Data acquisition control system and method for smart grid

Abstract

The application provides a data acquisition control system and method for a smart power grid, relates to the technical field of big data processing and analysis, and comprises the following steps: a data type determination module uploads a power grid supervision task, determines first and second data types through task interpretation and data sampling dimension conversion, and outputs a task causal diagram; a micro-generator construction module constructs a task supervision unit according to the task causal diagram; and a target power grid management and control module activates the task supervision unit when the task is started, collects and controls a power grid end through the second data type, generates the first data type based on the micro-generator, evaluates a state of the task causal diagram, and controls the target power grid. The application solves the technical problem that a traditional smart power grid data processing mode is difficult to adapt to complex supervision tasks, resulting in insufficient pertinence of data acquisition and precision of control, and achieves the technical effects of accurate data acquisition and efficient control, and improved operation monitoring and regulation capacity of the smart power grid.

Description

Technical Field

[0001] This application relates to the field of big data processing and analysis technology, specifically to a data acquisition control system and method for smart grids. Background Technology

[0002] In smart grids, accurate data collection and effective management are crucial for stable grid operation and efficient dispatch. Current technologies primarily employ traditional data acquisition equipment and centralized processing systems for grid data management. While these methods have proven effective in grid environments with simple structures and small data volumes, their limitations have become increasingly apparent as smart grids expand in scale and complexity. Traditional technologies, lacking precise interpretation of regulatory tasks and appropriate data categorization, struggle to adapt to the multi-dimensional data demands of complex grids. This results in insufficient data targeting, low processing efficiency, and an inability to meet the requirements of refined monitoring and dynamic management in smart grids. Summary of the Invention

[0003] This application provides a data acquisition control system and method for smart grids, which solves the technical problem that traditional smart grid data processing methods are difficult to adapt to complex regulatory tasks, resulting in insufficient targeting of data acquisition and accuracy of control. It achieves the technical effect of accurate data acquisition and efficient control, and improves the operation monitoring and regulation capabilities of smart grids.

[0004] In view of the above problems, on the one hand, this application provides a data acquisition and control system for smart grids, the system comprising: a data class determination module, used to upload grid supervision tasks to a grid management platform, and determine a first data class and a second data class by performing dimensional transformation of task interpretation and data sampling, wherein a task causal graph is output after task interpretation; a micro-generator construction module, used to construct a task supervision unit based on the first data class, the second data class and the task causal graph, wherein a micro-generator constructed by first-order adversarial training based on data class transformation and second-order lightweight distillation training based on directional transformation generation relationship is distributed between the first data class and the second data class; and a target grid control module, used to activate the task supervision unit embedded in the grid management platform upon task startup, perform data acquisition and control of the grid end using the second data class, and perform grid element status assessment based on the first data class generated by the micro-generator and the task causal graph to control the target grid based on the grid supervision task.

[0005] On the other hand, this application also provides a data acquisition and control method for smart grids. The method includes: uploading a power grid supervision task to a power grid management platform; determining a first data class and a second data class by performing dimensional transformation of task interpretation and data sampling, wherein a task causal graph is output after task interpretation; constructing a task supervision unit based on the first data class, the second data class, and the task causal graph, wherein a micro-generator constructed by first-order adversarial training based on data class transformation and second-order lightweight distillation training based on directional transformation generation relationship is distributed between the first data class and the second data class; and activating the task supervision unit embedded in the power grid management platform upon task startup, performing data acquisition and control of the power grid end using the second data class, and performing control of the target power grid based on the power grid supervision task by generating the first data class based on the micro-generator and assessing the state of power grid elements based on the task causal graph.

[0006] One or more technical solutions provided in this application have at least the following technical effects:

[0007] The data class determination module uploads power grid monitoring tasks to the power grid management platform, performs task interpretation and data sampling dimensional transformation, and determines the first data class, the second data class, and the task causal graph, providing the basic data and logical framework for subsequent data processing and monitoring. The micro-generator construction module, based on the first data class, the second data class, and the task causal graph, constructs a task monitoring unit containing a distributed micro-generator (formed through first-order adversarial training and second-order lightweight distillation training), establishing the core processing architecture for data transformation and power grid monitoring. The target power grid control module activates the task monitoring unit upon task startup, using the second data class for power grid data collection and control, generating the first data class through the micro-generator, and assessing the state of power grid elements in conjunction with the task causal graph. This achieves precise control of the target power grid based on the monitoring task, ensuring the efficient and stable operation of the smart grid.

