Power grid voltage out-of-limit cause diagnosis method and system based on big data mining

Abstract

The invention relates to a power grid voltage out-of-limit cause diagnosis method and system based on big data mining, and the method comprises the steps: collecting voltage monitoring data; performing voltage out-of-limit extraction on the voltage monitoring data to obtain a plurality of voltage out-of-limit fragments; carrying out dimension reduction processing on the plurality of voltage out-of-limit fragments by adopting a symbolic representation method, and carrying out rapid classification on the plurality of voltage out-of-limit fragments after the dimension reduction processing through Hash mapping to obtain a plurality of sequence groups with different durations; clustering the plurality of sequence groups with different durations by adopting a spectral clustering algorithm based on dynamic time warping method similarity to obtain a plurality of typical abnormal voltage sequences in each sequence group; and performing fault diagnosis on the typical abnormal voltage sequence by adopting the case library to obtain a fault reason of the typical abnormal voltage sequence. According to the invention, the voltage abnormity fault reason can be accurately and efficiently positioned.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid voltage monitoring and big data analysis, and in particular to a method and system for diagnosing causes of power grid voltage over-limit based on big data mining. Background Art

[0002] With the expansion of the power system and the diversification of power users' needs, the structure and operating environment of the power grid have become increasingly complex, and voltage quality issues have gradually become one of the key factors for the stable operation of the power system. Voltage over-limit, that is, the voltage value exceeds its prescribed safety range, may have a serious impact on the safety of power equipment and users, and even cause economic losses. Therefore, it is crucial to quickly and accurately diagnose the cause of voltage over-limit for the operation and maintenance of the power system.

[0003] Traditional voltage quality management and fault diagnosis mainly rely on manual experience and simple statistical analysis. This method requires high professional knowledge of power grid operation and maintenance personnel, and with the rapid growth of power grid data, manual analysis and processing are no longer efficient. In addition, due to the wide distribution of users and the complex and diverse causes of voltage anomalies, traditional methods cannot comprehensively and quickly locate the root cause of voltage exceeding the limit. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for diagnosing the cause of power grid voltage over-limit based on big data mining, which can accurately and efficiently locate the cause of abnormal voltage faults.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a method for diagnosing the cause of over-limit voltage of a power grid based on big data mining, comprising the following steps:

[0006] Collect voltage monitoring data;

[0007] Performing voltage over-limit extraction on the voltage monitoring data to obtain a plurality of voltage over-limit segments;

[0008] A symbolic representation method is used to perform dimensionality reduction processing on the multiple voltage-exceeding segments, and a hash map is used to quickly classify the multiple voltage-exceeding segments after the dimensionality reduction processing to obtain a plurality of sequence groups with different durations;

[0009] A spectral clustering algorithm based on the similarity of dynamic time warping method is used to cluster several sequence groups with different durations, and several typical abnormal voltage sequences in each sequence group are obtained.

[0010] The case library is used to perform fault diagnosis on typical abnormal voltage sequences and obtain the fault causes of typical abnormal voltage sequences.

[0011] Before the voltage monitoring data is subjected to voltage over-limit extraction, the voltage monitoring data is also preprocessed. The preprocessing includes: integrating the voltage monitoring data, removing abnormal values ​​in the voltage monitoring data, and filling missing values ​​in the voltage monitoring data.

[0012] The voltage monitoring data is subjected to voltage over-limit extraction to obtain a plurality of voltage over-limit segments, specifically:

[0013] Extracting voltage monitoring data in which the voltage exceeds the limit for a period exceeding a preset time, and obtaining multiple voltage exceeding limit segments;

[0014] Determine whether the time step between two adjacent voltage-over-limit segments is less than a preset step;

[0015] If the time step between two adjacent voltage-exceeding segments is smaller than a preset step, the two adjacent voltage-exceeding segments are merged into one voltage-exceeding segment.

[0016] The multiple voltage over-limit segments are subjected to dimensionality reduction processing by using a symbolic representation method, and the multiple voltage over-limit segments after dimensionality reduction processing are quickly classified by using a hash map to obtain a plurality of sequence groups with different durations, specifically including:

[0017] Calculate the voltage exceeding limit duration of each voltage exceeding limit segment;

[0018] Set N-1 cut-off points on the standard normal distribution curve to divide N equal probability intervals, and use the horizontal coordinates of the cut-off points as the mapping threshold;

[0019] According to the voltage exceeding time length and mapping threshold of each voltage exceeding time segment, the voltage exceeding time segment is mapped to a corresponding interval to obtain a plurality of sequence groups with different time durations.

[0020] The spectral clustering algorithm based on the similarity of the dynamic time warping method is used to cluster several sequence groups with different durations to obtain several typical abnormal voltage sequences in each sequence group, specifically including:

[0021] For each sequence group with different durations, the dynamic time warping method is used to calculate the similarity of the sequences within the group, and several similarity matrices are obtained;

[0022] Spectral clustering is used to cluster the voltage over-limit segments in each sequence group with different durations, a similarity graph is constructed according to the similarity matrix, and then dimension reduction is performed through the eigenvector of the Laplace matrix of the similarity graph. Clustering is performed using a clustering algorithm in a low-dimensional space to obtain several typical abnormal voltage sequences in each sequence group.

