Power secondary system operation data label extraction and management system

Abstract

The invention discloses an electric power secondary system operation data label extraction and management system, and the system comprises a data extraction module which is used for extracting electric power secondary system operation data, dividing the extracted electric power secondary system operation data into three types, and storing the three types in an intermediate database, meanwhile, recording a timestamp for storing the operation data of the power secondary system; the extraction interval division module is in butt joint with the intermediate database and sets data label extraction intervals for label acquisition of the three types of power secondary system operation data; the data label extraction module is used for performing data label extraction on the three types of power secondary system operation data stored in the intermediate database according to the data label extraction intervals divided by the extraction interval division module; and the label management module is connected with the data label acquisition module and is used for receiving and managing the acquired data labels.

Description

Technical Field

[0001] The present application relates to the technical field of label extraction, and in particular to a power secondary system operation data label extraction and management system. Background Art

[0002] The power secondary system refers to all equipment and systems within the power system that perform functions such as monitoring, control, protection, measurement, and dispatch, in addition to the primary equipment directly used for electricity generation, transmission, and distribution. These systems generate a wide variety of data, including but not limited to periodic data such as voltage, current, active power, and reactive power; event-driven data such as equipment anomalies and fault alarms; and specialized data such as secondary equipment status and configuration information. This data is a crucial foundation for the stable operation and efficient management of the power system.

[0003] Data label extraction involves extracting data segments with specific meaning or value from massive amounts of power secondary system data based on specific rules or algorithms, and assigning corresponding labels to them. These labels can be descriptions, classifications, associations, or time series of the data content. They help power system operators better understand the data, analyze the system's operating status, and promptly identify and address potential issues. Data label extraction plays a vital role in power secondary systems and is one of the key means of achieving intelligent and automated power system management.

[0004] Existing data tag extraction technologies for power secondary systems generally suffer from fixed extraction cycles and poor adaptability. Traditional data tag extraction methods often use fixed time intervals, and tag extraction is performed at preset time intervals regardless of data changes. However, due to the complexity and variability of power secondary system data, different types of data have different generation frequencies and change patterns. Therefore, fixed extraction cycles often cannot accurately match the actual data generation situation, resulting in the possibility of missing important information when the data is dense, and invalid tag extraction operations when the data is sparse. This not only wastes system resources, but also may affect the accuracy and completeness of data tags, thereby affecting the operation, maintenance and management efficiency of the power system. Summary of the Invention

[0005] In view of this, an embodiment of the present application provides a power secondary system operation data tag extraction and management system.

[0006] According to one aspect of the present application, a system for extracting and managing power secondary system operation data tags is provided, comprising:

[0007] A data extraction module is used to extract the power secondary system operation data, divide the extracted power secondary system operation data into three categories, and store them in the intermediate database. At the same time, it records the timestamp of the power secondary system operation data storage. The three categories are special data, periodic data, and event-driven data. Special data includes secondary equipment abnormality data and equipment error data. Periodic data includes voltage data, current data, active power data, reactive power data, forward active electric energy data, and reverse active electric energy data. Event-driven data includes switch position, knife switch status, and transformer tap position.

[0008] The extraction interval division module is connected to the intermediate database and sets the data label extraction interval for label collection of three categories of power secondary system operation data in the following way:

[0009] For special data: the data label extraction interval is set to regular extraction;

[0010] For periodic data: the data label extraction interval is divided according to the following formula: A = T + TB, where B is the set threshold, A is the data label extraction interval for periodic data, and T = t n+1 -t n , t n and t n+1 is the timestamp of adjacent data generated by a power secondary device in the periodic data, and T is the timestamp difference of adjacent data generated by a power secondary device in the periodic data;

[0011] For event-driven data, determine the data label extraction interval through the following steps 1 to 3:

[0012] Step 1: Collect time-driven data from the power secondary system operation data in the intermediate database, arrange them in order of timestamps, and divide the arranged event-driven data into time intervals;

[0013] Step 2: Establish a point process model and optimize the parameters of the point process model using maximum likelihood estimation;

[0014] Step 3: Set the data label extraction interval for event-driven data based on the event density output by the point process model;

[0015] a data label extraction module, configured to extract data labels from the three categories of power secondary system operation data stored in the intermediate database according to the data label extraction intervals divided by the extraction interval division module;

[0016] The tag management module is connected to the data tag collection module and is used to receive and manage the collected data tags.