[0008] In summary, this application uploads power grid monitoring tasks to the power grid management platform, determines the first data class, the second data class, and the task causal graph through task interpretation and data sampling dimensional transformation, constructs a task monitoring unit containing a distributed micro-generator based on adversarial training and lightweight distillation training, and collects power grid data using the second data class after unit activation. The micro-generator generates the first data class, and the power grid element status is evaluated in conjunction with the task causal graph, thereby achieving control over the target power grid based on the monitoring task, making smart grid data collection and control more accurate and efficient.

[0009] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the structure of a data acquisition and control system for a smart grid provided in an embodiment of this application.

[0012] Figure 2 This is a flowchart illustrating a data acquisition and control method for a smart grid provided in an embodiment of this application.

[0013] Figure labeling: Data class determination module 10, micro generator construction module 20, target power grid management module 30. Detailed Implementation

[0014] This application provides a data acquisition control system and method for smart grids, which solves the technical problem that traditional smart grid data processing methods are difficult to adapt to complex regulatory tasks, resulting in insufficient targeting of data acquisition and accuracy of control. It achieves the technical effect of accurate data acquisition and efficient control, and improves the operation monitoring and regulation capabilities of smart grids.

[0015] Example 1, as Figure 1 As shown in the figure, this application provides a data acquisition and control system for a smart grid, the system comprising:

[0016] The data class determination module 10 is used to upload power grid supervision tasks to the power grid management platform. By performing task interpretation and data sampling dimension transformation, it determines the first data class and the second data class. The task interpretation is accompanied by the output of a task cause-effect graph.

[0017] In this embodiment, the power grid monitoring task is a specific task used to monitor the smart grid and needs to be uploaded to the power grid management platform. The power grid management platform is a system platform that receives power grid monitoring tasks, embeds a task monitoring unit, processes power grid data, and realizes the monitoring of the smart grid. It can complete functions such as data acquisition, processing, and target power grid control.

[0018] Specifically, users or system maintenance personnel upload power grid monitoring tasks to the power grid management platform. The uploaded tasks must include key information such as specific monitoring objectives (e.g., peak load monitoring in a certain area, transformer operation status assessment), the affected power grid area, and the time period. Upon receiving the power grid monitoring task, the power grid management platform first verifies the format of the task to confirm whether it contains necessary parameters. If any information is missing, such as the monitoring time not being clearly defined, it automatically provides supplementary prompts until the power grid monitoring task information is complete.

[0019] Subsequently, the power grid management platform initiates the task interpretation process, using its built-in semantic parsing module to structure the task text. This semantic parsing module is a functional module used to interpret uploaded power grid monitoring tasks. Its core function is to use natural language processing algorithms to structure the monitoring tasks described in natural language. For example, for the task of monitoring the peak load and transformer temperature fluctuations of a certain area's distribution network, the semantic parsing module first identifies the task type as operational status monitoring, then extracts the core monitoring objects as distribution network load and transformer temperature, and further clarifies the monitoring indicators as peak load and temperature fluctuation amplitude. Simultaneously, by combining the information from the power grid equipment ledger to be monitored, it matches specific equipment identifiers related to the distribution network in that area, transforming the natural language description of the task into parameterized instructions that the system can recognize.

[0020] After task interpretation, the system connects to the power grid database and constructs a power grid causal graph using power grid regulatory elements and element data mining data. It separates spurious correlations and performs first-level pruning to determine the first-order causal graph. Then, it performs second-level pruning on the first-order causal graph using task interpretation information to determine the task causal graph. The specific steps are explained in detail in the power grid causal graph construction unit to the task causal graph determination unit.

[0021] Finally, based on the task cause-effect graph, the task element data class and task supervision element are identified, and after mapping to form multiple mapping groups, they are traversed and the corresponding dimension transformation is performed to determine the first data class and the second data class. The specific steps are explained in detail in the task supervision element identification unit to the first and second data class determination unit.

[0022] Through the above steps, the system accurately completed task interpretation and data classification, providing a structured data foundation and logical framework for the subsequent construction of task supervision units, and ensuring the relevance and effectiveness of data collection and processing.

