Fault detection algorithm based on multi-dimensional integrated power system

Abstract

The invention discloses a multi-dimensional integrated power system fault detection algorithm, and relates to the technical field of power system fault detection, and the algorithm comprises the steps: S1, data collection and preprocessing, S2, correlation analysis, S3, mean square error calculation, S4, extended Kalman filter application, S5, principal component analysis dimension reduction, and S6, fault detection and visualization. According to the fault detection algorithm based on the multi-dimensional integrated power system, multiple correlation analysis methods are integrated, the operation state of the power system is analyzed from multiple angles, the fault detection accuracy can be improved, different correlation analysis methods can capture different types of relations among power parameters, and the fault detection accuracy is improved through the comprehensive analysis result. By means of the method, faults in the system can be recognized more comprehensively and more accurately, the situation of misjudgment or missed judgment caused by limitation of a single analysis method is avoided, and therefore stable operation of the power system is guaranteed.

Description

Technical Field

[0001] The invention belongs to the technical field of power system fault detection, and in particular relates to a power system fault detection algorithm based on multi-dimensional integration. Background Art

[0002] In modern data centers, the stability and reliability of the power system are key factors in ensuring business continuity and data security. During the operation of the data center, any interruption or abnormality in the power supply will directly affect the normal operation of the server and network equipment, and thus have a significant impact on business operations. Therefore, timely monitoring and prevention of power system failures is an important task in data center operation and maintenance.

[0003] Traditional power system monitoring methods mainly rely on manual inspections and regular maintenance, but this method has problems such as poor real-time performance, low accuracy and low efficiency. With the continuous expansion of data center scale and the increase in equipment complexity, operation and maintenance personnel urgently need a more intelligent and efficient power system monitoring and fault warning method. To address the above problems, the following solutions are proposed. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-dimensional integrated power system fault detection algorithm, which solves the problems of poor real-time performance, low accuracy and low efficiency of the existing methods by designing a multi-dimensional integrated power system fault detection (MIPFDA) algorithm.

[0005] To solve the above technical problems, the present invention is achieved through the following technical solutions:

[0006] The present invention is a multi-dimensional integrated power system fault detection algorithm, the detection algorithm comprises the following steps:

[0007] Step S1, data collection and preprocessing: collect data containing various electrical parameters and time information from multiple data sources, perform multi-faceted cleaning, processing and preparation work, and generate a data quality report;

[0008] Step S2, correlation analysis: three correlation coefficient matrices, namely, Pearson, Spearman rank and cosine similarity, are calculated respectively, and the relationship between power parameters under normal and fault conditions is analyzed from the perspectives of linear relationship, monotonic relationship and high-dimensional data similarity;

[0009] Step S3, mean square error calculation: find the difference between the correlation matrix under normal and fault conditions, then calculate the mean square error, set a threshold and compare it with the calculated mean square error value;

[0010] Step S4, application of extended Kalman filter: defining a power system state space model, and performing real-time state estimation of power data through prediction and update steps;

[0011] Step S5, principal component analysis dimensionality reduction: after standardizing the data, calculate the covariance matrix and decompose it, select the principal component according to the eigenvalue, and project the original data to obtain the principal component score;

[0012] Step S6, fault detection and visualization: compare the changes in the correlation matrix to determine the changed variable pairs, draw the correlation matrix heat map and the principal component analysis scatter plot to intuitively display the data relationship and distribution, and output the fault detection detailed information by combining the results of each step;

[0013] Preferably, the step S2, correlation analysis, specifically comprises the following steps:

[0014] Step S21, calculate the Pearson correlation coefficient matrix:

[0015] Get the voltage, current and total output power data from the preprocessed data, and calculate the Pearson correlation coefficient matrix for the merged data:

[0016]

[0017] In the formula, X and Y are two variables. and are the average values ​​of X and Y respectively, and i is used to traverse the index of samples in the data set;

[0018] Step S22, calculate the Spearman rank correlation coefficient matrix:

[0019] By collecting relevant data, calculate the Spearman rank correlation coefficient matrix:

[0020]

[0021] Where d is the rank difference of the corresponding values ​​of two power parameters, n is the sample size, and l is used to traverse the index of the sample;

[0022] Step S23, calculate the cosine similarity matrix:

[0023] After obtaining the data, standardize the data and calculate the cosine similarity matrix:

[0024]

[0025] Where A and B are power parameter combinations for different time periods or devices, r is the index used to traverse the vector elements, and m represents the number of elements in the vector.

[0026] Preferably, the step S3, calculating the mean square error, specifically comprises the following steps:

[0027] Step S31: Calculate the difference between the correlation matrix in the normal state and the fault state:

[0028]

[0029] In the formula, and are the correlation coefficients of the sth row and jth column in normal and fault states, respectively. s and j are the indices used to traverse the elements of the correlation matrix, representing the rows and columns of the matrix, respectively.

[0030] Step S32: Calculate the mean square error based on the matrix difference:

[0031]

[0032] Where N is the dimension of the correlation matrix;

[0033] Step S33: setting a mean square error threshold, and comparing the calculated mean square error value with the threshold;

[0034] Preferably, the step S4, applying the extended Kalman filter, specifically comprises the following steps:

[0035] Step S41: defining a power system state space model;

[0036] Step S42: prediction step;

[0037] Step S43: updating step;

[0038] Step S44: performing real-time state estimation on voltage, current and total output power data;

[0039] Preferably, the formula of the power system state space model in step S41 is:

[0040] State transition equation: x k =f(x k-1 )+w k-1 ;

[0041] In the formula, x k is the state vector of the system at time step k, f(·) is the state transfer function, w k-1 is the process noise;

[0042] Observation equation: z k =h(x k )+v k ;

[0043] In the formula, z k is the observation vector, h(·) is the observation function, v k is the observation noise.

[0044] Preferably, the specific steps of step S42 are:

[0045] Step S421: Using the state estimation of the previous time step and the state transition equation predicts the state at the current time step

[0046] Step S422: Calculate the predicted covariance matrix:

[0047] In the formula, F k-1 is the Jacobian matrix of the state transfer matrix, P k-1 is the covariance matrix of the previous time step, is the matrix F k-1 The transpose of Q k-1 is the process noise covariance matrix.

[0048] Preferably, the specific steps of step S43 are:

[0049] Step S431: Using the observed value z of the current time step k Update the state estimate:

[0050]

[0051] In the formula, is the state estimation vector predicted at time step k, K k is the Kalman gain;

[0052] Step S432: Calculate Kalman gain

[0053] In the formula, is the covariance matrix predicted at time step k, is the matrix H k The transpose of H k is the Jacobian matrix of the observation matrix, R k is the observation noise covariance matrix;

[0054] Step S433: Update the covariance matrix

[0055] Where I is the identity matrix.

