Substation equipment state identification method and system based on GKA-RF technology

Abstract

The invention discloses a GKA-RF technology-based electrical equipment state identification and classification technical method for a transformer substation, so as to assist the fault diagnosis of the transformer substation, and the method comprises the steps: S1, obtaining the equipment state data of the transformer substation, and carrying out the preprocessing of the equipment state data through the GMM-based K-means clustering; s2, dividing the equipment state data into a training set and a test set based on the preprocessed equipment state data; s3, a priori algorithm (Apriori) is applied to the preprocessed training set, and mining of association rules is carried out; and S4, carrying out classification training by using a Random Forest algorithm, and applying the trained model to a test set and a newly added sample to carry out state identification and classification, thereby improving the accuracy and discrimination capability of the model on equipment state identification and classification.

Description

Technical Field

[0001] The present application relates to the technical field of power system protection, and in particular to a method and system for identifying equipment status of a substation. Background Art

[0002] In recent years, the level of intelligence in high-voltage substations has been significantly improved. Compared with the previous method of relying on periodic inspections by operation and maintenance personnel or simulating the operating conditions and fault conditions of some equipment in the laboratory, the current substations have gradually installed some sensors on the equipment, which can collect the operating status and various parameters of the equipment in real time, and can provide a large amount of data with time continuity and breadth to the operation and maintenance personnel. Therefore, compared with the time-consuming and error-prone traditional manual analysis methods, today's methods combining big data and artificial intelligence can effectively process the collected data through data mining, machine learning, deep learning, etc., and convert the raw data into more valuable information.

[0003] Many existing state identification and evaluation methods for substation electrical equipment use relatively simple algorithms, such as simple K-means or simple SVM classification. Although these methods perform well in some scenarios, they may not be able to provide accurate classification and evaluation for complex distribution or high-dimensional, nonlinear data. The few combined algorithms used are usually trained and evaluated based on a set of fixed parameters and rules, which makes it difficult to quickly adapt to the dynamic changes in the state of electrical equipment, thus greatly reducing flexibility and scalability. Summary of the invention

[0004] In view of the defects existing in the prior art, the present application provides a method, system and storage medium for identifying the equipment status of a substation to solve the problems of low accuracy and effectiveness of the existing electrical equipment status identification and evaluation methods, as well as insufficient flexibility in the distribution of different data in the substation.

[0005] To achieve the above objectives, the present application provides a substation equipment status identification and classification method based on GKA-RF technology, comprising the following steps:

[0006] S1. Obtain the equipment status data of the substation and pre-process the equipment status data using K-means clustering based on GMM;

[0007] S2, dividing the preprocessed device status data into a training set and a test set;

[0008] S3, applying the prior algorithm to the preprocessed training set to mine association rules;

[0009] S4. Use the Random Forest algorithm for classification training, and apply the trained model to the test set and newly added samples for state recognition and classification.

[0010] Further, S1, the step of obtaining the equipment status data of the substation and preprocessing the equipment status data using K-means clustering based on GMM specifically includes:

[0011] Gaussian mixture model (GMM) is used to divide the original data into discrete values ​​and continuous values ​​at multiple scales;

[0012] Specifically, the parameters of GMM are initialized by k-means clustering:

[0013]

[0014] Furthermore, the expectation maximization (EM) algorithm is used to summarize the distribution characteristics of the data and filter outliers, including:

[0015] Calculate the probability density function of the sample:

[0016]

[0017] Furthermore, the posterior probability of each sample belonging to each Gaussian component is calculated:

[0018]

[0019] Furthermore, the parameters of the GMM are updated in the M-step.

[0020] Furthermore, the step of performing association rule mining on the pre-processed device status data specifically includes:

[0021] Use the a priori algorithm to iteratively generate candidate sets;

[0022] Furthermore, the support value of the generated candidate set is calculated:

[0023]

[0024] Then the support value of the generated candidate set is compared with the preset threshold of the requirement realization, and those that meet the threshold requirement are recorded as frequent itemsets;

[0025] The loop is iterated until the frequent itemsets are generated;

[0026] Furthermore, the following indicators are used to evaluate the discovered association rules:

[0027]

[0028] Furthermore, classification evaluation is performed after mining the equipment status data;

[0029] Specifically, based on the mined association rules, new features matching the rules are generated for each piece of data, and then these new features are used together with the original features for subsequent classification.

[0030] Finally, test it on the test set.