[0023] When clustering is performed using a clustering algorithm in a low-dimensional space, the number of clusters when the silhouette coefficient and the Calinski-Harabasz index reach a relatively optimal value is used as the final number of clusters of the clustering algorithm.

[0024] The method of using the case library to perform fault diagnosis on the typical abnormal voltage sequence to obtain the fault cause of the typical abnormal voltage sequence specifically includes:

[0025] Calculate the similarity between the typical abnormal voltage sequence and the cases in the case library to obtain a first similarity result;

[0026] Calculate the similarity between the index characteristics of the typical abnormal voltage sequence and the index characteristics of the cases in the case library to obtain a second similarity result;

[0027] The first similarity result and the second similarity result are weightedly summed, and the case in the case library corresponding to the highest score is used as the fault cause of the typical abnormal voltage sequence.

[0028] When the fault cause of the typical abnormal voltage sequence does not match the actual situation, the W′ kj =W kj (1±Δ×R kj ) to modify the weight, where W kj represents the weight of the jth indicator feature in the kth case in the case library before correction, W′ kj represents the weight of the jth indicator feature in the kth case in the case library after correction, Δ represents the amplitude coefficient of each weight adjustment, R kj Indicates the importance of the j-th indicator feature on the k-th case.

[0029] The technical solution adopted by the present invention to solve the technical problem is: to provide a power grid voltage over-limit cause diagnosis system based on big data mining, comprising:

[0030] A collection module, used for collecting voltage monitoring data;

[0031] An extraction module, used for performing voltage over-limit extraction on the voltage monitoring data to obtain a plurality of voltage over-limit segments;

[0032] A dimensionality reduction classification module is used to perform dimensionality reduction processing on the multiple voltage over-limit segments by using a symbolic representation method, and quickly classify the multiple voltage over-limit segments after the dimensionality reduction processing by using a hash map to obtain a plurality of sequence groups with different durations;

[0033] A clustering module is used to cluster a number of sequence groups with different durations using a spectral clustering algorithm based on the similarity of a dynamic time warping method, and obtain a number of typical abnormal voltage sequences in each sequence group;

[0034] The diagnosis module is used to perform fault diagnosis on typical abnormal voltage sequences by using a case library to obtain the fault causes of the typical abnormal voltage sequences.

[0035] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the above-mentioned method for diagnosing the cause of grid voltage over-limit based on big data mining are implemented.

[0036] The technical solution adopted by the present invention to solve its technical problem is: providing a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing the cause of grid voltage over-limit based on big data mining are implemented.

[0037] Beneficial Effects

[0038] Due to the adoption of the above technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention introduces big data mining technology and utilizes an automated voltage data analysis method to significantly improve the speed and accuracy of fault diagnosis and reduce dependence on manual experience; the present invention adopts symbolic representation and hash mapping technology to first symbolize a complex voltage sequence and aggregate similar data through hash mapping, which greatly reduces the amount of data processing calculations and improves the analysis efficiency of voltage data; the present invention introduces a spectral clustering algorithm, combines the silhouette coefficient and the Calinski-Harabasz index, automatically determines the number of cluster centers, realizes accurate clustering of voltage out-of-limit segments, and significantly improves the accuracy of anomaly identification; the present invention combines a case-based reasoning system and utilizes similarity analysis of historical cases and current voltage data to quickly determine the cause of voltage out-of-limit, thereby realizing intelligent and automated fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of a method for diagnosing causes of over-limit voltage in a power grid based on big data mining according to a first embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of an IQR detection outlier in the first embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of completing missing voltage data in the first embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of a hash map based on symbolic representation in a first embodiment of the present invention;

[0043] Figure 5is a schematic diagram of a similarity matrix calculated by FastDTW in the first embodiment of the present invention;

[0044] Figure 6 is a schematic diagram of a typical abnormal voltage sequence obtained by clustering in the first embodiment of the present invention;

[0045] Figure 7 This is a diagram of a voltage big data waveform display interface in an embodiment of the present invention;

[0046] Figure 8 1 is a schematic diagram showing the proportion of abnormal voltage causes in each line according to an embodiment of the present invention;

[0047] Fig. 9 Schematic diagram of the proportion of abnormal voltage causes in each substation in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0049] The first embodiment of the present invention relates to a method for diagnosing the cause of voltage over-limit in a power grid based on big data mining. The method analyzes and processes massive voltage data based on big data mining technology, combines spectral clustering algorithm to implement cluster analysis of voltage anomalies, and performs intelligent diagnosis of the cause of voltage over-limit through case reasoning. Figure 1 As shown, the specific steps include:

[0050] Step 1: Collect voltage monitoring data.

[0051] Due to the complex operating environment of the power grid, there may be abnormal values ​​or missing values ​​in the voltage monitoring data, and the data is exported from different metering systems. Therefore, before analyzing the voltage monitoring data, it is necessary to preprocess the voltage monitoring data as follows:

[0052] First, match and integrate the data in multiple data tables according to the station area name:

[0053] Due to the different naming conventions of the substations, multi-character matching is performed to associate them and the data is stored in the database. These data include information such as lines, substations, users, voltage amplitude, collection time, distribution transformer capacity, access capacity, etc. According to the data model and query requirements, a suitable database table structure is designed. According to the line name, substation number, and user number, the data is multi-classified to structure the data, and the data collection time is standardized into a unified time format, and the data of each user is sorted in chronological order.

