Classification and state diagnosis method for direct current equipment
Through a machine learning method combining one-dimensional convolutional neural network and CIM model, the automated classification and state evaluation of DC devices are realized, solving the problems of low efficiency and insufficient accuracy in the existing technology, and improving the accuracy of equipment management and The level of intelligence of the system.
Patent Information
- Application Number
- CN202411045339.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-12-03
AI Technical Summary
The prior art has problems such as low efficiency, insufficient accuracy, lack of verification mechanism and insufficient technical depth in DC equipment classification and status diagnosis, and it is impossible to achieve automated, efficient and accurate equipment management and maintenance.
A one-dimensional convolutional neural network is used to extract features, build a power graph model based on CIM model, and combine machine learning to build a DC device classification recognition framework. Through data acquisition, preprocessing and pattern recognition model verification, it is integrated into the monitoring and monitoring of the power system and In the management platform, automated classification and status evaluation are realized.
It improves the accuracy and reliability of equipment classification, enhances the intelligence and safety of the system, and ensures the stable operation of the power system.
Smart Images

Figure CN119066547A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of power systems, and in particular to a classification and state diagnosis method for direct current equipment. Background Art
[0002] In modern power systems, there are many types of DC equipment, and different types of DC equipment have significant differences in function and structure. Traditional DC equipment classification methods usually rely on manual experience and manual marking, which are inefficient and inaccurate. This not only increases management costs, but may also lead to misoperation, affecting the safe and stable operation of the power system. Therefore, there is an urgent need for an automated, efficient and accurate DC equipment classification method to improve the operation and management level of the power system. Chinese patent CN202311119977.0 discloses a DC equipment operation status evaluation platform and detection method, which sets an application parameter extraction and analysis module, a connection cable basic parameter extraction and analysis module, a data comprehensive analysis module, an application terminal load status detection and analysis module, a protection mechanism trigger module, an application suitability abnormal feedback prompt module and an information base, and analyzes and calculates the application loss evaluation index of the connection cable to which the specified DC equipment belongs, thereby improving the level of detailed analysis of the basic parameters of the connection cable to which the DC equipment belongs. However, there are some shortcomings. The descriptions of data collection, preprocessing and analysis methods are relatively simple, lacking specific operational details and technical implementation processes. There is a lack of verification mechanism, and there is a lack of verification and optimization methods for the evaluation results, which cannot ensure the accuracy and reliability of the evaluation results. There is also insufficient technical depth, as advanced pattern recognition and machine learning technologies are not involved. The overall technical solution is relatively basic and lacks innovation and depth.
[0003] The detection methods of DC equipment are often unable to comprehensively and accurately evaluate the operating status of the equipment. They rely on the analysis of a single parameter and are difficult to reflect the actual condition of the equipment in actual applications. There is also a lag in the detection and feedback process, which makes it impossible to carry out effective equipment management and maintenance in a timely manner. Summary of the invention
[0004] Purpose of the invention: The purpose of the present invention is to provide a method for classifying and diagnosing the status of DC equipment.
[0005] Technical solution: The present invention comprises the following steps:
[0006] (1) Data collection and preprocessing: cleaning the collected raw data to remove noise and outliers;
[0007] (2) Feature extraction based on one-dimensional convolutional neural network;
[0008] (3) Construct a power diagram model based on the CIM model;
[0009] (4) Construct a DC device classification and recognition framework based on machine learning and input the extracted DC device feature data into the pattern recognition model;
[0010] (5) Verify and evaluate the classification results of the pattern recognition model;
[0011] (6) Integrate the trained pattern recognition model into the monitoring and management platform of the power system.
[0012] Furthermore, the data collection in step (1) includes:
[0013] Equipment characteristic data, including voltage, current, power, temperature and operating time, are collected from various DC devices in the power system. At the same time, the connection relationship and topological structure data of the DC devices are collected. Sensors and data acquisition devices are used to obtain this information in real time to provide basic data support for subsequent analysis.
[0014] Furthermore, the data preprocessing in step (1) includes:
[0015] The collected raw data is cleaned to remove noise and outliers to ensure the accuracy and reliability of the data; data normalization and denoising operations are performed. Among them, data normalization uses z-score standardization to standardize data of different modes to between 0 and 1, as shown in the following formula:
[0016]
[0017] Among them, x is the data of a certain mode; μ is the mean of the historical data of the mode; σ is the standard deviation of the historical data of the mode; the data of each mode is integrated into a multi-channel time series, as shown in the following formula:
[0018] X=[I t , V t , T t , V i b t ]
[0019] Where X is the data at a single time sampling point of the input data; t is the time step.