[0031] The present application also provides a substation equipment status identification system based on GKA-RF technology, the equipment status identification system comprising:

[0032] An acquisition module, used to acquire equipment status data of the substation and pre-process the equipment status data;

[0033] An association rule mining module is used to mine association rules for the preprocessed device status data, identify frequent item sets and derive corresponding association rules;

[0034] The equipment status classification and evaluation module uses random forest to evaluate the data of unknown status based on the results of mining association rules of the equipment status evaluation data.

[0035] The result analysis module analyzes the accuracy and effectiveness of the model by comparing the results of the trained model test with the results on the real data set.

[0036] The present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying the equipment status of a substation as described in any one of the above items is implemented.

[0037] Compared with the prior art, the advantages of this application are:

[0038] This application uses a combination of GMM-based k-means clustering, prior algorithm mining association rules, and random forest classification to solve the problems of insufficient accuracy and flexibility of existing technologies. Our technology can handle complex distributions more flexibly, because the Gaussian mixture model GMM allows each cluster to have different variances and covariances, which can more accurately identify the device status than a single Gaussian model. At the same time, on the basis of k-means clustering, through further refinement of GMM, it can better capture nonlinear relationships and ensure that the status classification of the equipment is more refined and accurate. The random forest classifier has a strong ability to process multidimensional features. The feature space of the electrical equipment in our substation is relatively complex. The random forest can accurately distinguish between normal and faulty and abnormal states in the relatively complex feature space of electrical equipment. Through the concentration of multiple decision trees, it has a strong ability to resist overfitting and can process high-dimensional data. Moreover, by combining the two, it can dynamically process different substation equipment data and different operating conditions, while maintaining high accuracy and stability, and also has a certain degree of flexibility and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 A schematic diagram of a flow chart of a method for identifying and evaluating equipment status of a substation according to an embodiment of the present application;

[0041] Figure 2 A GMM clustering diagram of a method for identifying and evaluating equipment status in a substation according to an embodiment of the present application;

[0042] Figure 3 A schematic diagram of association rule mining of a method for identifying and evaluating equipment status of a substation according to an embodiment of the present application;

[0043] Figure 4 A random forest flow chart of a method for identifying and evaluating equipment status of a substation according to an embodiment of the present application;

[0044] Figure 5 A system block diagram of a device status identification system for a substation according to an embodiment of the present application;

[0045] Figure 6 A graph showing the importance of all features after random forest training;

[0046] Figure 7 This is a graph showing the results of running a random forest on the test set.

[0047] Description of reference numerals:

[0048] 100, acquisition module; 200, association rule mining module; 300, equipment status classification and evaluation module; 400, result analysis module. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0050] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0051] In the description of the present application, the terms "first / second" involved are merely used to distinguish similar objects and do not represent a specific order for the objects. It can be understood that "first / second" can be interchanged with a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0052] It should be understood that the orientation or position relationship is based on the orientation or position relationship shown in the drawings. These orientation terms are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application.

[0053] Power system equipment status identification and classification refers to the detection, analysis and classification of the operating conditions of electrical equipment in order to determine the health status and fault risk of the equipment. This plays a key role in "early warning" and "precise maintenance" in fault warning, and can reduce the risk of unplanned downtime to a certain extent, and improve the reliability and economic benefits of equipment.

[0054] In order to help the operation and maintenance personnel in the substation to more accurately and effectively identify and evaluate the status of electrical equipment, we study the status recognition and classification technology of electrical equipment based on GKA-RF. The key to equipment status recognition and classification is to accurately detect the health status of the equipment and to extract potential abnormalities and fault signals through data analysis technology, so as to ensure the efficient and safe operation of the equipment.

[0055] According to the difference in driving methods, the existing electrical equipment status recognition and classification technologies can be divided into the following two categories: data-driven methods and model-driven methods; and according to the difference in data processing methods, the existing electrical equipment status recognition and classification technologies can be divided into the following two categories: threshold-based and statistical analysis-based and feature extraction and dimensionality reduction-based. The above two classification methods are currently the more mainstream classification methods.

[0056] Data-driven methods are a type of method that has developed rapidly in recent years. They mainly rely on large-scale data during the operation of equipment. Through various machine learning and statistical models, we can extract certain features from them for classification and identification. The key is to use advanced algorithms to automatically discover patterns in data without requiring us to have sufficient experience or rely on traditional physical models or experts. Some of the technologies in this type of method include machine learning algorithms such as SVM, random forest, neural network, clustering and unsupervised learning methods such as k-means, GMM, and rule mining algorithms such as Apriori algorithm.