[0054] Then, outliers are removed:

[0055] Since the data collection process may be affected by meter failure or signal transmission, some users’ data may contain abnormal values, which will affect the stability of the subsequent clustering algorithm. Therefore, these user data need to be deleted. In this implementation, the abnormal user data is excluded by the threshold method. or greater than When the voltage value reaches the threshold value, all voltage data of the user will be deleted. and The interquartile range (IQR) method was used to determine the

[0056] The IQR method is a statistical method for detecting outliers based on the quartiles of data, which mainly includes the following steps:

[0057] a. Calculate the first quartile (Q1) and the third quartile (Q3) of the data;

[0058] b. Calculate the interquartile range IQR = Q3-Q1;

[0059] c. Calculate the outlier boundary, i.e. the threshold

[0060]

[0061] Among them, λ is a custom constant;

[0062] d. The data that is less than the lower limit Or greater than the upper limit The value of is marked as an outlier, and the detected outlier is considered to be a voltage anomaly (see Figure 2 ).

[0063] Finally, fill in the missing values:

[0064] Missing voltage monitoring data are filled by linear interpolation. If the missing values ​​appear at the beginning or end of the sequence, they are filled with the average nighttime voltage value to ensure the integrity and consistency of the data.

[0065] When filling, first traverse all user voltage monitoring data with missing values. The number of consecutive missing data between two valid data is defined as the missing value segment length N. If N≤3, it is marked as a missing segment that can be filled. Take the valid data at both ends of the missing segment that can be filled for linear interpolation, denoted as (x j ,y j ), (x k ,y k), where j and k satisfy 2≤kj≤4.

[0066] Assume x i (j<i<k) is the horizontal coordinate of a missing point, and its vertical coordinate y i The calculation is as follows:

[0067]

[0068] The above method cannot effectively fill in the small missing values ​​at the beginning or end of the voltage sequence. Considering that the voltage data at night is generally stable, the method of calculating the average value for forward or backward filling is adopted to complete the sequence. The operation window range is between the 1st to 4th sequence or the 93rd to 96th sequence, that is, 0:00-1:00 and 23:00-24:00 at night.

[0069] Assume that a voltage sequence has n missing data at the beginning and but

[0070]

[0071] The same is true for the case where there are missing values ​​at the end of the sequence, which will not be discussed here. Figure 3 A schematic diagram of completing missing voltage monitoring data is given.

[0072] Step 2: extract voltage exceeding limit from the voltage monitoring data to obtain multiple voltage exceeding limit segments.

[0073] If the voltage exceeds the limit for more than one hour, it is considered that the user may have a problem. Therefore, this step extracts the voltage monitoring data of the voltage exceeding the limit for more than the preset time to obtain multiple voltage exceeding limit segments. The reasons for exceeding the limit between two adjacent segments may be related. Therefore, if the time step interval between two adjacent voltage exceeding limit segments is less than 3, the two voltage exceeding limit segments are merged into one voltage exceeding limit segment, so as to ensure the integrity of the segment data as much as possible.

[0074] Step 3: perform dimensionality reduction processing on the multiple voltage-exceeding segments by using a symbolic representation method, and quickly classify the multiple voltage-exceeding segments after the dimensionality reduction processing by using a hash map to obtain a plurality of sequence groups with different durations.

[0075] like Figure 4As shown in the figure, hash mapping based on symbolic representation means: symbolizing the voltage sequence, and converting it into a series of symbols (such as a, b, c, d, etc.) according to key features such as voltage mean, voltage fluctuation coefficient, voltage fluctuation trend or voltage over-limit duration, where each symbol represents a different voltage characteristic. Using the local sensitive hashing (LSH) technology, the symbol sequence is mapped to the corresponding hash bucket, and the data in the same hash bucket has similar characteristics. This step greatly reduces the amount of calculation for subsequent similarity calculations and avoids global similarity calculations for all data.

[0076] The key features such as voltage mean, voltage fluctuation coefficient, voltage fluctuation trend and voltage limit exceeding duration are extracted as follows:

[0077] The average voltage value reflects the severity of the over-limit behavior to a certain extent and is an important statistical indicator. i = {u1,u 2,…, u n} represents a certain voltage sequence fragment that exceeds the limit, and the voltage mean The calculation method is as follows:

[0078]

[0079] The duration of voltage over-limit is of great guiding significance for fault diagnosis. Usually, a short-term over-limit may be caused by user load. If the duration is long, it is necessary to consider whether there is a problem with the line. Assume that the over-limit voltage sequence segment is U i = {u1,u 2,…, u n}, then the duration of the voltage exceeding the limit is expressed as L=n.