[0017] By means of the above technical solution, the embodiment of the present application provides a power secondary system operation data tag extraction and management system, which has the following beneficial effects:

[0018] Data tag extraction is performed based on time intervals: The system uses segmentation modules to categorize data into special data, periodic data, and event-driven data, and sets different tag extraction intervals based on the characteristics of each type of data. This categorized processing method prevents the system from performing ineffective tag extraction operations when no data is generated, significantly reducing system resource usage.

[0019] Improved data processing efficiency: For periodic data, the system automatically adjusts the frequency of tag extraction based on the time interval between data generation, ensuring timely tag extraction as data is generated. This tag extraction strategy, which adapts to the frequency of data generation, effectively improves data processing efficiency.

[0020] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0022] Figure 1 A schematic diagram of the structure of a power secondary system operation data tag extraction and management system provided by an embodiment of the present application is shown;

[0023] Figure 2 A structural diagram of another power secondary system operation data label extraction and management system provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that, unless there is a conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0025] In this embodiment, a power secondary system operation data tag extraction and management system is provided. Figure 1 As shown, it includes: a data extraction module, an extraction interval division module, a data label extraction module, and a label management module.

[0026] The data extraction module is used to extract the power secondary system operation data, and divide the extracted power secondary system operation data into three categories and store them in the intermediate database, while recording the timestamp of the power secondary system operation data storage. The three categories are special data, periodic data and event-driven data. Special data includes secondary equipment abnormality data and equipment error data. Periodic data includes voltage data, current data, active power data, reactive power data, forward active electric energy data, and reverse active electric energy data. Event-driven data includes switch position, knife switch status, and transformer tap position.

[0027] Optionally, the intermediate database adopts a storage method that combines a relational database and a non-relational database, wherein the relational database is used to store periodic data, and the non-relational database is used to store event-driven data and special data.

[0028] Optionally, the data extraction module adopts an interface protocol that complies with the power secondary system communication standard, including but not limited to the IEC 61850 protocol and the Modbus protocol.

[0029] In an embodiment of the present application, the data extraction module is used to extract the data generated by the power secondary system, and store the extracted data in an intermediate database according to the categories divided by the division module, and record the data storage timestamp. Relational databases (such as MySQL, Oracle, etc.) have strict table structures and data constraints, and are suitable for storing highly structured periodic data. Periodic data, such as voltage, current, power, etc., usually have fixed formats and rules. Relational databases can well ensure the consistency and integrity of the data, while non-relational databases (such as MongoDB, Redis, etc.) have flexible data models and high scalability, and are suitable for storing event-driven data. Event-driven data often has irregular structures and dynamic changes, such as equipment fault alarm information, switch position records, etc. The data extraction module adopts an interface protocol that complies with the power secondary system communication standard, including but not limited to IEC 61850 and Modbus protocols, to adapt to different manufacturers and different types of power secondary equipment.

[0030] It should be noted that to improve data extraction efficiency, the data extraction module can employ multithreading or asynchronous I / O technology. Multithreading allows the module to handle multiple data extraction tasks simultaneously, avoiding the impact of individual task blockages on overall efficiency. Asynchronous I / O allows the module to perform other operations simultaneously during data transfer, fully utilizing system resources. Furthermore, the data extraction module should also possess data filtering and preprocessing capabilities. During data extraction, it should be able to filter out invalid or unnecessary data according to preset rules, while also performing preliminary data cleaning and conversion, such as standardizing data formats and checking data ranges, to improve the efficiency and accuracy of subsequent processing.

[0031] Alternatively, as Figure 2 As shown, the power secondary system operation data label extraction and management system also includes: a classification module, connected to the data extraction module, for using a machine learning model to classify the power secondary system operation data extracted by the data extraction module, so as to classify the power secondary system operation data into special data, periodic data and event-driven data, wherein the machine learning model is constructed and trained through the scikit-learn library of Python or the caret package of R language.