[0057] The model-driven electrical equipment state identification method relies on a deep understanding of the physical characteristics, operating mechanisms and failure modes of the equipment. It predicts the behavior of the equipment and detects faults by establishing a mathematical model or physical model of the equipment. The difference between this type of method and the data-driven method is that it does not rely entirely on historical data, but needs to be modeled and analyzed based on the actual working principle of the equipment, so it requires our model builders to have certain professional knowledge in the field. Some of the technologies in this type of method include physical model-driven methods such as state space models, equivalent circuit models, thermodynamic models, vibration and structural mechanics models, etc.; there are also models based on failure modes such as fault tree analysis and failure mode and effect analysis; there are also signal processing-based models such as Fourier transform and wavelet transform, as well as signal processing models and other methods.

[0058] In the identification of electrical equipment status, methods based on thresholds and statistical analysis are usually used to monitor the operating parameters of the equipment and determine whether the equipment is in normal operation by setting thresholds. This method is simple and easy to implement, so we often use this type of method in other methods. This type of method includes static threshold and dynamic threshold detection methods; there are also statistical analysis methods such as basic statistics, control chart method, and statistical process control.

[0059] The method based on feature extraction and dimensionality reduction is a common method in electrical equipment state identification and plays a key role. It is mainly used to extract meaningful features from high-dimensional data to reduce the data dimension and improve the efficiency and accuracy of subsequent analysis. These methods can effectively reduce computational complexity, remove redundant information, and enhance the generalization ability of the model. Therefore, researchers often combine such methods with data-driven methods and apply them in the data preprocessing stage. Common feature extraction methods include time domain feature extraction, frequency domain feature extraction, wavelet feature extraction, etc., while dimensionality reduction methods include principal component analysis, linear discriminant analysis, t-SNE, autoencoders, etc. There are also some methods that combine feature extraction with dimensionality reduction, such as filtering-based methods, wrapping-based methods, and embedding-based methods.

[0060] Figure 1 A schematic diagram of a flow chart of a method for identifying and evaluating equipment status of a substation according to an embodiment of the present application;

[0061] Figure 2 A GMM clustering diagram of a method for identifying and evaluating equipment status in a substation according to an embodiment of the present application; Figure 3 A schematic diagram of association rule mining of a method for identifying and evaluating equipment status of a substation according to an embodiment of the present application; Figure 4 A random forest flow chart of a method for identifying and evaluating the equipment status of a substation according to an embodiment of the present application; Figure 1 - Figure 4 As shown, the embodiment of the present application provides a method for identifying the equipment status of a substation based on the GKA-RF technology, comprising the following steps:

[0062] S1. Obtain the equipment status data of the substation and pre-process the equipment status data using K-means clustering based on GMM;

[0063] S2, dividing the preprocessed device status data into a training set and a test set;

[0064] S3, applying the prior algorithm to the preprocessed training set to mine association rules;

[0065] S4. Use the Random Forest algorithm for classification training, and apply the trained model to the test set and the newly added samples for state recognition and classification. Specifically, first obtain the state detection data of the electrical equipment of the substation. There are two sources of the equipment state data. One is the data collected on-site by the operation and maintenance personnel after the periodic power outage, and the other is the real-time data collected by the online monitoring equipment using various sensors during the operation of the power grid. The data collected by the operation and maintenance personnel during the power outage mainly reflects some inherent characteristics of the equipment. This type of state information includes the measurement date, the name of the measurement equipment, the measurement value, the environmental conditions (temperature and humidity, etc.) and the potential analysis results of the measurement personnel. The data obtained by online monitoring mainly represents the operation information of the electrical equipment. This type of data information includes some basic parameters (current, frequency, voltage, etc.) and advanced indicators (partial discharge conditions, dissolved gas, contact resistance, oil temperature, operation time, circuit breaker operation, etc.). Compared with the previous type of data information, this type of data information is more continuous in time, and the data set generated by real-time monitoring is more extensive and the data volume is larger. The acquired original data is then divided into a training set and a data set in a ratio of 7:3, and then the state data in the training set is preprocessed. We discretize the values ​​in the original data set through the Gaussian mixture algorithm GMM, so as to achieve a certain similarity in quantitative terms.