[0080] The Mann-Kendall (MK) method is a nonparametric statistical method used to analyze the trend of time series data. This method does not require the data to meet a specific distribution and is insensitive to outliers. The steps for using the MK method to analyze the trend of over-limit voltage segments are as follows:

[0081] a. For each data point u i (i=1,2,…,n-1), calculate it and each subsequent data point u j The sign of the difference of (j=i+1,i+2,…,n) is summed to obtain the Mann-Kendall statistic:

[0082]

[0083] Among them, sign(u j -u i ) is the sign function of the difference, if u j >ui , then +1, if u j <u i , then -1, if u j =u i , then it is 0.

[0084] b. Calculate the variance:

[0085]

[0086] c. Use the statistic S and its variance to calculate the standard normal deviation Z:

[0087]

[0088] d. Determine the significance of the trend change based on the statistic S and the standard normal distribution table. If the absolute value of the standard normal deviation Z is greater than the critical value, the trend change is considered significant. The direction of the trend is determined by the positive or negative value of the standard normal deviation Z. A positive value indicates an upward trend, and a negative value indicates a downward trend. The absolute size of the standard normal deviation Z indicates the significance of the trend change. If the absolute value of the standard normal deviation Z is less than the critical value, the trend change is considered insignificant.

[0089] The standard deviation is a statistic that measures the degree of data dispersion. It indicates the deviation between the data and its average value and is used to quantify the degree of dispersion of a set of values. Using it as a feature can reflect the voltage fluctuation to a certain extent. Therefore, this embodiment uses the standard deviation as the voltage fluctuation coefficient, and its calculation method is:

[0090]

[0091] Where n is the number of data in the acquisition segment, u i represents the i-th voltage sampling value in the segment, and U represents the voltage mean.

[0092] Large-scale sample data usually obeys normal distribution. Z-Score standardization is performed on it to make it satisfy the standard normal distribution, and the extracted sequence features are mapped into symbolic representation. N-1 cut-off points are set on the standard normal distribution curve to divide N equal probability intervals. The horizontal coordinates of the cut-off points are used as the mapping threshold {A1, A2, ..., A N-1 The mapping symbol is determined based on the comparison of the over-limit voltage mean, the voltage fluctuation coefficient and a plurality of mapping thresholds.

[0093] The mapping relationship between the voltage over-limit duration and the voltage fluctuation trend is shown in Table 1.

[0094] Table 1 Symbol mapping relationship table

[0095]

[0096] Finally, symbol sequences are constructed and each sequence is represented in the form of a string. Each sequence is represented by 4 letter symbols, and different symbol combinations are mapped to different hash buckets.

[0097] In this implementation, the voltage-exceeding duration of each voltage-exceeding segment is calculated; three dividing points are set on the standard normal distribution curve to divide four intervals of equal probability, and the horizontal coordinates of the dividing points are used as mapping thresholds, that is, 16, 32 and 48 are used as mapping thresholds, and according to the voltage-exceeding duration of each voltage-exceeding segment and the mapping threshold, the voltage-exceeding segment is mapped to the corresponding interval to obtain a number of sequence groups with different durations.

[0098] However, the above four key features alone cannot fully characterize the sequence characteristics. For example, the trend obtained by the MK method cannot effectively distinguish between sequences with both rising and falling trends. The voltage mean and fluctuation coefficient will be affected by outliers. Therefore, it is necessary to perform spectral clustering based on the coarse classification of symbol sequence hash mapping to obtain more refined classification results.

[0099] Step 4: Cluster several sequence groups with different durations using a spectral clustering algorithm based on the similarity of a dynamic time warping method to obtain several typical abnormal voltage sequences in each sequence group.

[0100] Spectral clustering requires calculating the similarity matrix between sequences, but since the duration of the voltage limit exceeded by the user varies, the length of the extracted segments is also uneven. The traditional method of calculating the segment similarity based on the Euclidean distance is not feasible. Therefore, this implementation uses the dynamic time warping (DTW) method to measure the similarity between sequences. DTW allows the time axis to be distorted to achieve local alignment, so it can capture the time offset and stretching in the sequence, so as to more flexibly adapt to changes in different speeds or time.

[0101] When the DTW algorithm is applied to the similarity calculation of user voltage anomaly segments, each voltage anomaly segment can be regarded as a time series. By calculating the DTW similarity between different segments, voltage segments with similar abnormal patterns can be identified. The core idea of ​​the DTW algorithm is to find the best matching path between two time series through dynamic programming technology, so that the sum of the distances between the point pairs on this path is minimized. This best matching path is called a "warping path", which defines a nonlinear mapping method that allows the two sequences to be aligned as much as possible under the mapping, so that the similarity between them can be flexibly measured. The algorithm steps are as follows:

[0102] a. Initialize the distance matrix: construct a matrix D of size m×n, where D(i,j) represents the distance between the i-th element of sequence A and the j-th element of sequence B, usually calculated using Euclidean distance.

[0103] b. Calculate the cumulative distance matrix: Use dynamic programming to fill the matrix D and calculate the minimum cumulative distance D' from D(1,1) to D(m,n). For each element D'(i,j) in the matrix, its value is updated to D(i,j) plus the minimum cumulative distance of its three adjacent elements to the left, above, and above the left. The specific formula is:

[0104] D'(i,j)=D(i,j)+min(D(i-1,j),D(i,j-1),D(i-1,j-1));

[0105] c. Find the best path: Start from D'(m,n) and trace back to find the path with the smallest cumulative distance. This path is the best matching path.