[0032] In an embodiment of the present application, the partitioning module uses a machine learning model to classify the input data, and the machine learning model is constructed and trained through the Python scikit-learn library / R language caret package to identify and classify special data, periodic data and event-driven data. With the help of the Python scikit-learn library or the R language caret package, a powerful classification model can be constructed and trained. The working principle of these models is based on the study of a large amount of historical data, mining the hidden features and patterns in the data, wherein special data is unpredictable and irregular data, including but not limited to secondary equipment abnormal data and equipment error data, periodic data is data generated according to a certain periodic pattern, including but not limited to voltage, current, active power, reactive power, forward active electric energy, reverse active electric energy data, event-driven data is triggered by events, has no fixed period but implicitly has regular data, including but not limited to switch position, knife switch status, transformer tap position. For example, for periodic data, the model learns recurring patterns in time series, such as the cyclical variations in voltage and current data that occur throughout the day, often in response to peaks and troughs in electricity consumption. For special data, the model identifies features associated with equipment failures and abnormal conditions, such as the sudden and irregular nature of secondary equipment abnormality data and equipment error reports. For event-driven data, the model focuses on features that correlate data generation with specific events. By learning and analyzing these features, the model can accurately classify new input data. When building and training machine learning models using the Python scikit-learn library or the R caret package, the first step is data preprocessing. This involves cleaning the collected secondary power system data to remove noise, missing values, and outliers. Feature extraction and engineering are also performed to select features that are important for classification, such as the mean, variance, and rate of change. Next, the preprocessed data is divided into a training set and a test set. The training set is used for model training, while the test set is used to evaluate model performance. Next, based on the data characteristics and the requirements of the classification task, an appropriate machine learning algorithm is selected from the scikit-learn library or the caret package, such as a decision tree, support vector machine, or random forest. During training, model parameters are continuously adjusted and methods such as cross-validation are used to optimize model performance and improve classification accuracy and stability. Finally, the trained model is evaluated using a test set. If the model performs as expected, it can be applied to the classification module to classify new power secondary system data.

[0033] The extraction interval division module is connected to the intermediate database and sets the data label extraction interval for label collection of three categories of power secondary system operation data in the following way:

[0034] For special data: the data label extraction interval is set to regular extraction;

[0035] For periodic data: the data label extraction interval is divided according to the following formula: A = T + TB, where B is the set threshold, A is the data label extraction interval for periodic data, and T = t n+1 -t n , t n and t n+1 is the timestamp of adjacent data generated by a power secondary device in the periodic data, and T is the timestamp difference of adjacent data generated by a power secondary device in the periodic data;

[0036] For event-driven data, determine the data label extraction interval through the following steps 1 to 3:

[0037] Step 1: Collect time-driven data from the power secondary system operation data in the intermediate database, arrange them in order of timestamps, and divide the arranged event-driven data into time intervals;

[0038] Step 2: Establish a point process model and optimize the parameters of the point process model using maximum likelihood estimation;

[0039] Step 3: Set the data label extraction interval for event-driven data based on the event density output by the point process model.

[0040] Optionally, the extraction interval division module sets the data tag extraction interval for event-driven data through the following steps:

[0041] Step 1 includes: dividing the arranged event-driven data into time intervals according to months, wherein the data set of event-driven data of the mth month is set as D m ={t1, t2, t3, ..., t m}, where ti is the timestamp of the event-driven event and m is the total number of timestamps in the month;

[0042] Step 2 includes: Step 2 (1): For each time interval D m The event base rate μ is calculated using the following formula:

[0043]

[0044] Where N is the dataset D m The total number of events in , T is the total number of hours in the month;

[0045] Step 2 (2): Establish the following strength model:

[0046]

[0047] Where t is the current time, ti is the timestamp of each event i in the events before the current time t, e -β(t-ti) is the weight of the trigger effect that decays exponentially with time from event ti to the current time t, λ(t) is the predicted occurrence rate of the event at time t, that is, the event density, is the basic trigger strength, initialized to 0.5μ, β is the decay rate, △t is D m The mean of the timestamp interval;

[0048] For D m ={t1, t2, t3, ..., t m}, construct the following log-likelihood function:

[0049]

[0050] Where λ(ti) is the predicted occurrence rate of the event at time ti;

[0051] Then perform the integral operation using the following formula:

[0052]

[0053] Finally, μ is calculated by gradient descent algorithm. β is optimized;

[0054] Step 3 includes: after calculating λ(t), calculating the data tag collection interval δ of the event-driven event by the following formula:

[0055]

[0056] Where C is a reference constant, which represents the number of events expected to be covered in each acquisition cycle, and ε is a smoothing term that prevents the denominator from being zero.