[0023] The micro-generator construction module 20 is used to construct a task supervision unit based on the first data class, the second data class and the task causal graph. The first data class and the second data class are distributed with micro-generators constructed by first-order adversarial training based on data class transformation and second-order lightweight distillation training based on directional transformation generation relationship.

[0024] Specifically, firstly, using the task cause-effect graph as the base layer, the first data class and the second data class are mapped to graph nodes to determine the first data layer and the second data layer, respectively. The three are then cascaded to determine the task supervision architecture, thereby constructing the task supervision unit. The specific steps for determining the unit in the second data layer and the task supervision architecture unit are explained in detail.

[0025] After determining the task supervision architecture, the generation plugin and the judgment plugin are deployed. After cascading, the data generator is determined through adversarial supervision training. Then, it is migrated and trained and distributed in the task supervision architecture to determine the task supervision unit of the built-in functional area of ​​the power grid management platform. The specific steps are detailed in the description of the sub-units from the judgment plugin deployment to the task supervision unit determination.

[0026] The target power grid control module 30 is used to activate the task supervision unit embedded in the power grid management platform when the task starts. It uses the second data class to collect and control data on the power grid side, and uses the first data class based on the micro generator to generate and assess the state of power grid elements based on the task causal graph to control the target power grid based on the power grid supervision task.

[0027] Specifically, a grid sampling instruction is generated upon the initiation of the grid monitoring task. Based on this instruction, the task monitoring unit is activated and multiple sampling instructions are determined based on the second data layer. Then, grid-side sampling is transmitted back and distributed in the second data layer. The specific steps are explained in detail in the grid sampling instruction generation unit to the distributed writing execution unit.

[0028] Next, the target power grid is generated based on the first data class generated by the micro-generator and the power grid element status assessment and control based on the task causal graph. The first data layer is updated by the distributed micro-generator, and the regulatory results are determined based on its mapping with the base layer and displayed in a pop-up window on the power grid management platform. The specific steps are explained in detail from the first data layer update unit to the task regulatory result display unit.

[0029] In one possible implementation, the data class determination module 10 further includes:

[0030] The power grid causal graph construction unit connects to the power grid database, performs data mining based on power grid regulatory elements and element data classes, and constructs a power grid causal graph, wherein the power grid regulatory elements are the smallest element units; the spurious correlation coefficient determination unit performs first-level pruning on the power grid causal graph by separating spurious correlations to determine the first-order causal graph, and determines the spurious correlation coefficient by balancing data relationships and data values; the task causal graph determination unit performs second-level pruning on the first-order causal graph based on task interpretation information to determine the task causal graph.

[0031] In this embodiment, the power grid database is a data source that stores power grid regulatory elements and their corresponding element data classes. After the system connects, it can perform data mining to extract the relationships between elements, providing foundational data for constructing a power grid causal graph. The power grid causal graph is a graph constructed by mining the relationships between power grid regulatory elements and element data classes after the system connects to the power grid database, containing the potential influence relationships between each element. False correlations refer to combinations of elements in the power grid causal graph that appear to be related but have no actual causal relationship.

[0032] Specifically, after the task interpretation is completed, the system first connects to the power grid database to extract power grid regulatory elements and their corresponding element data classes. Power grid regulatory elements are the smallest unit of data, such as line voltage, equipment load, and transformer temperature. Element data classes include data types such as real-time monitoring values, historical fluctuation curves, and threshold ranges for these power grid regulatory elements. Through data mining techniques, the system analyzes the relationships between these elements and data classes, such as the linkage between line voltage and equipment load changes, and the correlation between transformer temperature and operating time. This process initially constructs a power grid causal graph containing multiple sets of element relationships, visually presenting the potential impact relationships between various elements.

[0033] Taking association rule mining technology as an example, real-time current values ​​and line joint temperatures of a 110kV line were selected. Both are power grid monitoring factors, and the data types consist of current and temperature data collected continuously for 72 hours, once every 10 minutes, totaling 432 samples. Analysis using the Apriori algorithm, with a support of ≥3% and a confidence level of ≥75%, revealed that when the current value exceeded 350A for 15 consecutive minutes, the joint temperature exceeded 65℃ 18 times after 30 minutes. The support was 3.7% and the confidence level was 83%, meeting the threshold conditions and indicating a significant correlation between the two. This correlation was incorporated into the power grid causal graph to visually present the potential impact of current overload on the increase in line joint temperature.

