematic of the power supply system; Analyze the flow of current and voltage signals and mark key nodes, establish the timing logic of component actions, and determine the functional dependency chain; Typical functional areas are divided into: input processing area, power conversion area, output voltage regulation area, control and protection area.

6. The power failure rapid detection system based on intelligent diagnosis engine according to claim 4 is characterized in that: The current model building module specifically includes: Based on historical power supply operation data, a current time series prediction model is constructed, including: Data preprocessing unit: The data preprocessing unit is mainly used to perform wavelet denoising and outlier removal on historical current data; Feature extraction unit: The feature extraction unit is mainly used to extract the fundamental component and harmonic distribution characteristics of the current signal using a sliding window Fourier transform; Model training unit: The model training unit is trained through an LSTM neural network and constructs a nonlinear regression model of the current of each component changing with the load; Model validation unit: The model validation unit is mainly used to ensure that each sub-model has a high goodness of fit through cross-validation.

Images