[0057] The system uses the extraction interval classification module to categorize data into special data, periodic data, and event-driven data, and sets different tag extraction intervals based on the characteristics of each type of data. This categorized processing method allows the system to avoid ineffective tag extraction operations when no data is generated, significantly reducing system resource usage.

[0058] It should be noted that, in order to further improve the accuracy of data label collection and reduce the pressure on the device, the extraction interval division module is optionally used to perform the following steps to optimize the data label extraction interval for event-driven data:

[0059] Step 4: When the data tag extraction module extracts data tags from the data in the intermediate database, the event frequency F (unit: times) of event-driven events in the past day, the remaining storage capacity ratio D (unit: %) of the intermediate database, the data transmission delay L (unit: milliseconds), the device health status index H (the value range is 0-1, 0 indicates abnormality, 1 indicates normal), the CPU usage Uc (unit: %) and the memory usage Um (unit: %) are recorded. At the same time, according to step 2 (2), the intensity function output result λ(t) is found to construct the following initial state vector:

[0060] S t =[λ(t),F,D,L,H,U C ,U M ]

[0061] Step 5: Construct the following adjustment action set:

[0062] A = {+5%, -5%, +10%, -10%, hold}, which represents the percentage adjustment amplitude for the current acquisition interval δ. Then, St from step 4 is input into the neural network, which outputs the probability of each adjustment action in set A and selects the maximum probability.

[0063] Step 6: Adjust the value a of the action corresponding to the maximum probability in step 5 t Adjust the current data tag collection interval δ:

[0064] δ n =δ×(1+a t ), δ n is the adjusted label collection interval;

[0065] Step 7: Collect data labels for event-driven data using the adjusted data label collection interval, monitor the status of the data label extraction module, and construct the following function:

[0066] R t =w1×coverage-w2×resource consumption+w3×real-time score,

[0067] Then, coverage, resource consumption, and real-time performance scores are calculated using the following formula:

[0068]

[0069] Real-time score = e -L / 100 ,

[0070] Then, according to the new data tag collection interval, the state vector S is re-recorded t+1 , and (S t,a t ,R t ,S t+1 ) is stored as a complete set of interaction data;

[0071] Step 8: Repeat steps 4 to 7 until you have multiple sets of interaction data. Then, use these multiple sets of interaction data to construct the following advantage function:

[0072] A t =R t +γV(S t+1 )-V(S t ),

[0073] Among them, V(S t+1 ) and V(S t ) are respectively estimated by the Critic network state vector S t+1 With the estimated value of St, γ is set to 0.99, and then V(S t+1 ) and V(S t ) and the parameters of the neural network are optimized:

[0074] L ppo =min(θ×At, clipθ, 1-ε, 1+ε)×At,

[0075] Where θ is the probability ratio of the new and old adjustment actions, ε is the shear parameter, and its value is 0.2. Repeat steps 4 to 8 until the volatility of the loss function is less than 1%.

[0076] In the embodiment of the present application, the adjustment action is expressed as a percentage adjustment of the current acquisition interval δ, and then St in step 4 is input into the neural network (Actor network), and the probability of each adjustment action in the adjustment action set A is output through the network, and the percentage adjustment of the adjustment action corresponding to the maximum value of the probability is selected as the adjustment value at. For example, if the probability of 10% of the adjustment action in set A is the largest, then the adjustment value is determined to be 0.1. The input layer of the neural network architecture (Actor network) here is the state vector S t , with a dimension of 7. The hidden layer is a fully connected layer (128 neurons) with a ReLU activation function, and the output layer is a fully connected layer (5 neurons). It corresponds to 5 adjustment actions, and the action probability distribution is output through the Softmax function. The current collection interval δ is adjusted according to the action with the highest probability. Data labels are collected for event-driven data using the adjusted collection interval. At the same time, the status of the data label extraction module is monitored by the monitoring subunit, and the following function is constructed:

[0077] R t= w1×coverage-w2×resource consumption+w3×real-time score, where w1, w2 and w3 can be 0.7, 0.15 and 0.15, which are weight parameters. Since the coverage requirement is the greatest when extracting data tags, the coverage weight is the greatest, while the resource consumption and real-time requirements are not so great. The remaining weights are evenly distributed, and then, by calculating the coverage, resource consumption and real-time score. Then, re-record S according to the new collection interval t+1 , and (S t ,a t ,R t ,S t+1 ) is stored as a complete set of interaction data. Repeat steps 4 to 7 until there are multiple sets of interaction data (for example, 100 sets). The number of sets can also be other, which can be determined according to the specific amount of data. For example, if there are 10,000 switch position data in the intermediate database, then the interaction data here can be set to 500 sets. Then, through multiple sets of interaction data, the following advantage function is constructed: t =R t +γV(S t+1 )-V(S t ), where V(S t+1 ) and V(S t ) is the state vector S estimated by the Critic network t+1 With the estimated value of St, γ is set to 0.99, and then V(S t+1 ) and V(S t ) and the parameters of the neural network (Actor) are optimized: L ppo =min(θ×At, clipθ, 1-ε, 1+ε)At, where θ is the probability ratio of the new and old adjustment actions, that is, the ratio of the maximum probability of each adjustment action in set A predicted at time t to the maximum probability of each adjustment action in set A predicted at time t-1, ε is the clipping parameter, which limits the update amplitude and takes a value of 0.2. Here, the value can be between 0.1 and 0.3, and the middle value is taken here. Clip represents a hard truncation operation, and then steps 4 to 8 are repeated until the volatility of the loss function is less than 1%. At this time, it can be said that the loss function converges, and the setting of the data label collection interval for event-driven data is completed.

[0078] Optionally, the extraction interval division module is also used to perform the following steps: when extracting data labels for event-driven data through the optimized data label extraction interval, intermittently calculate the coverage rate; if the coverage rate is less than 99% for more than 3 times, the call point process model is used to extract subsequent data labels, and restart steps 4 to 8.

[0079] It should also be noted that even if the data label collection interval for event-driven data is set, the coverage rate needs to be counted regularly. Once the coverage rate is lower than 99% three times, the fixed formula in the point process model is used to calculate the extraction interval to temporarily collect data labels for event-driven data, and steps four to seven are started again until the loss function converges and the data label collection interval remains unchanged.

[0080] The data label extraction module is used to extract data labels from the three categories of power secondary system operation data stored in the intermediate database according to the data label extraction intervals divided by the extraction interval division module.

[0081] Optionally, the data label extraction module uses the distributed computing framework Apache Spark to perform parallel label extraction on the power secondary system operation data in the intermediate database.

[0082] In an embodiment of the present application, the data label extraction module extracts data labels from the three categories of data stored in the intermediate database according to the extraction intervals divided by the extraction interval division module, and the data label extraction module uses a distributed computing framework (Apache Spark) to perform parallel label extraction on the data in the intermediate database, which can decompose large-scale data processing tasks into multiple subtasks and execute them simultaneously on multiple nodes in the cluster. For the data label extraction module, this means that label extraction operations can be performed on different data partitions in the intermediate database at the same time, greatly shortening the processing time. Taking periodic data as an example, this type of data is usually stored in time series and the data volume is huge. By using the parallel processing function of Apache Spark, data in different time periods can be allocated to different computing nodes for label extraction, thereby significantly improving processing efficiency.

[0083] When the data label extraction module performs parallel label extraction based on Apache Spark, it first reads the corresponding data from the intermediate database, dividing the extraction intervals into intervals determined by the module. This data is then represented using Apache Spark's Resilient Distributed Datasets (RDDs) or DataFrames, both of which support distributed storage and parallel operations. When performing label extraction, the module defines a series of label extraction rules and algorithms and encapsulates them into functions. These functions are applied to the distributed dataset through Apache Spark transformation operations (such as map and filter), extracting labels for each data element. For example, for device error information in special data, the error type can be determined and assigned a corresponding label based on pre-set rules.

[0084] The tag management module is connected to the data tag collection module and is used to receive and manage the collected data tags.

[0085] Optionally, the tag management module is further used to display statistical information and change trends of data tags through visualization tools.