[0020] Furthermore, the feature extraction in step (2) includes:
[0021] The powerful feature extraction capability of convolutional neural networks is used to extract classification features of different DC devices from historical data. The convolution layer is the core component of the one-dimensional convolutional neural network, responsible for extracting local features from time series data. For each convolution layer, the convolution kernel is slid on the input data to calculate the dot product between the convolution kernel and the data. The convolution operation at time step t can be expressed as:
[0022]
[0023] Where D is the input data; K is the convolution kernel; k is the convolution kernel size;
[0024] The activation function is used to introduce nonlinear relationships. In the process of DC device classification and recognition, the nonlinear relationship between the input sensor data and the DC device classification category is enhanced within the 1D-CNN through the activation function;
[0025] Using the optimized ReLu function, it is expressed as:
[0026] f(x)=sigmoid(x)
[0027] f(x) = max(0, x)
[0028] Among them, x is the output of the current neuron, and the pooling layer is used to reduce the feature dimension.
[0029] Furthermore, the method for constructing the power diagram model in step (3) includes:
[0030] A DC equipment network topology structure based on the CIM basic model is established, and the structural characteristics and connection relationships of the equipment are converted into graph structure data. The topological characteristics of the equipment are analyzed through graph calculation methods. Then, the connection relationships between devices are mapped to edges. In addition, vertices with voltage level and reference voltage identification are established, and their association with load vertices is described by edges. In the topology structure, line information is defined on Line edges, and multi-time load data and equipment status information are defined in Time nodes and Time connect edges, showing the evolution of the system in space and time dimensions. Finally, a complete graph data model is formed through metadata mapping to construct a DC equipment network topology structure.
[0031] Furthermore, the CIM basic model is a specification for describing a power system model, and for related DC equipment, the DC equipment types in the CIM / E model are mapped to vertices in a graph database.
[0032] Furthermore, step (4) includes constructing a DC device classification and recognition framework based on machine learning, inputting the extracted DC device feature data into a pattern recognition model, training and optimizing model parameters, forming a classifier that can distinguish different types of DC devices, and realizing efficient and accurate device classification.
[0033] Furthermore, the pattern recognition in step (4) includes:
[0034] (4.1) The processed DC equipment operating status data is divided into four status sets: severe, abnormal, caution and normal, marked as s1, s2, s3, s4. The data in each set is:
[0035] s i ={s i1 ,s i2 ,…,s in}
[0036] In the formula, n is the number of data in the set;
[0037] (4.2) Construct the orthogonal correlation matrix of each state set as follows:
[0038]
[0039] And perform singular value decomposition on it to get the matrix Y i The eigenvalues and eigenvectors are arranged in descending order as follows:
[0040] λ i1 ≥λ i2 ≥…≥λ ik >0
[0041] α i1 , α i2 , …, α ik
[0042] Among them, λ is the eigenvalue; α is the eigenvector;
[0043] (4.3) The cumulative contribution rate of each eigenvalue above is determined by the following formula
[0044]
[0045] Select m principal component vectors of the operating status of the DC equipment;
[0046] (4.4) Calculate the projection of the orthogonal correlation matrix of each state set of the original power supply equipment in the m-dimensional feature space,
[0047] P=A i s i =[α i1 , α i2 , …, α ik ] T ·[s i1 ,s i2 ,…,s in ]
[0048] Among them, A i =[α i1 , α i2, …, α ik ] r is the characteristic factor matrix;
[0049] (4.5) Calculate the projection of the state information Z of the substation equipment to be evaluated in each state feature space,
[0050] w i =A i Z
[0051] Among them, w i To normalize the projection distance, take the state set with the minimum value as the operating state of the substation;
[0052] (4.6) According to the following formula, the operating status of the DC equipment is obtained, and the faults or defects of the equipment are clearly identified.
[0053]
[0054] Based on the graph computing model and combined with the DC equipment classification and identification algorithm, the computational efficiency and accuracy of the model are improved through parallel technology, and the position of the DC equipment in the topology structure is finally determined. In addition, the power flow calculation can be used to verify whether the DC equipment classification and identification algorithm meets the requirements of system operation, and to determine whether the line power or node voltage exceeds the limit. If an out-of-limit phenomenon occurs, timely alarm and early warning processing can be carried out.