[0056] Preferably, the step S5, principal component analysis dimensionality reduction, specifically comprises the following steps:

[0057] Step S51: Obtain voltage, current and power data and perform standardization processing:

[0058]

[0059] In the formula, x′ is the data after standardization, x is the original data, μ is the data mean, and σ is the data standard deviation;

[0060] Step S52: Calculate the covariance matrix of the standardized data:

[0061]

[0062] In the formula, C is the covariance matrix, u is the number of samples, and x is a is the ath sample, is the sample mean;

[0063] Step S53: Perform eigenvalue decomposition C on the covariance matrix v =λv;

[0064] In the formula, λ is the eigenvalue and v is the eigenvector;

[0065] According to the size of the eigenvalue, the eigenvectors corresponding to the first L largest eigenvalues ​​are selected as the principal components. The number of principal components L is selected depending on the proportion of data variance that needs to be retained;

[0066] Step S54: Projecting the original data onto the selected principal component to obtain the principal component score y=XW;

[0067] Where y is the principal component score matrix, X is the standardized data matrix, and W is the selected eigenvector matrix.

[0068] The present invention has the following beneficial effects:

[0069] The multi-dimensional integrated power system fault detection algorithm of the present invention combines a variety of correlation analysis methods, analyzes the operating status of the power system from multiple angles, and can improve the accuracy of fault detection. Different correlation analysis methods can capture different types of relationships between power parameters. Through comprehensive analysis results, faults in the system can be identified more comprehensively and accurately, avoiding misjudgment or missed judgment due to the limitations of a single analysis method, thereby ensuring the stable operation of the power system.

[0070] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0072] Figure 1 This is an architecture diagram of a multi-dimensional integrated power system fault detection algorithm according to the present invention;

[0073] Figure 2A comparison diagram between the original data and the processed data of the present invention in the true state;

[0074] Figure 3 A comparison diagram between the original data and the processed data in the fa l se state of the present invention;

[0075] Figure 4 The present invention performs dimensionality reduction processing on the voltage, current and power data of the device AN8_UPS1 and extracts two principal components to draw a scatter plot;

[0076] Figure 5 This is a scatter plot drawn by extracting two principal components after dimensionality reduction processing on the voltage, current and power data of the device AN9_UPS4 in the present invention. DETAILED DESCRIPTION

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. 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.

[0078] See also Figure 1 As shown, the present invention is a fault detection algorithm based on multi-dimensional integrated power system, and the detection algorithm includes the following steps:

[0079] Step S1, data collection and preprocessing: collect data containing various electrical parameters and time information from the data source, perform various cleaning, processing and preparation work, and generate a data quality report;

[0080] Step S2, correlation analysis: three correlation coefficient matrices, namely, Pearson, Spearman rank and cosine similarity, are calculated respectively, and the relationship between power parameters under normal and fault conditions is analyzed from the perspectives of linear relationship, monotonic relationship and high-dimensional data similarity;

[0081] Step S3, mean square error calculation: find the difference between the correlation matrix under normal and fault conditions, then calculate the mean square error, set a threshold and compare it with the calculated mean square error value;

[0082] Step S4, application of extended Kalman filter: defining a power system state space model, and performing real-time state estimation of power data through prediction and update steps;

[0083] Step S5, principal component analysis dimensionality reduction: after standardizing the data, calculate the covariance matrix and decompose it, select the principal component according to the eigenvalue, and project the original data to obtain the principal component score;

[0084] Step S6, fault detection and visualization: compare the changes in the correlation matrix to determine the changed variable pairs, draw the correlation matrix heat map and the principal component analysis scatter plot to intuitively display the data relationship and distribution, and output the fault detection detailed information by combining the results of each step;

[0085] Step S2, correlation analysis specifically includes the following steps:

[0086] Step S21, calculate the Pearson correlation coefficient matrix:

[0087] Get the voltage, current and total output power data from the preprocessed data, and calculate the Pearson correlation coefficient matrix for the merged data:

[0088]

[0089] In the formula, X and Y are two variables. and are the average values ​​of X and Y respectively, and i is used to traverse the index of samples in the data set;

[0090] Step S22, calculate the Spearman rank correlation coefficient matrix:

[0091] By collecting relevant data, calculate the Spearman rank correlation coefficient matrix:

[0092]

[0093] Where d is the rank difference of the corresponding values ​​of two power parameters, n is the sample size, and l is used to traverse the index of the sample;

[0094] Step S23, calculate the cosine similarity matrix:

[0095] After obtaining the data, standardize the data and calculate the cosine similarity matrix:

[0096]

[0097] Where A and B are power parameter combinations for different time periods or devices, r is the index used to traverse the vector elements, and m represents the number of elements in the vector.

[0098] Step S3, mean square error calculation specifically includes the following steps:

[0099] Step S31: Calculate the difference between the correlation matrix in the normal state and the fault state:

[0100]

[0101] In the formula, and are the correlation coefficients of the sth row and jth column in normal and fault states, respectively. s and j are the indices used to traverse the elements of the correlation matrix, representing the rows and columns of the matrix, respectively.

[0102] Step S32: Calculate the mean square error based on the matrix difference:

[0103]

[0104] Where N is the dimension of the correlation matrix;

[0105] Step S33: setting a mean square error threshold, and comparing the calculated mean square error value with the threshold;

[0106] Step S4, application of the extended Kalman filter specifically includes the following steps:

[0107] Step S41: defining a power system state space model;

[0108] Step S42: prediction step;

[0109] Step S43: updating step;

[0110] Step S44: performing real-time state estimation on voltage, current and total output power data;

[0111] The formula of the power system state space model in step S41 is:

[0112] State transition equation: x k =f(x k-1 )+w k-1 ;

[0113] In the formula, x k is the state vector of the system at time step k, f(·) is the state transfer function, w k-1 is the process noise;

[0114] Observation equation: z k =h(x k )+v k ;

[0115] In the formula, z k is the observation vector, h(·) is the observation function, v k is the observation noise.

[0116] The specific steps of step S42 are:

[0117] Step S421: Using the state estimation of the previous time step and the state transition equation predicts the state at the current time step

[0118] Step S422: Calculate the predicted covariance matrix:

[0119] In the formula, F k-1 is the Jacobian matrix of the state transfer matrix, P k-1 is the covariance matrix of the previous time step, is the matrix F k-1 The transpose of Q k-1 is the process noise covariance matrix.

[0120] The specific steps of step S43 are:

[0121] Step S431: Using the observed value z of the current time step k Update the state estimate:

[0122]

[0123] In the formula, is the state estimation vector predicted at time step k, K k is the Kalman gain;

[0124] Step S432: Calculate Kalman gain

[0125] In the formula, is the covariance matrix predicted at time step k, is the matrix H k The transpose of H k is the Jacobian matrix of the observation matrix, R k is the observation noise covariance matrix;

[0126] Step S433: Update the covariance matrix

[0127] Where I is the identity matrix.

[0128] Step S5, principal component analysis dimensionality reduction specifically includes the following steps:

[0129] Step S51: Obtain voltage, current and power data and perform standardization processing:

[0130]

[0131] In the formula, x′ is the data after standardization, x is the original data, μ is the data mean, and σ is the data standard deviation;

[0132] Step S52: Calculate the covariance matrix of the standardized data:

[0133]

[0134] In the formula, C is the covariance matrix, u is the number of samples, and x isa is the ath sample, is the sample mean;

[0135] Step S53: Perform eigenvalue decomposition C on the covariance matrix v =λv;

[0136] In the formula, λ is the eigenvalue and v is the eigenvector;

[0137] According to the size of the eigenvalue, the eigenvectors corresponding to the first L largest eigenvalues ​​are selected as the principal components. The number of principal components L is selected depending on the proportion of data variance that needs to be retained;

[0138] Step S54: Projecting the original data onto the selected principal component to obtain the principal component score y=XW;

[0139] Where y is the principal component score matrix, X is the standardized data matrix, and W is the selected eigenvector matrix.