[0066] The data set we collected is represented by X = {x1,…,x n} means that each x i represents a sample. Each sample usually consists of multiple features. In order to better represent and process the multiple features of each sample, we further transform the dataset X into the following matrix form:

[0067]

[0068] Where n is the number of samples, each row represents a sample; m is the number of features of each sample, each column represents the value of the feature in different samples; x ij Represents the j-th eigenvalue of the ith sample.

[0069] We then use k-means for initial clustering, and then use the means and covariances of these clusters to initialize the parameters of the GMM. Our parameter initialization is as follows:

[0070]

[0071] Among them, μ krepresents the mean vector of the kth Gaussian component. Initially, we use the mean of each cluster as the initial mean; ∑k represents the covariance matrix of the kth Gaussian component. Initially, we calculate the variance of each sample and the mean μ based on the samples currently assigned to the component. k The difference between them, then find the outer product matrix, and finally take the average of these differences; π k Represents the weight of the kth Gaussian component in the entire mixture model. Initially, we assume that each component contributes equally to the overall model.

[0072] Then the parameters are iteratively updated through the expectation maximization (EM) algorithm until the model converges. The EM algorithm is mainly divided into the following two steps:

[0073] E-step: First calculate the probability density function of the sample:

[0074]

[0075] Furthermore, under the currently given parameters, the posterior probability of each data point belonging to each Gaussian component is calculated:

[0076]

[0077] It is calculated by calculating the proportion of the product of the weight of the current k-th Gaussian component and the probability density function in the current iteration.

[0078] M-step: This step is to update the parameters of GMM based on the results calculated in the previous E-step.

[0079] The update parameters are calculated as follows:

[0080]

[0081] Next, we use the prior algorithm to mine association rules for the preprocessed data. The core idea is to use the properties of frequent itemsets to reduce the search space by continuously recursively constructing itemsets.

[0082] Our specific approach is as follows:

[0083] First, we set two key parameters, the minimum support threshold minsup and the minimum confidence threshold minconf, according to the size of the actual data set. We set minsup to 5% and minconf to 90%.

[0084] Then we mine frequent itemsets. We give k candidate items in the entire item set and define them as I = {i1, i2, ..., i k}. The candidate item set for the first iteration is C1 = {{i1}, {i2}, ..., {i k}}.

[0085] After the first iteration of candidate item sets are created, the algorithm scans the transaction database to calculate the support value support of each candidate item set. The calculation formula is as follows:

[0086]

[0087] count({i k}) represents the number of item sets in the candidate set, and |D| represents the total number of elements in the candidate set.

[0088] Then, the items whose support values ​​are greater than the threshold minsup we set are recorded as frequent itemsets L1:

[0089] L1={{i k}∈C1|support({i k})≥minsup}

[0090] Furthermore, we need to generate a larger candidate set. The result of L1 is recorded as {{j1}, {j2}, ..., {j m}}, based on L1, select the candidate item set C2 for the second iteration = {{j1, j2}, {j1, j3}, ..., {j m-1 , j m}}.

[0091] Similarly, calculate the support value of each candidate item set:

[0092]

[0093] Then, the items whose support value in this iteration is greater than minsup are recorded as frequent itemsets L2:

[0094] L2={{j m-1 , j m}∈C2|support({j m-1 , j m})≥minsup}

[0095] Continue iterating pruning until or

[0096] We record the frequent itemsets obtained from the last iteration as L = L1UL2U...UL i .

[0097] After obtaining the final frequent item sets, the next step is to generate association rules and discover the conditional dependencies between item sets.

[0098] Then use the following indicators to evaluate the association rules found. For each rule A→B, calculate its confidence, such as:

[0099]

[0100] If the calculated confidence is higher than the minconf we set, we keep the rule, otherwise we discard it.

[0101] We then use lift to assess whether A and B are independent:

[0102]

[0103] If the calculated Lift value is greater than 1, it means that x2 and x1 are positively correlated.

[0104] Then use cosine similarity to define the strength of the rule, and the calculation formula is:

[0105]

[0106] If the calculated result is close to 1, it means that x2 and x1 appear at the same time most of the time, and x2→x1 is strong.

[0107] Then, based on the mined association rules, new features that match the rules are generated for each piece of data.

[0108] For a rule i , its condition can be converted into a binary feature:

[0109]

[0110] The feature set after being enhanced by association rules is denoted as X * ={x1 * ,x2 * ,…,x n *},in:

[0111] x i * =[x1,r1(x i ),r2(x i ),…,r k (x i )]

[0112] Furthermore, we use Synthetic Minority Oversampling (SMOTE) to balance our augmented data to minimize the class imbalance problem.