[0055] Furthermore, the step (5) comprises:
[0056] (5.1) The data set is randomly divided into multiple subsets through cross-validation, and training and validation are performed alternately in a cycle to calculate the average classification accuracy and evaluate the generalization ability of the model;
[0057] (5.2) Construct a confusion matrix, compare the actual category with the predicted category, record the classification details, and calculate the accuracy, precision, recall and F1-score indicators to comprehensively evaluate the model performance.
[0058] (5.3) Based on the verification results, identify any faults or defects in the equipment, analyze the deficiencies in the model in terms of classification accuracy and stability, adjust model parameters, perform feature engineering, retrain and verify the model, and ensure that the optimized model has improved classification accuracy and stability.
[0059] Furthermore, step (5) includes integrating the trained pattern recognition model into the monitoring and management platform of the power system to realize the automatic classification of DC equipment, and provide a visual display and report generation function of the classification results. It also integrates alarm and early warning functions. When the equipment classification results are abnormal, an alarm is promptly issued to the operation and maintenance personnel.
[0060] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: by comprehensively collecting and processing various feature data of the equipment, a pattern recognition model based on machine learning is constructed to realize automatic classification and status evaluation of DC equipment types, which not only improves the accuracy and reliability of equipment classification, but also enhances the intelligence and safety of the system through real-time monitoring and early warning functions, effectively ensuring the stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 is a flow chart of the present invention;
[0062] Figure 2 It is the data model framework diagram of the DC equipment system diagram;
[0063] Figure 3 Flowchart of DC device identification using CNN-LSTMD. DETAILED DESCRIPTION
[0064] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.
[0065] like Figure 1 As shown, the present invention comprises the following steps:
[0066] (1) Data collection and preprocessing: cleaning the collected raw data to remove noise and outliers;
[0067] Data collection:
[0068] Equipment characteristic data is collected from various DC devices in the power system, including but not limited to parameters such as voltage, current, power, temperature, and operating time. At the same time, the connection relationship and topological structure data of the DC devices are collected. Sensors and data acquisition devices are used to obtain this information in real time to provide basic data support for subsequent analysis.
[0069] Data preprocessing:
[0070] The collected raw data is cleaned to remove noise and outliers to ensure the accuracy and reliability of the data; in order to optimize the quality of data of different modalities, these data are simultaneously input into the network to avoid the influence of irrelevant factors such as dimension on the final classification effect. A series of preprocessing steps are required to facilitate subsequent feature extraction and pattern recognition. These steps include data normalization and denoising. Among them, data normalization uses z-score standardization to standardize data of different modalities to between 0 and 1, as shown in the following formula:
[0071]
[0072] Among them, x is the data of a certain mode; μ is the mean of the historical data of the mode; σ is the standard deviation of the historical data of the mode. In order to effectively use these data, the data of each mode is integrated into a multi-channel time series, as shown in the following formula:
[0073] X=[I t , V t , T t , V i b t ]
[0074] Among them, X is the data at a single time sampling point of the input data; t is the time step. Through this multimodal data processing method, the unique information contained in each data can be fully utilized to provide a more comprehensive perspective for accurate diagnosis of DC equipment classification.
[0075] (2) Feature extraction based on one-dimensional convolutional neural network;
[0076] By using a feature extraction network based on a one-dimensional convolutional neural network (1D-CNN), the powerful feature extraction ability of convolutional neural networks can be used to extract classification features of different DC devices from historical data. 1D-CNN is designed for processing time series data (such as sensor data, etc.) and extracts local features by sliding the convolution kernel on the time series. In the context of DC device classification and identification, 1D-CNN can effectively capture the time-varying patterns of data such as voltage, current, power, temperature, and operating time, which is crucial for identifying different types of faults. The convolution layer is the core component of the one-dimensional convolutional neural network and is responsible for extracting local features from time series data. For each convolution layer, a convolution kernel (or filter) is used to slide on the input data and the dot product of the convolution kernel and the data is calculated. The convolution operation at time step t can be expressed as:
[0077]
[0078] Among them, D is the input data; K is the convolution kernel; k is the convolution kernel size.