[0034] Subsequently, the system enters the first-level pruning phase, focusing on separating spurious correlations. Spurious correlations refer to combinations of elements that appear to be related but lack actual causal relationships, such as the synchronous fluctuations in light intensity in a certain area and the switching status of a substation over a certain period. By balancing data relationships and data value, spurious correlations are calculated, where the balanced data relationship represents the strength of the actual causal influence between each element, and the data value represents the actual reference significance for power grid supervision. Simultaneously, a coefficient threshold is set, and correlations with coefficients below this value are judged as spurious correlations. After calculation, spurious correlations are identified and removed from the initial multiple sets of correlations, resulting in a first-order causal graph of the remaining true correlations.

[0035] Specifically, when calculating the spurious correlation coefficient, a causal inference model is first constructed. This model is based on power grid regulatory elements (such as voltage and load) and corresponding data types (real-time monitoring values, historical fluctuation data, etc.) extracted from the power grid database. After data cleaning and standardization, the model is trained using a causal forest algorithm. Its input consists of correlation data between elements, including time series and numerical features. The output is a quantified strength of causal influence between elements, ranging from 0 to 1, where 1 represents extremely strong causality and 0 represents no causality. This supports the calculation of the spurious correlation coefficient and helps separate spurious correlations.

[0036] Next, the data relationships are quantified using a causal inference model: for each group of elements in the power grid causal diagram, the actual causal influence strength is calculated and represented by a value in the range of 0-1, where 1 represents extremely strong causality and 0 represents no causality. For example, the causal strength of the relationship between line current and equipment temperature in a certain group is 0.7. Then, data value assessment is combined: based on the core objectives of current power grid supervision tasks, such as the safe operation of equipment, those skilled in the art score the actual reference significance of the relationship, also in the range of 0-1. The above relationship, because it directly affects equipment safety, has a value score of 0.8. Subsequently, the two are weighted and balanced using the formula: spurious correlation coefficient = causal strength × value score. The calculated spurious correlation coefficient for this group is 0.56.

[0037] When setting the coefficient threshold, the lowest coefficient value verified as a valid association in historical monitoring is considered. Assuming the lowest coefficient of the past 300 valid associations is 0.25, and referring to the importance assessment of key power grid element associations by experts in the field, the coefficient threshold is set at 0.25. When the calculated coefficient of a certain association is below 0.25, it indicates either weak causal strength, low value to power grid monitoring tasks, or both. For example, the causal strength of regional humidity and disconnector status is 0.3, the value score is 0.6, and the calculated coefficient is 0.18, which is below the coefficient threshold and is therefore judged as a spurious association. This method accurately distinguishes between genuine and spurious associations.

[0038] Finally, secondary pruning is performed based on the task interpretation information. Assuming this power grid supervision task focuses on load balancing in the distribution network, the task interpretation information clearly identifies the core focus elements as line load, switching switch status, and regional electricity consumption. Based on this, correlations unrelated to load balancing in the first-order causal graph are eliminated, such as the correlation between transformer insulation resistance and ambient humidity. After removing these non-core correlations, a task causal graph containing multiple sets of core correlations is finally determined. This task causal graph retains only the causal relationships of elements directly related to the current power grid supervision task.

[0039] By connecting to the database to build an initial association graph, first-level pruning to remove false associations, and second-level pruning to focus on the core associations of the task, the causal relationships highly related to the power grid supervision task were accurately determined, providing a reliable logical foundation for subsequent data classification and supervision unit construction.

[0040] In one possible implementation, the data class determination module 10 further includes:

[0041] The task supervision element identification unit is used to identify task element data classes and task supervision elements based on the task cause-effect graph; the mapping group determination unit is used to map the task element data classes and task supervision elements to determine multiple mapping groups; the first and second data class determination unit is used to traverse the multiple mapping groups, perform a first data dimension transformation to determine a first data class, and perform a second data transformation to determine a second data class.

[0042] Specifically, based on the clearly defined core element relationships in the task causal graph, the system begins to identify task element data classes and task monitoring elements. For each set of relationships, the system breaks down the objects requiring key monitoring, i.e., the task monitoring elements, and simultaneously breaks down the basic data types supporting its analysis, i.e., the task element data classes. For example, in the core relationship corresponding to line load fluctuation → switching switch operation frequency, the task monitoring elements are the line load fluctuation amplitude and the switching switch operation frequency. The corresponding task element data classes include the real-time sampled value of the line load, the 5-minute moving average, the 24-hour peak record, as well as the timestamp of the switch operation and the daily operation count statistics. After traversing all sets of core element relationships, multiple task monitoring elements and task element data classes were identified. Among them, the task element data class corresponding to the task monitoring element of transformer load rate includes instantaneous load value, load rate exceeding the standard duration, and three-phase load imbalance.