[0106] d. Calculate similarity: The sum or average of the distances between the point pairs on the best matching path is used as the similarity measure of the two sequences.

[0107] DTW uses dynamic programming to calculate the best matching path between two sequences, which results in a computational complexity of O(mn), where m and n are the sequence lengths. This quadratic complexity makes it very time-consuming to calculate long sequences. In order to speed up the DTW calculation, the improved algorithm FastDTW (Fast Dynamic TimeWarping) can be used. FastDTW reduces the computational complexity by limiting the search space and introducing approximate methods, thereby improving the computational speed to a certain extent, but this also introduces some loss of accuracy, reducing its complexity to O(m*log(n)). FastDTW uses a hierarchical strategy to decompose the original sequence into multiple levels of subsequences. Then, it calculates DTW at different levels to reduce the computational complexity. The algorithm adopts a "rough to fine" strategy, first calculating DTW at a coarse level, and then gradually improving the accuracy through refined calculations until the required similarity level is reached.

[0108] Based on the hash bucket classification, this implementation method calls the anomaly with a time limit of more than 1 hour and less than 4 hours as a short-term anomaly, the anomaly with a time limit of more than 4 hours and less than 8 hours as a long-term anomaly, the anomaly with a time limit of more than 8 hours and less than 12 hours as a long-term anomaly, and the anomaly with a time limit of more than 8 hours as an ultra-long-term anomaly, corresponding to the sequences represented by symbols a, b, c, and d. For each sequence group with different durations, FastDTW is used to calculate the similarity of the sequences in the group to obtain the similarity matrices Sa, Sb, Sc, and Sd. In order to integrate the sequence relationship obtained by the previous symbolic representation, weights are added to the sequence similarities represented by the same symbol between different hash buckets in the same group, and the importance of the selected feature representation in the similarity calculation is strengthened.

[0109] Then, spectral clustering is used to cluster the voltage-limited segments in each sequence group with different durations, and a similarity graph is constructed based on the similarity matrix obtained above (see Figure 5 ), then the eigenvectors of the Laplace matrix of the similar graph are used for dimensionality reduction, and finally the traditional clustering algorithm (such as K-means) is used for clustering in the low-dimensional space. In this embodiment, the number of cluster centers is selected according to the parameters when the two indicators of silhouette coefficient and Calinski-Harabasz index reach the relative optimality, so as to ensure the rationality and accuracy of the clustering results. Finally, several typical abnormal voltage sequences will be obtained for each time length segment (see Figure 6 ).

[0110] Step 5: Use the case library to perform fault diagnosis on the typical abnormal voltage sequence to obtain the fault cause of the typical abnormal voltage sequence.

[0111] This implementation method designs a voltage intelligent diagnosis method based on case reasoning, and uses a historical case library to perform fault diagnosis on the current voltage over-limit problem.

[0112] Case-Based Reasoning (CBR) is a strategy that uses historical cases to guide the solution of current cases. Its core idea is to find historical cases similar to the current problem and solve new problems based on these cases. The CBR process for fault diagnosis mainly includes two core steps: retrieval and revision.

[0113] After reusing a case solution, you may need to revise it to ensure that the case matching process is optimized. If new problems are encountered during the process, or if new solutions are available, you need to consider whether to add them to the case library.

[0114] When implementing this step, the case library is first constructed: the case library stores typical cases of different types of voltage over-limit faults, such as too long line power supply radius, concentrated load, insufficient reactive power compensation, photovoltaic power reverse transmission, etc. Each case corresponds to the characteristic sequence of the voltage over-limit fragment.

[0115] Then, case retrieval is performed: retrieval is to find the case that is most similar to the current problem in the case library, which requires a clear problem description and an effective similarity evaluation standard, so as to find similar cases in the case library accurately and quickly. When a case (i.e., user voltage sequence) is input, the over-limit fragment is first extracted, and then the similarity is calculated with the typical sequence obtained by spectral clustering, and it is classified into a certain typical abnormal voltage sequence. At this time, the similarity of the typical abnormal voltage sequence and the case in the case library is calculated to obtain the first similarity result, and then the index characteristics of the typical abnormal voltage sequence (such as the distance between the user and the power supply point, the diameter of the cable used in the line, and the transformer distribution capacity) are calculated with the index characteristics of the case in the case library to obtain the second similarity result; finally, the first similarity result and the second similarity result are weighted and summed, and the case in the case library corresponding to the highest score is used as the fault cause of the typical abnormal voltage sequence.

[0116] After obtaining the diagnosis result, if it does not match the actual situation, the search rules need to be revised to ensure that the case matching is more accurate. If new problems are encountered in the process, and new diagnosis results need to be considered whether to add to the case library, the search rules also need to be revised after adding to the case library. In this embodiment, the revision of the search rules is achieved by the system automatically adjusting the weights of each feature item, and the weight adjustment formula is as follows:

[0117] W k ' j =W kj (1±Δ×R kj );

[0118] Among them, W kj represents the weight of the jth indicator feature in the kth case in the case library before correction, W k ' j represents the weight of the jth indicator feature in the kth case in the case library after correction, Δ represents the amplitude coefficient of each weight adjustment, R kj Indicates the importance of the j-th indicator feature on the k-th case, expressed as:

[0119]

[0120] Where V k It represents the score of each feature when calculating the similarity of the kth case, and V0 represents the score of the current case.