[0140] A specific application of this embodiment is:

[0141] Step S1, data collection and preprocessing: obtain data from the low-voltage cabinets and UPS measuring points in the data center. The data set covers the period from June 14, 2024 to July 8, 2024, including data of different equipment (such as AN4-1, AN4-2 low-voltage cabinets and UPS1-3, UPS1-4, etc.) in multiple areas (such as 5B14, 1A31, and 3A31 areas of Building No. 7) in normal operation, fault state, and unknown state. The data samples have different time lengths (3 days, 5 days, 7 days, and 10 days);

[0142] The main variables in the data set are time (Time, format is YYYY-MM-DDHH:MM:SS, sampling interval is 5 minutes), input voltage (Input_Ua_V, unit volt), input current (I nput_Ia_A, unit ampere) and total output power (Total_Power_output_KV, unit kilowatt). Each CSV file is named according to the acquisition device, data status and time length;

[0143] Clean the data, check and delete duplicate records, unify variable names, and handle obvious outliers (such as negative values ​​or data outside the normal range);

[0144] Ensure the continuity of timestamps, check for missing time points, and convert time columns to standard date and time formats; for missing values, short-term (less than 30 minutes) missing values ​​are filled using linear interpolation, and long-term missing values ​​are marked and paid special attention to in subsequent analysis;

[0145] Normalize the voltage, current, and power data to make the data comparable between different devices; split the data set into different subsets according to the data status and time length, and create corresponding data samples; use statistical methods (such as Z-score or IQR) to preliminarily mark potential outliers, do not delete them directly, and pay attention to them in subsequent analysis; generate a data quality report, including basic statistical information of each variable, the proportion of missing values, the number of outliers, etc.;

[0146] Step S2: Correlation analysis:

[0147] Step S21, calculate the Pearson correlation coefficient matrix:

[0148] The voltage, current and total output power data are obtained from the preprocessed data. For the merged data (normal state and fault state data are merged on the same time axis), the Pearson correlation coefficient matrix is ​​calculated. The formula is:

[0149]

[0150] In the formula, X and Y are two variables. and are the average values ​​of X and Y respectively, and i is used to traverse the index of samples in the data set;

[0151] Analyze the difference between the correlation coefficient matrix in normal state and fault state, and observe the change of correlation between variables. For example, the correlation between input current and total output power may be enhanced in fault state, and the correlation between voltage, current and power may change. If an abnormality occurs (such as abnormal correlation in normal state or the correlation matrix in fault state does not meet expectations), further check the data collection or processing process, re-evaluate the definition of fault state and collection method, and consider supplementing other detection methods or collecting more samples.

[0152] Step S22, calculate the Spearman rank correlation coefficient matrix:

[0153] Through the collected relevant data, the Spearman rank correlation coefficient matrix is ​​calculated, and the formula is:

[0154]

[0155] Where d is the rank difference of the corresponding values ​​of two power parameters, n is the sample size, and l is used to traverse the index of the sample;

[0156] The results of the coefficient matrix analysis are consistent with the trend of the Pearson correlation coefficient, but provide additional insights in some aspects; for example, the negative correlation between the input voltage and input current in the fault state of the 5B14 equipment group may be higher, and the correlation of some variables in the 1A31 equipment group changes (such as the correlation between input current and total output power increases, and the correlation between input voltage and input current decreases). It can also capture subtle changes that are ignored by the Pearson correlation coefficient, reflecting the importance of combining multiple correlation analysis methods;

[0157] Step S23, calculate the cosine similarity matrix:

[0158] After acquiring the data, first standardize the data (to make the data comparable under different capacity devices or load conditions), and calculate the cosine similarity matrix. The formula is:

[0159]

[0160] In the formula, A and B are power parameter combinations of different time periods or devices, r is the index used to traverse the vector elements, and m represents the number of elements in the vector;

[0161] Analyzing the matrix, the similarity between the input voltage and other variables in the 5B14 equipment group under fault conditions may decrease, while the similarity between the input current and the total output power increases; the 1A31 equipment group under fault conditions may experience a polarization phenomenon in the cosine similarity of variables (some increase sharply, some decrease sharply), which can capture subtle changes that are ignored by traditional correlation analysis and serve as a powerful tool to supplement other methods in building a fault detection framework;

[0162] Step S3, mean square error calculation:

[0163] Step S31: Calculate the difference between the correlation matrix in the normal state and the fault state, the formula is:

[0164]

[0165] In the formula, and are the correlation coefficients of the sth row and jth column in normal and fault states, respectively. s and j are the indices used to traverse the elements of the correlation matrix, representing the rows and columns of the matrix, respectively.

[0166] Step S32: Calculate the mean square error based on the matrix difference, the formula is:

[0167]

[0168] Where N is the dimension of the correlation matrix;

[0169] Step S33: Set the MSE threshold, compare the calculated MSE value with the threshold, if the MSE value exceeds the threshold, it is determined that the system has detected a fault and outputs the fault detection result; otherwise, it is considered that there is no obvious fault and monitoring continues;

[0170] Step S4, extended Kalman filter:

[0171] Step S41: define the power system state space model, state transition equation:

[0172] x k =f(x k-1 )+w k-1 ;

[0173] In the formula, x k is the state vector of the system at time step k, f(·) is the state transfer function, w k-1 is the process noise;

[0174] Observation equation: z k =h(x k )+v k ;

[0175] In the formula, z k is the observation vector, h(·) is the observation function, v k is the observation noise;

[0176] Step S42: prediction step, specifically:

[0177] Step S421: Using the state estimation of the previous time step and the state transition equation predicts the state at the current time step

[0178] Step S422: Calculate the predicted covariance matrix:

[0179] In the formula, F k-1 is the Jacobian matrix of the state transfer matrix, P k-1 is the covariance matrix of the previous time step, is the matrix F k-1 The transpose of Q k-1 is the process noise covariance matrix;

[0180] Step S43: Update step, specifically:

[0181] Step S431: Using the observed value z of the current time step k Update the state estimate:

[0182]

[0183] In the formula, is the state estimation vector predicted at time step k, K k is the Kalman gain;

[0184] Step S432: Calculate Kalman gain

[0185] In the formula, is the covariance matrix predicted at time step k, is the matrix H k The transpose of H k is the Jacobian matrix of the observation matrix, R k is the observation noise covariance matrix;

[0186] Step S433: Update the covariance matrix

[0187] Where I is the identity matrix;

[0188] Step S44: Real-time state estimation is performed on the voltage, current and total output power data, and the original data is compared with the data processed by EKF to observe the effect of EKF in smoothing noise and reducing errors. For example, in different states (true and false), EKF can effectively smooth the high-frequency noise in the original data, make the voltage, current and power data more stable, reduce fluctuations, improve data readability, and enhance the accuracy and reliability of fault detection;