[0113] Then initialize a RandomForestClassifier instance and train an initial model to obtain some important features.

[0114] Furthermore, we visualize the feature importance after random forest training, so that we can understand which features contribute more to the model prediction.

[0115] Then use grid search hyperparameter optimization to automatically select the best hyperparameters such as n_estimators, max_depth, min_samples_split, min_samples_leaf and bootsrap through code.

[0116] Next, the random forest model is retrained based on the best parameters obtained from the previous grid search.

[0117] Finally, the model is tested on the test set to evaluate its performance.

[0118] To further improve the prediction performance, after evaluating on the test set, we used a voting classifier for model integration to see if we could get better results.

[0119] X mn In order to better understand the device status identification method of the substation in the embodiment of the present application, Figures 2 to 4 Provide a detailed description.

[0120] The equipment status recognition technology adopted in the embodiment of the present application is designed in combination with specific application scenarios, and is aimed at the recognition, classification and evaluation of the status data of the power equipment in the substation. The main technical points used are the Gaussian mixture model GMM to discretize the original data, and then use the prior algorithm to mine association rules. Finally, the mined association rules are classified and identified through random forests, which provides a certain degree of support for the subsequent fault discrimination and analysis technology using deep learning.

[0121] The following case analysis is based on sample data of a power station equipment to reflect the characteristics of this application.

[0122] For equipment status recognition and evaluation, we used the data of a 110kv substation of a power company to analyze the effectiveness of the electrical equipment recognition and status evaluation method based on GKA-RF technology. Here, we selected 1,000 sample data in the experiment, each sample data has 15 attribute features, a small part of which is shown in Table 1 below:

[0123] Table 1 Part of equipment status data

[0124] Temperature Current Voltage Frequency … Oil_Level 74.9671 177.9871 106.6241 49.0461 … 42.5883 68.6173 168.4926 109.2774 49.5698 … 58.0774 76.4768 151.1926 106.0379 49.7931 … 65.8034 65.3658 176.3478 101.9622 49.9531 … 50.5469 72.4196 191.5052 106.1542 49.3564 … 58.5301 50.8671 136.2162 105.3004 49.3014 … 47.2188 52.7508 184.7192 114.1473 49.7082 … 30.9624 64.3771 153.9582 109.0308 50.5191 … 45.1712 84.6564 160.4567 110.2996 50.6001 … 52.9008 … … … … … … 67.7422 138.5259 111.3868 49.7684 … 50.4390

[0125] After GMM discretization and clustering of the above original data, we obtained the following data set as shown in Table 2:

[0126] Table 2 GMM clustering result data

[0127]

[0128]

[0129] Further, we plot the importance of all features as follows Figure 6 shown.

[0130] Finally, we apply the trained model to the test set and get the following test results: Figure 7 shown.

[0131] Through the above analysis combined with the example data, the results are compared with the state characteristics of the sample data, and it is found that the electrical equipment state recognition and classification technology based on the GKA-RF technology in the embodiment of the present application has relatively good state recognition accuracy and effectiveness.

[0132] Figure 5 A system block diagram of a device status identification system for a substation according to an embodiment of the present application is shown in FIG. Figure 5 As shown, the embodiment of the present application also provides an equipment state identification system for a substation based on the GKA-RF technology, and the equipment state identification system includes an acquisition module 100, an association rule mining module 200, an equipment state classification evaluation module 300, and a result analysis module 400. Among them, the acquisition module 100 is used to acquire the equipment state data of the substation and preprocess the equipment state data, the association rule mining module 200 is used to mine association rules for the preprocessed equipment state data, identify frequent item sets and derive corresponding association rules, the equipment state classification evaluation module 300 uses random forests to evaluate and classify data of unknown states based on the results of the previous association rule mining of the equipment state evaluation data, and the result analysis module 400 compares the results of the trained model test with the results on the real data set to obtain the accuracy and effectiveness of the model.

[0133] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned substation fault diagnosis method is implemented.

[0134] It should be noted that the storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk or an optical disk.

[0135] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable storage medium, which may include a computer-readable storage medium (or a non-transitory medium) and a communication medium (or a temporary medium).

[0136] As is known to those of ordinary skill in the art, the term computer-readable storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer-readable storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is known to those of ordinary skill in the art that communication media typically contain computer-readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and may include any information delivery medium.