[0121] In order to verify the effectiveness of this embodiment in actual power grid operation, a specific example of voltage over-limit diagnosis based on a city's low-voltage distribution network is provided below. This example shows how to automatically diagnose voltage over-limit problems occurring in a power grid using the method of this embodiment and provide corresponding solutions.

[0122] 1. Implementation environment:

[0123] The subjects of this experiment are low-voltage users in 10 substations in a city's distribution network, and voltage data of more than 5,000 users were collected. The system collects voltage data from smart meters and advanced measurement equipment every 15 minutes, and each user generates 96 voltage data per day. The experimental period is 1 month, and a total of about 150 million voltage data were collected. The key hardware equipment in the experiment includes:

[0124] Data collection terminals: smart meters and advanced measurement equipment.

[0125] Data processing server: used for data preprocessing, symbolic representation, hash mapping and spectral clustering.

[0126] Diagnostic server: runs the case-based reasoning system to provide diagnosis of voltage over-limit causes.

[0127] User terminal: A visual interface for operation and maintenance personnel to view diagnostic results.

[0128] 2. Implementation steps

[0129] 2.1 Data collection and preprocessing

[0130] The system collects voltage data for each user in the distribution network through smart meters and advanced measurement equipment. The system generates 96 voltage data for each user every day. During the collection process, the system removes abnormal data (such as voltage exceeding the specified threshold) and performs linear interpolation to complete a small amount of missing data to ensure data integrity.

[0131] 2.2 Symbolic Representation and Hash Mapping

[0132] The voltage data of each user is symbolized. According to the voltage fluctuation, mean, duration and other characteristics, the data is converted into a symbol sequence. Subsequently, these symbol sequences are classified into different hash buckets using hash mapping technology to reduce the computational complexity of the data. The data in each hash bucket represents users with similar voltage characteristics.

[0133] 2.3 Similarity calculation and spectral clustering

[0134] The dynamic time warping (DTW) algorithm is used to calculate the similarity of the symbolized voltage over-limit segments. The system automatically constructs a similarity matrix and performs cluster analysis on these voltage data through the spectral clustering algorithm to extract typical voltage anomaly patterns in different time periods. The system automatically determines the number of cluster centers through the silhouette coefficient and the Calinski-Harabasz index. Finally, the system identifies multiple typical segments of voltage over-limit, including the voltage over-limit problem caused by photovoltaic power generation reverse transmission and the voltage drop problem caused by load concentration.

[0135] 2.4 Fault Diagnosis of Case-Based Reasoning Systems

[0136] The system uses the built case library to match historical voltage limit-exceeding cases to intelligently diagnose the specific causes of voltage limit-exceeding. Figure 7 As shown, the system identifies the following two main types of faults:

[0137] Photovoltaic power backfeed: This problem caused some users in certain areas to experience overvoltage during the peak period of photovoltaic power generation. Diagnosis results showed that this problem was caused by the photovoltaic power generation equipment connected to the user feeding power back to the grid.

[0138] Load concentration during peak hours: Some users experienced voltage drops during peak hours (summer afternoons). Diagnosis results showed that this was due to load concentration and insufficient transformer capacity, which led to insufficient voltage.

[0139] 2.5 Diagnosis results and feedback

[0140] The diagnostic results are displayed through a visual interface (see Figure 8-Figure 9 ), operation and maintenance personnel can view the voltage over-limit information, causes and recommended solutions for each user in each substation. For the problem of photovoltaic power generation reverse transmission, the system recommends adjusting the reactive power compensation equipment in the substation and increasing the reactive power compensation capacity. For voltage drops caused by load concentration, the system recommends increasing transformer capacity or optimizing power dispatching.

[0141] 3. Implementation effect:

[0142] After one month of experimental verification, the system's diagnostic accuracy reached over 90%, and it was able to effectively identify the cause of voltage over-limit and provide corresponding solutions. By adjusting reactive power compensation equipment and optimizing power dispatch, the grid's operational stability was improved by 20%, and the voltage over-limit phenomenon was significantly reduced. In addition, the system's automatic diagnosis function reduced the workload of operation and maintenance personnel by about 35%, effectively reducing operation and maintenance costs.

[0143] It is not difficult to find that the present invention significantly improves the speed and accuracy of fault diagnosis and reduces the dependence on manual experience by introducing big data mining technology and utilizing an automated voltage data analysis method; the present invention adopts symbolic representation and hash mapping technology to first symbolize the complex voltage sequence and aggregate similar data through hash mapping, which greatly reduces the amount of data processing calculations and improves the analysis efficiency of voltage data; the present invention introduces a spectral clustering algorithm, combines the silhouette coefficient and the Calinski-Harabasz index, automatically determines the number of cluster centers, realizes accurate clustering of voltage out-of-limit segments, and significantly improves the accuracy of anomaly identification; the present invention combines a case-based reasoning system and utilizes similarity analysis of historical cases and current voltage data to quickly determine the cause of voltage out-of-limit, thereby realizing intelligent and automated fault diagnosis.