[0189] Step S5: principal component analysis dimensionality reduction:

[0190] Step S51: Obtain voltage, current and power data and perform standardization processing. The formula is:

[0191]

[0192] In the formula, x′ is the data after standardization, x is the original data, μ is the data mean, and σ is the data standard deviation;

[0193] Step S52: Calculate the covariance matrix of the standardized data:

[0194]

[0195] In the formula, C is the covariance matrix, u is the number of samples, and x is a is the ath sample, is the sample mean;

[0196] Step S53: Perform eigenvalue decomposition C on the covariance matrix v =λv, where λ is the eigenvalue and v is the eigenvector;

[0197] According to the size of the eigenvalue, the eigenvectors corresponding to the first L largest eigenvalues ​​are selected as the principal components. The number of principal components L is selected depending on the proportion of data variance that needs to be retained;

[0198] Step S54: Projecting the original data onto the selected principal component to obtain the principal component score y=XW;

[0199] Where y is the principal component score matrix, X is the standardized data matrix, and W is the selected eigenvector matrix;

[0200] Through PCA dimensionality reduction, principal component analysis, and observation of data distribution differences between normal and fault states (data distribution is concentrated under normal conditions, and dispersed under fault conditions), this provides a basis for fault diagnosis and analysis;

[0201] Step S6: Fault detection and visualization:

[0202] Compare the changes in correlation matrices: Detailed comparison of the correlation matrices (Pearson, Spearman, cosine similarity) under normal and fault conditions, in-depth analysis of the changes in correlation between variables, and identification of variable pairs with significant changes. These changes are of great significance for determining the occurrence and impact of faults.

[0203] Draw a heat map of the correlation matrix: The correlation matrix under normal and fault conditions is intuitively displayed in the form of a heat map. The strength and change of the correlation between variables are intuitively presented through the depth of color, etc., helping engineers to quickly and clearly understand the system operation status and potential problems;

[0204] Draw PCA scatter plot: Draw a scatter plot after principal component analysis to show the distribution characteristics of the data after dimensionality reduction, compare the scatter distribution patterns under normal and fault conditions, effectively identify changes in equipment operating status, and provide strong support for fault diagnosis;

[0205] Output fault detection results: Based on comprehensive information such as the comparison results of the MSE value and the threshold value, and the analysis of the change in the correlation matrix, the fault detection results are accurately output, including detailed information such as whether a fault is detected, the variables and equipment that may be involved in the fault, etc., to provide a clear decision-making basis for power system operation and maintenance; at the same time, the fault detection results are intuitively displayed through visualization methods (such as correlation matrix heat maps, PCA scatter plots, etc.), making it easier for engineers to understand and deeply analyze system fault conditions, and to formulate targeted maintenance strategies;

[0206] Test verification:

[0207] Through 15 sets of experiments, the purpose is to comprehensively evaluate the fault detection methods of UPS and low-voltage wiring cabinets in power systems. The experimental design follows the following logical progression:

[0208] 1. Basic comparison (experiments 1-3):

[0209] Purpose: To establish a baseline comparison between fault and normal conditions;

[0210] Logic: Gradually extend the observation window from 3 days to 10 days to evaluate the impact of the time span on fault detection;

[0211] 2. Normal state analysis (experiments 4-6):

[0212] Purpose: To gain in-depth understanding of the characteristics of different devices under normal operating conditions;

[0213] Logic: Compare the two groups of equipment, 5B14 and 1A31, and extend the observation time from 3 days to 10 days to explore the differences between the equipment and the influence of time factors;

[0214] 3. Fault status analysis (experiments 7-9):

[0215] Purpose: To study the behavior patterns of different devices under fault conditions;

[0216] Logic: Also compare 5B14 and 1A31, observe the time from 3 days to 10 days, and analyze how the fault evolves over time;

[0217] 4. Analysis of unknown state (experiments 10-12):

[0218] Purpose: To test the effectiveness of the method under unknown conditions;

[0219] Logic: Use the 3A31 equipment group from 3 days to 10 days to verify the performance of the algorithm in the absence of prior information;

[0220] 5. Data anomaly verification (experiment 13):

[0221] Purpose: To verify the cause of abnormal data in 5B14 equipment group;

[0222] Logic: Use the updated data set to reanalyze the fault and normal status of the 7-day window;

[0223] 6. Short time window analysis (experiments 14-15):

[0224] Purpose: To evaluate the effectiveness of the method over a shorter time frame;

[0225] Logic: shorten the time window to 5 days, test the 5B14 and 1A31 equipment groups separately, and introduce the Multi-dimensional Integrated Power System Fault Detection Algorithm (MIPFDA);

[0226] The experimental design sequence shows a systematic exploration process from basic comparison to complex scenarios. By gradually increasing the complexity and introducing new variables (such as different device types, time window lengths, and unknown states), the robustness and adaptability of the proposed method can be fully evaluated. Experiments (14-15) reflect the response to the demand for fast fault detection in practical applications, and also provide a platform for the verification of new algorithms.

[0227] The above progressive experimental design can deeply study the advantages and disadvantages of each method, and also help understand their applicability in different situations, laying a solid foundation for developing more effective and reliable fault detection systems;

[0228] Result analysis:

[0229] Pearson correlation coefficient analysis results:

[0230] In experiments 1-3 and 13, it was observed that there were significant differences between the fault state and the normal state; in experiment 13, using the updated data, the 5B14 device group still showed significant differences before and after the fault, but the correlation coefficient in the normal state was still abnormal;

[0231] This result suggests that further investigation is needed to determine whether the data anomaly in the 5B14 device group is due to problems with the device itself or errors in the data collection process;

[0232] Usually, the correlation between variables changes significantly under expected fault conditions, but in the subsequent experiment 4-14, the correlation matrices of AN81 and UPS11, as well as AN92 and UPS14 in the 1A31 room, show high similarity in both fault and normal conditions, which is not common in actual systems;

[0233] Secondly, under the fault state, it is observed that the correlations between many variables are abnormally strong, almost close to perfect correlation (1 or -1); it is worth noting that the input current (I nput_la_A) and the total output power (Total_Power_output_KV) always show perfect correlation (1.0) under the fault state; in addition, the correlation between the voltage (i nput_Ua_V) and the current and power is stronger under the fault state than under the normal state; such a high correlation is unlikely to persist in a normally operating system; and the correlation matrix under the fault state is almost exactly the same between the two different groups of equipment (AN8_1 / UPS1_1 and AN9_2 / UPS1_4); such a high degree of consistency is rare in actual situations, because different equipment usually show some slight differences;

[0234] In contrast, the correlation matrices under normal conditions are more reasonable; they show more diverse and moderate correlations between variables, which is more in line with expectations;

[0235] The above observations reflect that there may be some potential problems; there may be some systematic errors in the data collection or processing process, especially for the data of fault status; another possibility is that the "fault status" may not represent the real fault situation, or the fault simulation method may not be accurate enough, resulting in such abnormally consistent results; in view of this situation, re-examine the entire data collection and processing process; first, carefully check the raw data to ensure that no errors were introduced during the data processing or merging process; second, re-evaluate the definition of fault status and data collection methods to ensure that they accurately reflect the real fault situation; in addition, consider using other fault detection methods or indicators to supplement the Pearson correlation coefficient analysis to obtain a more comprehensive perspective; finally, it may be necessary to collect more samples or different types of fault data to verify whether this phenomenon is universal or limited to this specific data set;