[0137] Exemplarily, the computer-readable storage medium may be an internal storage unit of the electronic device of the aforementioned embodiment, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device.

[0138] The above is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A method for identifying and classifying substation equipment status based on GKA-RF technology, characterized in that: The steps include: S1. Obtain the equipment status data of the substation and pre-process the equipment status data using K-means clustering based on GMM; S2, dividing the preprocessed device status data into a training set and a test set; S3, applying the prior algorithm to the preprocessed training set to mine association rules; S4. Use the random forest algorithm for classification training, and apply the trained model to the test set and newly added samples for state recognition and classification.

2. The method for identifying and classifying equipment status of a substation according to claim 1, characterized in that: There are two sources of equipment status data: one is the data collected on-site by operation and maintenance personnel after a certain periodic power outage; the other is real-time data collected by online monitoring equipment using various sensors during the operation of the power grid.

3. The device status identification and classification method of a substation according to claim 1, characterized in that: S1. Obtaining the equipment status data of the substation and preprocessing the equipment status data using K-means clustering based on GMM, specifically including: Divide the original data into discrete and continuous values ​​on multiple scales; The expectation maximization algorithm is used to summarize the distribution characteristics of the data and filter outliers.

4. The method for identifying equipment status of a substation according to claim 3, characterized in that: The mathematical model used to discretize the data is a Gaussian mixture model.

5. The method for identifying equipment status of a substation according to claim 3, characterized in that: The k-means clustering of the Gaussian mixture model parameters is initialized as follows: Among them, x i represents a sample, μ k represents the mean, ∑k represents the covariance matrix, π k It represents the mixing coefficient.

6. The method for identifying equipment status of a substation according to claim 3, characterized in that: The steps of summarizing the data distribution characteristics for the expectation maximization algorithm include: Calculate the probability density function N(x i |μ k ,∑k); Calculate the posterior probability P(x) that each sample belongs to each Gaussian component i |μ k ,∑k); Use EM to solve and update the parameters of the Gaussian mixture model in the M-step.

7. The method for identifying equipment status of a substation according to claim 6, characterized in that: The probability density function of the sample is calculated using the following formula: Among them, |Σ| represents the determinant of the covariance matrix, Σ -1 represents the inverse of the covariance matrix.

8. The method for identifying equipment status of a substation according to claim 6, characterized in that: The step of performing association rule mining on the state data of the device after preprocessing specifically includes: Use the a priori algorithm to iteratively generate candidate item sets; Compare the support value of the generated candidate item set with the preset threshold of the required implementation, and record those that meet the threshold requirements as frequent item sets; The loop is iterated until the frequent itemsets are generated; Multiple metrics are used to evaluate the discovered association rules.

9. The method for identifying equipment status of a substation according to claim 8, characterized in that: The support value of the candidate item set is calculated as follows: Among them, count({i k }) represents the number of item sets in the candidate set, and |D| represents the total number of elements in the candidate set.

10. The method for identifying equipment status of a substation according to claim 8, characterized in that: The indicators for evaluating association rules are as follows: Among them, Confidence represents the confidence level. If the confidence level is higher than the minimum confidence threshold minconf, the association rule can be considered acceptable. Lift represents the lift. A Lift value greater than 1 indicates a positive correlation, indicating that the appearance of x2 increases the possibility of x1. Cosine represents the cosine similarity, and the value range is [0,1]. The larger the value, the stronger the correlation.

11. The method for identifying equipment status of a substation according to claim 8, characterized in that: The steps of classification and evaluation after mining the electrical equipment status data include: Combine the extracted association rules and then label each piece of data in the training set with a rule-matching label; Perform random forest classification on the data in the training set; Finally, test it on the test set.

12. A substation equipment status identification system based on GKA-RF technology, characterized in that: The substation equipment status identification system comprises: An acquisition module, used to acquire equipment status data of the substation and pre-process the equipment status data; An association rule mining module is used to mine association rules for the preprocessed device status data, identify frequent item sets and derive corresponding association rules; The equipment status classification and evaluation module uses random forest to evaluate the data of unknown status based on the results of the previous association rule mining of the equipment status evaluation data; The result analysis module analyzes the accuracy and effectiveness of the model by comparing the results of the trained model test with the results on the real data set.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for identifying the equipment status of a substation according to any one of claims 1 to 11 is implemented.

Images