[0144] A second embodiment of the present invention relates to a power grid voltage over-limit cause diagnosis system based on big data mining, comprising:

[0145] A collection module, used for collecting voltage monitoring data;

[0146] An extraction module, used for performing voltage over-limit extraction on the voltage monitoring data to obtain a plurality of voltage over-limit segments;

[0147] A dimensionality reduction classification module is used to perform dimensionality reduction processing on the multiple voltage over-limit segments by using a symbolic representation method, and quickly classify the multiple voltage over-limit segments after the dimensionality reduction processing by using a hash map to obtain a plurality of sequence groups with different durations;

[0148] A clustering module is used to cluster a number of sequence groups with different durations using a spectral clustering algorithm based on the similarity of a dynamic time warping method, and obtain a number of typical abnormal voltage sequences in each sequence group;

[0149] The diagnosis module is used to perform fault diagnosis on typical abnormal voltage sequences by using a case library to obtain the fault causes of the typical abnormal voltage sequences.

[0150] The grid voltage over-limit cause diagnosis system based on big data mining also includes a preprocessing module for preprocessing the voltage monitoring data. The preprocessing includes: integrating the voltage monitoring data, eliminating abnormal values ​​in the voltage monitoring data, and filling missing values ​​in the voltage monitoring data.

[0151] The extraction module comprises:

[0152] An extraction unit is used to extract voltage monitoring data when the voltage exceeds the limit for a period of time exceeding a preset time, and obtain multiple voltage exceeding limit segments;

[0153] A judging unit, used to judge whether the time step between two adjacent voltage-over-limit segments is less than a preset step;

[0154] The merging unit is used to merge two adjacent voltage-exceeding segments into one voltage-exceeding segment when the time step between the two adjacent voltage-exceeding segments is less than a preset step.

[0155] The dimension reduction classification module includes:

[0156] A calculation unit, used for calculating the voltage over-limit duration of each voltage over-limit segment;

[0157] A segmentation unit is used to set N-1 segmentation points on the standard normal distribution curve to divide N equal probability intervals, and use the horizontal coordinates of the segmentation points as mapping thresholds;

[0158] The mapping unit is used to map the voltage-exceeding segment to a corresponding interval according to the voltage-exceeding duration of each voltage-exceeding segment and a mapping threshold, so as to obtain a plurality of sequence groups with different durations.

[0159] The clustering module comprises:

[0160] A similarity calculation unit is used to calculate the similarity of the sequences in each sequence group with different durations by using a dynamic time warping method to obtain a plurality of similarity matrices;

[0161] The clustering unit is used to cluster the voltage over-limit segments in each sequence group with different durations by using spectral clustering, construct a similarity graph according to the similarity matrix, and then reduce the dimension by the eigenvector of the Laplace matrix of the similarity graph, and cluster by using a clustering algorithm in the low-dimensional space to obtain several typical abnormal voltage sequences in each sequence group.

[0162] The clustering unit uses the number of clusters when the silhouette coefficient and the Calinski-Harabasz index reach a relatively optimal value as the final number of clusters of the clustering algorithm.

[0163] The diagnostic module comprises:

[0164] A first calculation unit, configured to calculate the similarity between the typical abnormal voltage sequence and the cases in the case library to obtain a first similarity result;

[0165] A second calculation unit is used to perform similarity calculation on the index characteristics of the typical abnormal voltage sequence and the index characteristics of the cases in the case library to obtain a second similarity result;

[0166] The summing unit is used to perform weighted summation on the first similarity result and the second similarity result, and take the case in the case library corresponding to the highest score as the fault cause of the typical abnormal voltage sequence.

[0167] When the fault cause of the typical abnormal voltage sequence does not match the actual situation, k ' j =W kj (1±Δ×R kj ) to modify the weight, where W kj represents the weight of the jth indicator feature in the kth case in the case library before correction, W k ' j represents the weight of the jth indicator feature in the kth case in the case library after correction, Δ represents the amplitude coefficient of each weight adjustment, R kj Indicates the importance of the j-th indicator feature on the k-th case.

[0168] A third embodiment of the present invention relates to an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for diagnosing the cause of grid voltage over-limit based on big data mining of the first embodiment when executing the computer program.

[0169] The fourth embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for diagnosing causes of grid voltage over-limit based on big data mining of the first embodiment.

[0170] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0171] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0172] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction method, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0173] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0174] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for diagnosing the cause of power grid voltage over-limit based on big data mining, characterized in that: The following steps are involved: Collect voltage monitoring data; Performing voltage over-limit extraction on the voltage monitoring data to obtain a plurality of voltage over-limit segments; A symbolic representation method is used to perform dimensionality reduction processing on the multiple voltage-exceeding segments, and a hash map is used to quickly classify the multiple voltage-exceeding segments after the dimensionality reduction processing to obtain a plurality of sequence groups with different durations; A spectral clustering algorithm based on the similarity of dynamic time warping method is used to cluster several sequence groups with different durations, and several typical abnormal voltage sequences in each sequence group are obtained. The case library is used to perform fault diagnosis on typical abnormal voltage sequences and obtain the fault causes of typical abnormal voltage sequences.