[0236] Overall, these results, while unexpected, provide important insights; they remind us that more caution is needed in processing and interpreting the data, while also providing directions for further improving our fault detection methods;

[0237] When analyzing the four devices of 5B13 in the 13th experiment, this new set of Pearson correlation coefficient matrices showed more reasonable and expected results; there were obvious differences between the normal state and the fault state, which was consistent with our expectations of the impact of the fault on the system behavior; it was particularly noteworthy that the correlation between some variables changed significantly under the fault state, such as the increase in the correlation between AN4 and the input voltage of the UPS, which may indicate an abnormal synchronous change in voltage caused by the fault;

[0238] The more negative correlations that occur under normal conditions, especially between input voltage and input current and output power, are consistent with the basic principles of power systems; this complex correlation pattern, coupled with the slight differences between the two sets of data in AN41 / UPS1_3 and AN4_2 / UPS1_4, is closer to the characteristics of the actual system;

[0239] Under fault conditions, the enhancement of some correlations (such as input current and output power) and the weakening of other correlations reflect the specific impact of the fault on system operation; the distribution of correlation coefficients is reasonable, without extreme cases of complete correlation or no correlation, which further confirms the credibility of the data;

[0240] The above observations not only provide a reliable basis for fault detection, but also provide valuable insights into the behavioral characteristics of the system under different states; this analysis method demonstrates its potential in identifying system anomalies and guiding fault diagnosis;

[0241] Spearman rank correlation coefficient analysis results:

[0242] Spearman rank correlation coefficient analysis provides another angle to observe the relationship between variables, especially when dealing with data that may have nonlinear relationships; the results of the Spearman rank correlation coefficient are generally consistent with the trend of the Pearson correlation coefficient, but show unique insights in some key aspects; in Experiments 13-15, the Spearman rank correlation coefficient also effectively reflects the obvious differences before and after the failure in these experiments, but the characteristics it shows are slightly different from those of the Pearson correlation coefficient;

[0243] For the 5B14 device group, the Spearman rank correlation coefficient has a stronger correlation between variables under fault conditions; for example, the negative correlation between input voltage and input current becomes more obvious under fault conditions; this change suggests that under fault conditions, the relationship between voltage and current becomes more complex and may involve nonlinear interactions;

[0244] For the analysis results of the 1A31 equipment group, under normal operating conditions, the Spearman rank correlation coefficient shows that there is a moderate correlation between the variables; under fault conditions, the correlation between some variable pairs (such as input current and total output power) is significantly enhanced, while the correlation between other variable pairs (such as input voltage and input current) is weakened; this pattern change may reflect how the fault changes the dynamic balance within the system;

[0245] It is worth noting that the Spearman rank correlation coefficient captures subtle changes that the Pearson correlation coefficient may miss in some cases; for example, a pair of variables that is weakly correlated by the Pearson correlation coefficient under normal conditions may show a monotonic relationship of moderate strength in the Spearman rank correlation analysis; this difference highlights the importance of using multiple correlation analysis methods, because they can jointly provide a more comprehensive understanding of the data;

[0246] For the 5-day time window, we can observe how the fault evolves over time and affects the relationship between system variables; by comparing the Spearman rank correlation matrix at different time points, we can track the dynamic changes of the correlation, which is crucial for understanding the progression of the fault and formulating timely maintenance strategies;

[0247] The Spearman rank correlation coefficient analysis not only validates the findings of the Pearson correlation coefficient, but also provides additional insights, especially when dealing with data that may have nonlinear relationships; the robustness of this analysis method makes it a powerful tool for power system fault detection and diagnosis, especially when performing rapid assessments within a short time window; by combining the Pearson and Spearman methods, we are able to build a more comprehensive and reliable fault detection framework, providing solid data support for preventive maintenance and system optimization;

[0248] Cosine Similarity Analysis Results Cosine similarity analysis can provide another method to evaluate the relationship between power system variables, which is applicable to high-dimensional data space; in Experiments 13-15, this method demonstrated its effectiveness in distinguishing normal and fault states within a short time window;

[0249] For the 5B14 device group, the cosine similarity analysis reveals a significant decrease in the similarity between the input voltage and other variables under the fault state, while the similarity between the input current and the total output power increases; this reflects how the fault changes the pattern of energy flow in the system;

[0250] The results for the 1A31 device group are more prominent; under fault conditions, a polarization phenomenon can be observed: the cosine similarities of some pairs of variables increase sharply, while others decrease significantly; this change may indicate the nonlinear response of the system under fault conditions;

[0251] It is worth noting that cosine similarity is able to capture some subtle changes that may be overlooked in traditional correlation analysis; this property makes it a powerful tool to supplement Pearson and Spearman methods and help build a more comprehensive fault detection framework;

[0252] Cosine similarity analysis not only verifies the findings of the first two methods, but also provides unique insights, especially in dealing with high-dimensional data and capturing nonlinear relationships. This multi-angle analysis method can provide an opportunity to gain a deeper understanding of system behavior, which is of great significance for preventive maintenance and system optimization.

[0253] Data visualization results:

[0254] Under normal operating conditions, both the 5B14 and 1A31 equipment groups exhibited relatively stable operating modes; the voltage of the 5B14 equipment group fluctuated regularly between 220 and 235 V, showing obvious periodic changes, which may reflect daily load changes; the current and power also showed similar periodic fluctuations, but the fluctuation range was relatively small, between 200-220 A and 130-150 kW respectively; the 1A31 equipment group also showed similar periodic fluctuations under normal conditions, but the overall fluctuation range was slightly larger than that of 5B14, especially in terms of current and power, which may indicate that the load fluctuations borne by 1A31 were larger;

[0255] Under fault conditions, the 5B14 equipment group showed obvious abnormalities. The voltage dropped significantly from the normal 220-235V to around 200-220V, and the fluctuations became more intense. The more obvious abnormality was the current, which rose sharply during the fault period, from the normal 200-220A to 230-240A, and even peaked at more than 240A. The power also showed instability under fault conditions, with the overall level decreasing and the fluctuations intensifying, with the power falling below 130kW on several occasions. These phenomena collectively pointed to possible short circuits or equipment overloads.