[0079] The activation function is used to introduce nonlinear relationships. In the process of DC device classification and recognition, the nonlinear relationship between the input sensor data and the DC device classification category can be enhanced within the 1D-CNN through the activation function. The traditional activation function usually uses the sigmoid function, but the sigmoid function has the problem of gradient disappearance, which makes the network difficult to train when the model depth is high. Therefore, the present invention uses the optimized ReLu function to enhance the expression ability of the model:
[0080] f(x)=sigmoid(x)
[0081] f(x) = max(0, x)
[0082] Where x is the output of the current neuron. The pooling layer is used to reduce the feature dimension and retain important information while reducing the amount of calculation. Here, the maximum value of the region is selected by maximum pooling to reduce the dimension of the feature, thereby reducing the computational complexity and storage overhead.
[0083] (3) Construct a power diagram model based on the CIM model;
[0084] Establish a DC equipment network topology structure based on the CIM basic model, convert the structural characteristics and connection relationships of the equipment into graph structure data, analyze the topological characteristics of the equipment through graph calculation methods, provide effective characteristics for identifying various types of DC equipment, and have the ability to provide wiring methods.
[0085] The CIM / E model is a specification proposed by State Grid Corporation of China to describe the power system model. For relevant DC equipment, the DC equipment types in the CIM / E model (such as DC converters, DC circuit breakers, etc.) are mapped to vertices in the graph database. Then, the connection relationship between devices (such as DC cables) is mapped to edges. In the graph database, single-terminal components (such as DC converters) are defined as vertices, multi-terminal components (such as DC cables) are defined as edges, and the system topology is formed by connecting vertices and edges. In order to improve the efficiency of database access, switches and knife switches are converted into edges to reduce the number of vertices. In addition, vertices with voltage level and reference voltage identification are established, and their association with load vertices is described by edges. In the topological structure, line information is defined in Line edges, and multi-time load data and equipment status information are defined in Time nodes and Time_connect edges, showing the evolution of the system in spatial and temporal dimensions. Finally, a complete graph data schema is formed through metadata mapping to construct the DC equipment network topology.
[0086] like Figure 2 As shown, by using a graph database, the operating data at different times are collected, realizing the transformation from the traditional storage of "multiple interrelated tables" to the storage of "one interrelated table", and integrating all the spatiotemporal data of DC equipment into "one graph" for fast data storage, query, and modification, thereby improving data management efficiency.
[0087] (4) Construct a DC device classification and recognition framework based on machine learning and input the extracted DC device feature data into the pattern recognition model;
[0088] Construct a DC equipment classification and identification framework based on machine learning, input the extracted DC equipment feature data into the pattern recognition model, train and optimize the model parameters, form a classifier that can distinguish different types of DC equipment, and achieve efficient and accurate equipment classification. To this end, this problem is defined as a supervised learning task. The key is to process and interpret multimodal time series data, including parameters such as voltage, current, power, temperature, and operating time, to determine the specific category and diagnosis of DC equipment. The classification and diagnosis of DC equipment is achieved by fusing two deep learning models, multidimensional convolution and LSTM. Multidimensional convolution is used to extract the features of multimodal data fusion, and LSTM is used to process time series dependencies, thereby achieving a DC equipment classification and diagnosis method that is more suitable for the sensor data in this study. The overall architecture of this method is as follows: Figure 3 As shown,
[0089] (4.1) The processed DC equipment operating status data is divided into four status sets: severe, abnormal, caution and normal, marked as s1, s2, s3, s4. The data in each set is:
[0090] s i ={s i1 ,s i2 ,…,s in}
[0091] In the formula, n is the number of data in the set;
[0092] (4.2) Construct the orthogonal correlation matrix of each state set as follows:
[0093]
[0094] And perform singular value decomposition on it to get the matrix Y i The eigenvalues and eigenvectors are arranged in descending order as follows:
[0095] λ i1 ≥λ i2 ≥…≥λ ik >0
[0096] α i1 , α i2 , …, α ik
[0097] Among them, λ is the eigenvalue; α is the eigenvector;
[0098] (4.3) The cumulative contribution rate of each eigenvalue above is determined by the following formula
[0099]
[0100] Select m principal component vectors of the operating status of the DC equipment;
[0101] (4.4) Calculate the projection of the orthogonal correlation matrix of each state set of the original power supply equipment in the m-dimensional feature space,
[0102] P=A i s i =[α i1 , α i2 , …, α ik ] T ·[s i1 ,s i2 ,…,s in ]
[0103] Among them, A i =[α i1 , α i2 , …, α ik ] r is the characteristic factor matrix;
[0104] (4.5) Calculate the projection of the state information Z of the substation equipment to be evaluated in each state feature space,
[0105] w i =A i Z
[0106] Among them, w i To normalize the projection distance, take the state set with the minimum value as the operating state of the substation;
[0107] (4.6) According to the following formula, the operating status of the DC equipment is obtained, and the faults or defects of the equipment are clearly identified.