[0086] In an embodiment of the present application, the tag management module is connected to the data tag collection module to receive and manage the collected data tags. The tag management module further includes a visual management function. The visual management function uses visualization tools such as Tableau and Power BI to display the statistical information and change trends of data tags in the form of intuitive charts, graphs and reports. This enables managers to quickly grasp the distribution of data tags, such as the proportion of different types of data tags, the increase and decrease trend of the number of tags in a specific time period, etc. Through intuitive visual display, abnormal situations in data tags can be discovered in time, such as a sudden and significant increase or decrease in the number of tags of a certain device, which may indicate that there is a problem with the operation of the equipment, thereby providing strong support for the operation and management of the power system.

[0087] By applying the technical solution of this embodiment, the following beneficial effects can be achieved:

[0088] 1. Data label extraction based on time intervals: The system divides data into special data, periodic data, and event-driven data through segmentation modules, and sets different label extraction intervals based on the characteristics of each type of data. This categorized processing method allows the system to avoid ineffective label extraction operations when no data is generated, significantly reducing system resource usage.

[0089] 2. Improved data processing efficiency: For periodic data, the system automatically adjusts the frequency of tag extraction based on the time interval between data generation, ensuring timely tag extraction as data is generated. This tag extraction strategy, which adapts to the frequency of data generation, effectively improves data processing efficiency.

[0090] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A system for extracting and managing power secondary system operation data tags, characterized in that: include: A data extraction module is used to extract the power secondary system operation data, divide the extracted power secondary system operation data into three categories, and store them in the intermediate database. At the same time, it records the timestamp of the power secondary system operation data storage. The three categories are special data, periodic data, and event-driven data. Special data includes secondary equipment abnormality data and equipment error data. Periodic data includes voltage data, current data, active power data, reactive power data, forward active electric energy data, and reverse active electric energy data. Event-driven data includes switch position, knife switch status, and transformer tap position. The extraction interval division module is connected to the intermediate database and sets the data label extraction interval for label collection of three categories of power secondary system operation data in the following way: For special data: the data label extraction interval is set to regular extraction; For periodic data: the data label extraction interval is divided according to the following formula: A = T + TB, where B is the set threshold, A is the data label extraction interval for periodic data, and T = t n+1 -t n , t n and t n+1 is the timestamp of adjacent data generated by a power secondary device in the periodic data, and T is the timestamp difference of adjacent data generated by a power secondary device in the periodic data; For event-driven data, determine the data label extraction interval through the following steps 1 to 3: Step 1: Collect time-driven data from the power secondary system operation data in the intermediate database, arrange them in order of timestamps, and divide the arranged event-driven data into time intervals; Step 2: Establish a point process model and optimize the parameters of the point process model using maximum likelihood estimation; Step 3: Set the data label extraction interval for event-driven data based on the event density output by the point process model; a data label extraction module, configured to extract data labels from the three categories of power secondary system operation data stored in the intermediate database according to the data label extraction intervals divided by the extraction interval division module; The tag management module is connected to the data tag collection module and is used to receive and manage the collected data tags.

2. The system for extracting and managing power secondary system operation data tags according to claim 1, characterized in that: Also includes: The partitioning module is connected to the data extraction module and is used to classify the power secondary system operation data extracted by the data extraction module using a machine learning model to classify the power secondary system operation data into special data, periodic data and event-driven data, wherein the machine learning model is constructed and trained using the scikit-learn library of Python or the caret package of R language.

3. The system for extracting and managing power secondary system operation data tags according to claim 1, characterized in that: The intermediate database adopts a storage method that combines a relational database with a non-relational database, wherein the relational database is used to store periodic data, and the non-relational database is used to store event-driven data and special data.

4. The system for extracting and managing power secondary system operation data tags according to claim 1, characterized in that: The data extraction module adopts an interface protocol that complies with the power secondary system communication standard, including but not limited to the IEC61850 protocol and the Modbus protocol.

5. The system for extracting and managing power secondary system operation data tags according to claim 1, characterized in that: The data label extraction module uses the distributed computing framework Apache Spark to perform parallel label extraction on the power secondary system operation data in the intermediate database.

6. The system for extracting and managing power secondary system operation data tags according to claim 1, characterized in that: The tag management module is also used to display statistical information and change trends of data tags through visualization tools.