[0256] In contrast, the changes in the 1A31 equipment group under fault conditions were relatively small; although the voltage, current, and power all experienced a certain degree of fluctuation increase, there was no drastic change like that of 5B14; the average levels of current and power increased slightly, but were still within an acceptable range, which may indicate that 1A31 experienced a slight load increase or system fluctuation rather than a serious fault;

[0257] By comparing the two equipment groups, obvious differences can be observed. Fault severity: 5B14 showed more serious abnormalities in the fault state, especially the sharp rise in current and unstable drop in power; while 1A31's changes were relatively mild, just experiencing a minor fault or load fluctuation; Voltage stability: The voltage of 5B14 dropped significantly and became unstable in the fault state, while the voltage change of 1A31 was relatively small;

[0258] Load characteristics: Even in normal conditions, the current and power fluctuations of 1A31 are more obvious than those of 5B14, which reflects that the two equipment groups serve different types or sizes of loads;

[0259] Failure mode: The fault characteristics of 5B14 clearly point to short circuit or severe overload, which is manifested by a sharp increase in current and unstable power; while the changes of 1A31 are more like system fluctuations or slight load changes, without extreme values;

[0260] This detailed time series analysis not only reveals the operating characteristics of the two equipment groups under normal and fault conditions, but also provides an important basis for fault diagnosis and preventive maintenance; the abnormality of 5B14 requires immediate attention and processing, while the slight change of 1A31 requires further monitoring to prevent the deterioration of potential problems;

[0261] Voltage distribution comparison:

[0262] For the 5B14 equipment group, the voltage distribution in the normal state is relatively concentrated, with a median of about 220V and a small interquartile range, indicating that the voltage is stable; in the fault state, the voltage distribution drops significantly, with the median dropping to about 200V, and the interquartile range increases significantly, indicating increased voltage instability; this large difference may indicate a voltage-related fault; for the 1A31 equipment group, the difference in voltage distribution in the normal and fault states is not as obvious as that of the 5B14; the medians in both states are around 220V, but the interquartile range in the fault state is slightly larger, indicating increased voltage fluctuations;

[0263] Current distribution comparison:

[0264] The 5B14 device group showed obvious current anomalies in the fault state. In the normal state, the current distribution was concentrated around 200-220A, while in the fault state, the current rose sharply, the median exceeded 230A, and a large number of outliers appeared, some even exceeded 240A. This drastic current increase may indicate a short circuit or other serious electrical faults. The current distribution of the 1A31 device group changed relatively little. In the fault state, the median of the current distribution increased slightly, and the interquartile range also increased, indicating that the current fluctuation increased, but the change was not as drastic as that of 5B14.

[0265] Power distribution comparison:

[0266] The power distribution of the 5B14 equipment group changed significantly under the fault state; in the normal state, the power distribution was concentrated around 140-150kW, while in the fault state, the power distribution dropped significantly, the median dropped to about 130kW, and the distribution range expanded significantly, with more low-power abnormal values; this indicates that the equipment cannot work normally under the fault state, resulting in unstable or reduced power output; the power distribution of the 1A31 equipment group changed relatively little; in the fault state, the median of the power distribution increased slightly, and the distribution range also increased, but the change was not as obvious as 5B14;

[0267] The 5B14 device group showed more obvious abnormalities in the fault state, especially in current and power, which may indicate a more serious fault condition; while the 1A31 device group also showed some fault characteristics, the changes were relatively small, which may be caused by a minor fault or system fluctuation; these observations are consistent with the experimental record that "the 5B14 results are as expected, but the 1A31 core area has a correlation of all 1s, and the reason is unknown", reflecting the different performances of the two device groups in the fault state; this detailed box plot analysis provides an important visual basis for fault diagnosis and system performance evaluation, and helps to identify potential problem areas and abnormal patterns; by comparing the data distribution of different device groups and under different operating conditions, we can better understand the normal operating range and abnormal behavior characteristics of the system, providing an important basis for preventive maintenance and fault early warning;

[0268] Comprehensive comparison:

[0269] Fault impact: The 5B14 device group showed more obvious changes under fault conditions, especially a sharp increase in current and increased power instability; in contrast, the changes in the 1A31 device group were relatively mild;

[0270] Distribution characteristics: The distribution of parameters of 5B14 is more concentrated under normal conditions, while 1A31 shows a wider distribution range even under normal conditions, which may reflect the different load characteristics of the two equipment groups; Failure mode: The failure characteristics of 5B14 clearly point to possible short circuit or severe overload, while the changes of 1A31 are more like slight system fluctuations or load changes;

[0271] System stability: The parameter distribution changes of 1A31 in normal and fault conditions are relatively small, indicating that this equipment group may have better stability or stronger anti-interference ability;

[0272] This detailed distribution analysis not only reveals the performance characteristics of the two equipment groups under different operating conditions, but also provides an important basis for fault diagnosis and system optimization; the abnormality of 5B14 requires immediate attention and processing, while the slight change of 1A31 requires further monitoring to prevent the deterioration of potential problems;

[0273] Insights from data visualization methods:

[0274] For UPS and low voltage wiring cabinets:

[0275] Heat maps help to visually display the correlation changes between different electrical parameters; for example, we can clearly see how the correlation between input voltage, output voltage, battery voltage and load current changes before and after a UPS failure; box plots can effectively display abnormal voltage and current fluctuations; for example, in a low-voltage wiring cabinet, if the current box plot of a phase shows abnormally high or low values, it may indicate that the phase has an overload or circuit breaker problem;

[0276] Line charts are very useful for showing instantaneous changes when UPS switches to battery mode. We can clearly see the instantaneous fluctuations in voltage and current when power is switched, helping to identify problems such as unsmooth switching or insufficient battery capacity. Violin charts can show subtle changes in voltage and current distribution. For example, in a normally operating low-voltage wiring cabinet, the voltage distribution should be a symmetrical bell shape. If the distribution becomes skewed or has double peaks, it may indicate voltage instability or harmonic interference problems.

[0277] Practical impact on power system monitoring, maintenance and optimization:

[0278] Application of UPS and low voltage wiring cabinet:

[0279] Accurate fault detection: By combining multiple analysis methods, different UPS fault types such as battery failure, inverter failure or bypass failure can be distinguished more accurately; for example, a battery failure may appear as a sudden drop in battery voltage with correlation to other parameters;

[0280] Preventive maintenance: By monitoring the long-term trend of each phase current in the low-voltage wiring cabinet, possible overload problems can be predicted, and load balancing or equipment upgrades can be performed in advance to avoid failures;

[0281] System optimization: Based on the performance analysis of different UPS models (such as 5B14 and 1A31) under various load conditions, the UPS configuration and load distribution can be optimized to improve overall efficiency and reliability;

[0282] Improved real-time monitoring: New algorithms can more sensitively detect changes in the UPS battery health status, predict battery life and schedule replacement in a timely manner by analyzing small anomalies in the charge and discharge cycles;

[0283] Customized strategies: Specific monitoring values ​​and alarm strategies can be formulated for different UPS models and low-voltage wiring cabinets. For example, for the 1A31 model, you may need to pay more attention to the fluctuation of input current, while for the 5B14 model, you may need to monitor the stability of output voltage.