[0108]
[0109] Based on the graph computing model and combined with the DC equipment classification and identification algorithm, the computational efficiency and accuracy of the model are improved through parallel technology, and the position of the DC equipment in the topology structure is finally determined. In addition, the power flow calculation can be used to verify whether the DC equipment classification and identification algorithm meets the requirements of system operation, and to determine whether the line power or node voltage exceeds the limit. If an out-of-limit phenomenon occurs, timely alarm and early warning processing can be carried out.
[0110] (5) Verify and evaluate the classification results of the pattern recognition model;
[0111] The classification results of the pattern recognition model are verified and evaluated using methods such as cross-validation and confusion matrix to ensure classification accuracy and stability.
[0112] (5.1) The dataset is randomly divided into multiple subsets through cross-validation, and training and validation are performed alternately in a cycle. The average classification accuracy is calculated to evaluate the generalization ability of the model.
[0113] (5.2) Construct a confusion matrix, compare the actual category with the predicted category, record the classification details, and calculate indicators such as accuracy, precision, recall, and F1-score to comprehensively evaluate the model performance.
[0114] (5.3) Based on the verification results, identify the faults or defects of the equipment. Analyze the deficiencies of the model in classification accuracy and stability, adjust model parameters, perform feature engineering, retrain and verify the model, and ensure that the optimized model has improved classification accuracy and stability. Through continuous iterative optimization, the reliability and application value of the classification results are ultimately guaranteed.
[0115] (6) Integrate the trained pattern recognition model into the monitoring and management platform of the power system.
[0116] The trained pattern recognition model is integrated into the monitoring and management platform of the power system to realize the automatic classification of DC equipment, and provide visual display of classification results and report generation functions to facilitate management and decision-making by operation and maintenance personnel. In addition, the alarm and early warning functions are integrated. When the equipment classification results are abnormal, the operation and maintenance personnel are promptly alerted to improve the safety and intelligence level of the system.
Claims
1. A method for classifying and diagnosing the status of a DC device, characterized in that: The steps include: (1) Data collection and preprocessing: cleaning the collected raw data to remove noise and outliers; (2) Feature extraction based on one-dimensional convolutional neural network; (3) Construct a power diagram model based on the CIM model; (4) Construct a DC device classification and recognition framework based on machine learning and input the extracted DC device feature data into the pattern recognition model; (5) Verify and evaluate the classification results of the pattern recognition model; (6) Integrate the trained pattern recognition model into the monitoring and management platform of the power system.
2. The classification and status diagnosis method of DC equipment according to claim 1, characterized in that: The data collection in step (1) comprises: Equipment characteristic data, including voltage, current, power, temperature and operating time, are collected from various DC devices in the power system. At the same time, the connection relationship and topological structure data of the DC devices are collected. Sensors and data acquisition devices are used to obtain this information in real time to provide basic data support for subsequent analysis.
3. The classification and status diagnosis method of DC equipment according to claim 1, characterized in that: The data preprocessing in step (1) comprises: The collected raw data is cleaned to remove noise and outliers to ensure the accuracy and reliability of the data; data normalization and denoising operations are performed. Among them, data normalization uses z-score standardization to standardize data of different modes to between 0 and 1, as shown in the following formula: Among them, x is the data of a certain mode; μ is the mean of the historical data of the mode; σ is the standard deviation of the historical data of the mode; the data of each mode is integrated into a multi-channel time series, as shown in the following formula: X=[I t ,V t ,T t ,V i b t ] Where X is the data at a single time sampling point of the input data; t is the time step.
4. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The feature extraction of step (2) comprises: The powerful feature extraction capability of convolutional neural networks is used to extract classification features of different DC devices from historical data. The convolution layer is the core component of the one-dimensional convolutional neural network, responsible for extracting local features from time series data. For each convolution layer, the convolution kernel is slid on the input data to calculate the dot product between the convolution kernel and the data. The convolution operation at time step t can be expressed as: Where D is the input data; K is the convolution kernel; k is the convolution kernel size; The activation function is used to introduce nonlinear relationships. In the process of DC device classification and recognition, the nonlinear relationship between the input sensor data and the DC device classification category is enhanced within the 1D-CNN through the activation function; Using the optimized ReLu function, it is expressed as: f(x)=sigmoid(x) f(x) = max(0, x) Among them, x is the output of the current neuron, and the pooling layer is used to reduce the feature dimension.
5. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The method for constructing the power diagram model in step (3) includes: A DC equipment network topology structure based on the CIM basic model is established, and the structural characteristics and connection relationships of the equipment are converted into graph structure data. The topological characteristics of the equipment are analyzed through graph calculation methods. Then, the connection relationships between devices are mapped to edges. In addition, vertices with voltage level and reference voltage identification are established, and their association with load vertices is described by edges. In the topology structure, line information is defined on Line edges, and multi-time load data and equipment status information are defined in Time nodes and Time connect edges, showing the evolution of the system in space and time dimensions. Finally, a complete graph data model is formed through metadata mapping to construct a DC equipment network topology structure.
6. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The CIM basic model is a specification for describing a power system model. For related DC equipment, the DC equipment types in the CIM / E model are mapped to vertices in a graph database.
7. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The step (4) includes constructing a DC device classification and recognition framework based on machine learning, inputting the extracted DC device feature data into a pattern recognition model, training and optimizing model parameters, forming a classifier that can distinguish different types of DC devices, and realizing efficient and accurate device classification.
8. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The pattern recognition in step (4) comprises: (4.1) The processed DC equipment operating status data is divided into four status sets: severe, abnormal, caution and normal, marked as s1, s2, s3, s4. The data in each set is: s i ={s i1 ,s i2 ,…,s in } In the formula, n is the number of data in the set; (4.2) Construct the orthogonal correlation matrix of each state set as follows: And perform singular value decomposition on it to get the matrix Y i The eigenvalues and eigenvectors are arranged in descending order as follows: l i1 ≥λ i2 >…≥λ ik >0 α i1 ,Aα i2 ,…,α ik Among them, λ is the eigenvalue; α is the eigenvector; (4.3) The cumulative contribution rate of each eigenvalue above is determined by the following formula Select m principal component vectors of the operating status of the DC equipment; (4.4) Calculate the projection of the orthogonal correlation matrix of each state set of the original power supply equipment in the m-dimensional feature space, P=A i s i =[a i1 ,a i2 ,…,a ik ] T ·[s i1 ,s i2 ,…,s in ] Among them, A i =[α i1 , α i2 , …, α ik ] r is the characteristic factor matrix; (4.5) Calculate the projection of the state information Z of the substation equipment to be evaluated in each state feature space, In i =A i WITH Among them, w i To normalize the projection distance, take the state set with the minimum value as the operating state of the substation; (4.6) According to the following formula, the operating status of the DC equipment is obtained, and the faults or defects of the equipment are clearly identified. Based on the graph computing model and combined with the DC equipment classification and identification algorithm, the computational efficiency and accuracy of the model are improved through parallel technology, and the position of the DC equipment in the topology structure is finally determined. In addition, the power flow calculation can be used to verify whether the DC equipment classification and identification algorithm meets the requirements of system operation, and to determine whether the line power or node voltage exceeds the limit. If an out-of-limit phenomenon occurs, timely alarm and early warning processing can be carried out.
9. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The step (5) comprises: (5.1) The data set is randomly divided into multiple subsets through cross-validation, and training and validation are performed alternately in a cycle to calculate the average classification accuracy and evaluate the generalization ability of the model; (5.2) Construct a confusion matrix, compare the actual category with the predicted category, record the classification details, and calculate the accuracy, precision, recall and F1-score indicators to comprehensively evaluate the model performance; (5.3) Based on the verification results, identify any faults or defects in the equipment, analyze the deficiencies in the model in terms of classification accuracy and stability, adjust model parameters, perform feature engineering, retrain and verify the model, and ensure that the optimized model has improved classification accuracy and stability.
10. The method for classifying and diagnosing the status of DC equipment according to claim 1, characterized in that: The step (5) includes integrating the trained pattern recognition model into the monitoring and management platform of the power system to realize the automatic classification of DC equipment, and provide a visual display of the classification results and a report generation function, and also integrates an alarm and early warning function. When an abnormality occurs in the equipment classification result, an alarm is promptly issued to the operation and maintenance personnel.
Citation Information
Cited By
Virtual power plant productivity prediction method and system
CN119417071A