7. The system for extracting and managing power secondary system operation data tags according to claim 6, characterized in that: The extraction interval division module sets the data tag extraction interval for event-driven data through the following steps: Step 1 includes: dividing the arranged event-driven data into time intervals according to months, wherein the data set of event-driven data of the mth month is D m ={t1, t2, t3, ..., t m }, where t i is the timestamp of the event-driven event, and m is the total number of timestamps in the month; Step 2 includes: Step 2 (1): For each time interval D m The event base rate μ is calculated using the following formula: Where N is the dataset D m The total number of events in , T is the total number of hours in the month; Step 2 (2): Establish the following strength model: Where t is the current time, ti is the timestamp of each event i in the events before the current time t, e -β(t-ti) is the weight of the trigger effect that decays exponentially with time from event ti to the current time t, λ(t) is the predicted occurrence rate of the event at time t, that is, the event density, is the basic trigger strength, initialized to 0.5μ, β is the decay rate, △t is D m The mean of the timestamp interval; For D m ={t1, t2, t3, ..., t m }, construct the following log-likelihood function: Where λ(ti) is the predicted occurrence rate of the event at time ti; Then perform the integral operation using the following formula: Finally, μ is calculated by gradient descent algorithm. β is optimized; Step 3 includes: after calculating λ(t), calculating the data tag collection interval δ of the event-driven event by the following formula: Where C is a reference constant, which represents the number of events expected to be covered in each acquisition cycle, and ε is a smoothing term that prevents the denominator from being zero.

8. The system for extracting and managing power secondary system operation data tags according to claim 7, characterized in that: The extraction interval partitioning module is also used to perform the following steps to optimize the data label extraction interval for event-driven data: Step 4: When the data tag extraction module extracts data tags from the data in the intermediate database, the event frequency F (unit: times) of event-driven events in the past day, the remaining storage capacity ratio D (unit: %) of the intermediate database, the data transmission delay L (unit: milliseconds), the device health status index H (the value range is 0-1, 0 indicates abnormality, 1 indicates normal), the CPU usage Uc (unit: %) and the memory usage Um (unit: %) are recorded. At the same time, according to step 2 (2), the intensity function output result λ(t) is found to construct the following initial state vector: S t =[λ(t),F,D,L,H,UC,U M ] Step 5: Construct the following adjustment action set: A = {+5%, -5%, +10%, -10%, hold], where this set represents the percentage adjustment amplitude for the current acquisition interval δ. Then, St from step 4 is input into the neural network, which outputs the probability of each adjustment action in the adjustment action set A and selects the maximum probability. Step 6: Adjust the value a of the action corresponding to the maximum probability in step 5 t Adjust the current data tag collection interval δ: δ n =δ×(1+a t ), δ n is the adjusted label collection interval; Step 7: Collect data labels for event-driven data using the adjusted data label collection interval, monitor the status of the data label extraction module, and construct the following function: R t =w1×coverage-w2×resource consumption+w3×real-time score, Then, coverage, resource consumption, and real-time performance scores are calculated using the following formula: Real-time score = e -L / 100 , Then, according to the new data tag collection interval, the state vector S is re-recorded t+1 , and (S t ,a t ,R t ,S t+1 ) is stored as a complete set of interaction data; Step 8: Repeat steps 4 to 7 until you have multiple sets of interaction data. Then, use these multiple sets of interaction data to construct the following advantage function: From t =R t +γV(S t+1 )-V(S t ) Among them, V(S t+1 ) and V(S t ) are respectively estimated by the Critic network state vector S t+1 With the estimated value of St, γ is set to 0.99, and then V(S t+1 ) and V(S t ) and the parameters of the neural network are optimized: L ppo =min(θ×At, clipθ, 1-ε, 1+ε)×At, Where θ is the probability ratio of the new and old adjustment actions, and ε is the shear parameter, which ranges from 0.1 to 0.

3. Repeat steps 4 to 8 until the volatility of the loss function is less than 1%.

9. The system for extracting and managing power secondary system operation data tags according to claim 8, characterized in that: The extraction interval division module is also used to perform the following steps: when extracting data labels for event-driven data through the optimized data label extraction interval, the coverage is intermittently calculated. If the coverage is less than 99% for more than 3 times, the call point process model is used to extract subsequent data labels and restart steps 4 to 8.

Images