[0284] In-depth fault analysis: By comparing the data characteristics of UPS in normal operation and fault state, a more accurate fault model can be established; this helps to quickly diagnose problems, such as distinguishing between UPS faults and external power grid problems; these applications not only improve the reliability and efficiency of UPS and low-voltage wiring cabinet systems, but also provide smarter and more proactive solutions for power management in data centers; through these methods, unexpected downtime can be significantly reduced, equipment life can be extended, and energy use can be optimized, thereby reducing operating costs and improving overall system performance;

[0285] Analysis of Multidimensional Integrated Power System Fault Detection Algorithm (MIPFDA):

[0286] The analysis of the 5B14 equipment group shows that there are obvious correlation changes between almost all parameters, which indicates that the fault has a wide impact on the operating status of the entire system; in particular, the correlation between the AN input voltage (I nput_Ua_V_AN) and other parameters has changed most significantly, which may be the core cause or main manifestation of the fault; from the comparison of the correlation matrix, there is a strong correlation between the parameters in the normal state, especially between the corresponding parameters of AN and UPS; however, in the fault state, the overall correlation is reduced, especially the correlation between the AN input voltage and other parameters is significantly weakened; it is worth noting that the correlation change of UPS related parameters in the fault state is relatively small, which indicates that the UPS system may have a certain fault isolation capability; the mean square error (MSE) value is 0.045, which is relatively high, further confirming the obvious change of the system parameter relationship in the fault state;

[0287] In contrast, the analysis results of the 1A31 equipment group show that the correlation changes less, mainly concentrated between the input voltage, input current and total output power of AN and UPS; this indicates that the fault of 1A31 may be more localized, mainly affecting the relationship between voltage and current; from the comparison of the correlation matrix, there is already a high correlation between the parameters of 1A31 in the normal state; in the fault state, the correlation between some parameters is further enhanced, especially between the input current and total output power of AN; the correlation of UPS related parameters in the fault state does not change much, or even increases, which may indicate that the UPS system plays a stabilizing role in the fault state; the mean square error (MSE) value of 1A31 is 0.038, slightly lower than 5B14, indicating that the overall change of 1A31 in the fault state is relatively small;

[0288] Comparing the results of the two equipment groups, the following conclusions can be drawn: the fault of 5B14 seems to be more serious and comprehensive, affecting the relationship between almost all parameters; while the fault of 1A31 is relatively localized, mainly affecting the voltage and current parameters, with less impact on power; although the fault was detected in both equipment groups, the fault characteristics of 5B14 are more obvious; it is worth noting that the UPS system showed a certain stability in both equipment groups, which may have played a role in fault mitigation;

[0289] The Multi-dimensional Integrated Power System Fault Detection algorithm (MIPFDA) successfully detected the faults of the two equipment groups and revealed the different characteristics of the faults through correlation analysis; this analysis method provides valuable information for fault diagnosis and preventive maintenance; based on these results, it can be inferred that 5B14 may need a more comprehensive inspection and repair, while 1A31 may only need to check the voltage and current systems; this precise analysis helps to develop a more targeted maintenance strategy and improve the overall reliability and efficiency of the system;

[0290] The extended Kalman filter (EKF) is applied to power system data analysis, which can provide smoother and more accurate results through real-time state estimation. By comparing the original data with the data processed by EKF, the effect of EKF in smoothing noise and reducing errors is observed.

[0291] like Figure 2 and Figure 3 As shown in the figure, the performance of the device AN8UPS1 in the true and false states is compared and analyzed respectively; for InputUa_VAN and InputUa_VAN(AN8UPS1true), the comparison between the original voltage data and the voltage data processed by EKF is seen; EKF effectively smoothes the high-frequency noise in the original data, making the voltage data smoother and reducing fluctuations; at certain time points (such as where the voltage drops sharply in the figure), EKF can provide a more stable voltage estimate;

[0292] The voltage, current and total output power data are processed by the extended Kalman filter (EKF), and the smoothing effect of the data can be clearly seen; the EKF effectively reduces the high-frequency noise and mutations in the original data, providing a more stable and accurate state estimation; this method not only improves the readability of the data, but also enhances the accuracy and reliability of fault detection; especially in the comparison between different states (true and false), the EKF shows obvious smoothing effect and noise suppression ability in all cases; the voltage, current and power data are reduced in dimension by principal component analysis (PCA), two principal components are extracted and scatter plots are drawn, refer to Figure 4 and Figure 5, which can help us understand the distribution of data after dimensionality reduction;

[0293] Through PCA analysis, we can observe obvious differences in data distribution under normal state (true) and fault state (failure); under normal state, the data distribution of AN and UPS is relatively concentrated and similar, indicating that the equipment is running stably and the relationship between parameters is relatively consistent; while under fault state, the data distribution is obviously dispersed, and the data patterns of AN and UPS are obviously different, indicating that the equipment is running abnormally and there may be potential faults or problems;

[0294] These PCA results intuitively show the distribution of data after dimensionality reduction, which helps to identify changes in equipment operating status and provide a basis for improved fault diagnosis and analysis. By comparing the principal component distributions under normal and fault conditions, changes in equipment operating status and their impact can be more clearly understood;

[0295] Comparison of the new algorithm (Multi-dimensional Integrated Power System Fault Detection Algorithm (MIPFDA)) with traditional algorithms (Pearson, Spearman, Cosine Similarity):

[0296] The application of the multi-dimensional integrated power system fault detection algorithm (MIPFDA) in the power operation and maintenance system can significantly improve the accuracy and efficiency of fault detection; compared with traditional fault detection methods, this algorithm has obvious advantages and beneficial effects in many aspects;

[0297] First, MIPFDA comprehensively analyzes the operating status of the equipment from multiple angles by using the Pearson correlation coefficient, Spearman rank correlation coefficient and cosine similarity, significantly improving the accuracy of fault detection. The extended Kalman filter (EKF) provides smoother data in real-time state estimation, reduces noise and errors, and further improves the reliability of detection results. The traditional correlation coefficient test method is difficult to quantify the current and voltage standards when dealing with equipment of different brands and models, but MIPFDA overcomes this challenge.

[0298] Secondly, MIPFDA combines multiple analysis methods to comprehensively analyze power system data by calculating correlation matrix, mean square error (MSE), EKF and principal component analysis (PCA). This multi-dimensional integrated approach can identify more complex and subtle fault modes and provide a more comprehensive system status assessment. In contrast, traditional methods have poor adaptability to equipment types and operating conditions and have difficulty handling complex nonlinear relationships.

[0299] In addition, MIPFDA can handle nonlinear systems and is applicable to different types of power equipment and multiple operating states. Through PCA dimension reduction, MIPFDA can effectively reduce data dimensions, retain important information, and improve analysis efficiency and accuracy. Traditional methods are prone to false positives or false negatives when faced with interference factors such as electromagnetic interference, ambient temperature changes, load fluctuations, power equipment aging, voltage sags, and transient overvoltages. MIPFDA, by combining multiple analysis methods, can effectively filter out normal fluctuations, highlight real abnormal changes, and enhance the system's anti-interference ability.

[0300] Finally, M IPFDA provides a wealth of visualization tools. By drawing correlation matrix heat maps and PCA scatter plots, fault detection results are more intuitive and easy to understand. This visualization and easy explanation helps engineers quickly locate problems and develop maintenance plans.