2. The method for diagnosing causes of power grid voltage over-limit based on big data mining according to claim 1 is characterized in that: Before the voltage monitoring data is subjected to voltage over-limit extraction, the voltage monitoring data is also preprocessed. The preprocessing includes: integrating the voltage monitoring data, removing abnormal values ​​in the voltage monitoring data, and filling missing values ​​in the voltage monitoring data.

3. The method for diagnosing causes of voltage over-limit in power grid based on big data mining according to claim 1 is characterized in that: The voltage monitoring data is subjected to voltage over-limit extraction to obtain a plurality of voltage over-limit segments, specifically: Extracting voltage monitoring data in which the voltage exceeds the limit for a period exceeding a preset time, and obtaining multiple voltage exceeding limit segments; Determine whether the time step between two adjacent voltage-over-limit segments is less than a preset step; If the time step between two adjacent voltage-exceeding segments is smaller than a preset step, the two adjacent voltage-exceeding segments are merged into one voltage-exceeding segment.

4. The method for diagnosing causes of voltage over-limit in power grid based on big data mining according to claim 1, characterized in that: The multiple voltage over-limit segments are subjected to dimensionality reduction processing by using a symbolic representation method, and the multiple voltage over-limit segments after dimensionality reduction processing are quickly classified by using a hash map to obtain a plurality of sequence groups with different durations, specifically including: Calculate the voltage exceeding limit duration of each voltage exceeding limit segment; Set N-1 cut-off points on the standard normal distribution curve to divide N equal probability intervals, and use the horizontal coordinates of the cut-off points as the mapping threshold; According to the voltage exceeding time length and mapping threshold of each voltage exceeding time segment, the voltage exceeding time segment is mapped to a corresponding interval to obtain a plurality of sequence groups with different time durations.

5. The method for diagnosing causes of voltage over-limit in power grid based on big data mining according to claim 1 is characterized in that: The spectral clustering algorithm based on the similarity of the dynamic time warping method is used to cluster several sequence groups with different durations to obtain several typical abnormal voltage sequences in each sequence group, specifically including: For each sequence group with different durations, the dynamic time warping method is used to calculate the similarity of the sequences within the group, and several similarity matrices are obtained; Spectral clustering is used to cluster the voltage over-limit segments in each sequence group with different durations, a similarity graph is constructed according to the similarity matrix, and then dimension reduction is performed through the eigenvector of the Laplace matrix of the similarity graph. Clustering is performed using a clustering algorithm in a low-dimensional space to obtain several typical abnormal voltage sequences in each sequence group.

6. The method for diagnosing causes of voltage over-limit in power grid based on big data mining according to claim 5 is characterized in that: When clustering is performed using a clustering algorithm in a low-dimensional space, the number of clusters when the silhouette coefficient and the Calinski-Harabasz index reach a relatively optimal value is used as the final number of clusters of the clustering algorithm.

7. The method for diagnosing causes of voltage over-limit in power grid based on big data mining according to claim 1, characterized in that: The method of using the case library to perform fault diagnosis on the typical abnormal voltage sequence to obtain the fault cause of the typical abnormal voltage sequence specifically includes: Calculate the similarity between the typical abnormal voltage sequence and the cases in the case library to obtain a first similarity result; Calculate the similarity between the index characteristics of the typical abnormal voltage sequence and the index characteristics of the cases in the case library to obtain a second similarity result; The first similarity result and the second similarity result are weightedly summed, and the case in the case library corresponding to the highest score is used as the fault cause of the typical abnormal voltage sequence.

8. The method for diagnosing causes of voltage over-limit in power grid based on big data mining according to claim 7 is characterized in that: When the fault cause of the typical abnormal voltage sequence does not match the actual situation, the W′ kj =W kj (1±Δ×R kj ) to modify the weight, where W kj represents the weight of the jth indicator feature in the kth case in the case library before correction, W′ kj represents the weight of the jth indicator feature in the kth case in the case library after correction, Δ represents the amplitude coefficient of each weight adjustment, R kj Indicates the importance of the j-th indicator feature on the k-th case.

9. A power grid voltage over-limit cause diagnosis system based on big data mining, characterized in that: include: A collection module, used for collecting voltage monitoring data; An extraction module, used for performing voltage over-limit extraction on the voltage monitoring data to obtain a plurality of voltage over-limit segments; A dimensionality reduction classification module is used to perform dimensionality reduction processing on the multiple voltage over-limit segments by using a symbolic representation method, and quickly classify the multiple voltage over-limit segments after the dimensionality reduction processing by using a hash map to obtain a plurality of sequence groups with different durations; A clustering module is used to cluster a number of sequence groups with different durations using a spectral clustering algorithm based on the similarity of a dynamic time warping method, and obtain a number of typical abnormal voltage sequences in each sequence group; The diagnosis module is used to perform fault diagnosis on typical abnormal voltage sequences by using a case library to obtain the fault causes of the typical abnormal voltage sequences.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for diagnosing the cause of grid voltage over-limit based on big data mining as described in any one of claims 1-8 are implemented.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for diagnosing the cause of over-limit voltage in a power grid based on big data mining as described in any one of claims 1 to 8 are implemented.

Images