[0301] Compared with competitors' products, MIPFDA integrates multiple analysis methods, which is more accurate and reliable than competitors' single methods and can detect potential faults earlier; it can provide a more comprehensive system status assessment, cover a wider range of fault modes, and is applicable to various types of power equipment and multiple operating states, with stronger versatility; through the combination of multiple analysis methods, MIPFDA has stronger anti-interference ability in the face of various interference factors, reducing false alarms and missed alarms; in addition, MIPFDA provides detailed visualization tools to make fault detection results more intuitive and easy to interpret, which is convenient for engineers to analyze and make decisions;

[0302] In conclusion, the multi-dimensional integrated power system fault detection algorithm (M IPFDA) has brought significant improvements and enhancements to the power operation and maintenance system through its high accuracy, comprehensiveness, strong adaptability, strong anti-interference ability and good visualization effect, giving it a clear advantage in market competition;

[0303] Conclusions and Findings:

[0304] This study analyzed the power system data of two equipment groups, 5B14 and 1A31, and used the newly developed multi-dimensional integrated power system fault detection algorithm (MIPFDA) and traditional correlation analysis methods to obtain several important findings. First, the new algorithm demonstrated the ability to effectively detect abnormal conditions in the power system and reveal the characteristics and impact range of the fault through changes in the correlation matrix. In the analysis of specific equipment groups, the 5B14 equipment group showed more obvious and comprehensive parameter correlation changes under fault conditions, especially the relationship between input voltage and other parameters, which may indicate a more serious system fault. In contrast, the fault performance of the 1A31 equipment group was relatively localized, mainly affecting The voltage and current parameters indicate that different equipment groups may face different types or degrees of faults; it is worth noting that the UPS system showed a certain stability in both equipment groups and maintained a high parameter correlation under fault conditions, showing its important role in fault mitigation; this finding emphasizes the key position of the UPS system in power system reliability maintenance; in addition, by comparing with traditional correlation analysis methods (Pierce, Spearman, cosine similarity), the multi-dimensional integrated power system fault detection algorithm (MIPFDA) shows obvious advantages in capturing dynamic system changes and identifying specific fault characteristics, providing a more comprehensive and accurate tool for power system fault diagnosis.

[0305] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0306] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. A multi-dimensional integrated power system fault detection algorithm, characterized in that: The detection algorithm comprises the following steps: Step S1, data collection and preprocessing: collect data containing various electrical parameters and time information from the data source, perform various cleaning, processing and preparation work, and generate a data quality report; Step S2, correlation analysis: three correlation coefficient matrices, namely, Pearson, Spearman rank and cosine similarity, are calculated respectively, and the relationship between power parameters under normal and fault conditions is analyzed from the perspectives of linear relationship, monotonic relationship and high-dimensional data similarity; Step S3, mean square error calculation: find the difference between the corresponding matrices under normal and fault conditions, then calculate the mean square error, set a threshold and compare it with the calculated mean square error value; Step S4, application of extended Kalman filter: defining a power system state space model, and performing real-time state estimation of power data through prediction and update steps; Step S5, principal component analysis dimensionality reduction: after standardizing the data, calculate the covariance matrix and decompose it, select the principal component according to the eigenvalue, and project the original data to obtain the principal component score; Step S6, fault detection and visualization: compare the corresponding matrix changes to determine the changed variable pairs, draw the correlation matrix heat map and principal component analysis scatter plot to intuitively display the data relationship and distribution, and output the fault detection detailed information based on the results of each step.

2. The multi-dimensional integrated power system fault detection algorithm according to claim 1 is characterized in that: The step S2, correlation analysis specifically includes the following steps: Step S21, calculate the Pearson correlation coefficient matrix: Get the voltage, current and total output power data from the preprocessed data, and calculate the Pearson correlation coefficient matrix for the merged data: In the formula, X and Y are two variables. and are the average values ​​of X and Y respectively, and i is used to traverse the index of samples in the data set; Step S22, calculate the Spearman rank correlation coefficient matrix: By collecting relevant data, calculate the Spearman rank correlation coefficient matrix: Where d is the rank difference of the corresponding values ​​of two power parameters, n is the sample size, and l is used to traverse the index of the sample; Step S23, calculate the cosine similarity matrix: After obtaining the data, standardize the data and calculate the cosine similarity matrix: Where A and B are power parameter combinations for different time periods or devices, r is the index used to traverse the vector elements, and m represents the number of elements in the vector.

3. The multi-dimensional integrated power system fault detection algorithm according to claim 1 is characterized in that: The step S3, calculating the mean square error, specifically comprises the following steps: Step S31: Calculate the difference between the correlation matrix in the normal state and the fault state: In the formula, and are the correlation coefficients of the sth row and jth column in normal and fault states, respectively. s and j are the indices used to traverse the elements of the correlation matrix, representing the rows and columns of the matrix, respectively. Step S32: Calculate the mean square error based on the matrix difference: Where N is the dimension of the correlation matrix; Step S33: Setting a mean square error threshold, and comparing the calculated mean square error value with the threshold.

4. The multi-dimensional integrated power system fault detection algorithm according to claim 1 is characterized in that: The step S4, application of the extended Kalman filter, specifically comprises the following steps: Step S41: defining a power system state space model; Step S42: prediction step; Step S43: updating step; Step S44: Perform real-time state estimation on the voltage, current and total output power data.

5. The multi-dimensional integrated power system fault detection algorithm according to claim 4 is characterized in that: The formula of the power system state space model in step S41 is: State transition equation: x k =f(x k-1 )+w k-1 ; In the formula, x k is the state vector of the system at time step k, f(·) is the state transfer function, w k-1 is the process noise; Observation equation: z k =h(x k )+v k ; In the formula, z k is the observation vector, h(·) is the observation function, v k is the observation noise.

6. The multi-dimensional integrated power system fault detection algorithm according to claim 5, characterized in that: The specific steps of step S42 are: Step S421: Using the state estimation of the previous time step and the state transition equation predicts the state at the current time step Step S422: Calculate the predicted covariance matrix: In the formula, F k-1 is the Jacobian matrix of the state transfer matrix, P k-1 is the covariance matrix of the previous time step, is the matrix F k-1 The transpose of Q k-1 is the process noise covariance matrix.

7. The multi-dimensional integrated power system fault detection algorithm according to claim 6 is characterized in that: The specific steps of step S43 are: Step S431: Using the observed value z of the current time step k Update the state estimate: In the formula, is the state estimation vector predicted at time step k, K k is the Kalman gain; Step S432: Calculate Kalman gain In the formula, is the covariance matrix predicted at time step k, is the matrix H k The transpose of H k is the Jacobian matrix of the observation matrix, R k is the observation noise covariance matrix; Step S433: Update the covariance matrix Where I is the identity matrix.

8. The multi-dimensional integrated power system fault detection algorithm according to claim 1 is characterized in that: The step S5, principal component analysis dimensionality reduction, specifically comprises the following steps: Step S51: Obtain voltage, current and power data and perform standardization. The formula for standardization is: In the formula, x′ is the data after standardization, x is the original data, μ is the data mean, and σ is the data standard deviation; Step S52: Calculate the covariance matrix of the standardized data: In the formula, C is the covariance matrix, u is the number of samples, and x is a is the ath sample, and x is the sample mean; Step S53: Perform eigenvalue decomposition C on the covariance matrix v =λv; In the formula, λ is the eigenvalue and v is the eigenvector; According to the size of the eigenvalue, the eigenvectors corresponding to the first L largest eigenvalues ​​are selected as the principal components. The number of principal components L is selected depending on the proportion of data variance that needs to be retained; Step S54: Projecting the original data onto the selected principal component to obtain the principal component score y=XW; Where y is the principal component score matrix, X is the standardized data matrix, and W is the selected eigenvector matrix.

Images