[0043] The mapping phase then begins, matching task element data classes with task supervision elements based on the principle that each task supervision element must be fully described by its corresponding task element data class. For example, the task supervision element of regional electricity consumption peak-valley difference is matched with its task element data classes as regional hourly electricity consumption, daily peak electricity consumption, daily valley electricity consumption, and peak-valley occurrence time, forming one mapping group. The task supervision element of cable joint temperature trend is matched with real-time temperature, 1-hour temperature change, and historical temperature curves under the same operating conditions, forming another mapping group. During the process, the coverage of task element data classes with task supervision elements needs to be verified, with a coverage requirement of ≥90%. For combinations with insufficient coverage, corresponding task element data classes are added. Finally, multiple mapping groups are determined, each containing one core task supervision element and 2-4 corresponding task element data classes.

[0044] The calculation of the coverage of the aforementioned task element data class to the task supervision elements requires first identifying the core evaluation dimensions of the task supervision elements. For example, a task supervision element may include dimensions such as trend, threshold, and correlation impact. Then, the number of core dimensions already covered by the existing task element data class is counted, and the percentage is obtained by dividing the number of covered dimensions by the total number of core dimensions. A feature matching algorithm is used in the process to compare the descriptive features of the task element data class with the semantic and functional relationships of the core dimensions of the task supervision elements to help determine whether there is coverage, thereby calculating the coverage degree.

[0045] Finally, for each mapping group, the first data class is determined by the first data dimension transformation, which is guided by data value and performs dimensionality increase or decrease. The second data class is determined by the second data dimension transformation, which is guided by lightweight data collection and performs minimum-order dimensionality decrease. The specific steps for determining the first and second data class sub-units are explained in detail.

[0046] By accurately identifying key elements and data types from the task cause-effect graph and establishing a one-to-one mapping relationship, clear objects and association basis are provided for subsequent data dimension transformation, ensuring that the transformed first and second data classes can accurately support the power grid supervision needs.

[0047] In one possible implementation, the first second data class determining unit further includes:

[0048] The first data class determination subunit is used to perform a first data dimension transformation on the task element data within each mapping group based on the data value orientation of the task supervision elements within the group, and determine the first data class, wherein the first data dimension is transformed into either dimensionality increase or dimensionality reduction; the second data class determination subunit is used to perform a second data dimension transformation on the task element data within each mapping group based on the lightweight collection orientation, and determine the second data class, wherein the second data dimension is transformed into the minimum-order data dimensionality reduction processing, and the second data class is determined.

[0049] Specifically, when performing the first data dimension transformation for each mapping group, the core data requirements of the task supervision elements within the group are first analyzed, and the transformation method is determined with the goal of maximizing data value. For example, in a certain mapping group, the task supervision element is cable line aging risk assessment, and the corresponding task element data class is cable sheath temperature. The original data is a single-dimensional temperature sampling value of one value per hour. Because assessing aging risk requires combining temperature fluctuation trends and the duration of sustained high temperatures, the system chooses to upgrade the dimension transformation: expanding the data to include three dimensions: instantaneous temperature every 10 minutes, hourly average temperature, and the cumulative duration exceeding the threshold of 60℃ for three consecutive hours. The data volume increases from 12 records per day to 144 records per day, forming the first data class for this group, meeting the risk assessment's need for multi-dimensional data. Another mapping group's task monitoring element is bus voltage stability monitoring. The original data includes three dimensions: instantaneous voltage value, frequency, and harmonic content. After analysis, those skilled in the art found that frequency data has low value for current stability monitoring. Assuming that the correlation between frequency fluctuation and voltage instability in historical data is only 0.12, a dimensionality reduction transformation is adopted to remove the frequency dimension and retain the instantaneous voltage value and harmonic content, forming a concise but high-value first data class.

[0050] In determining the value of task element data classes to task regulatory elements, the core objectives of the regulatory elements are first clarified. For example, the core of voltage stability monitoring is to identify abnormal voltage fluctuations. Then, historical correlation data between the task element data class and the task regulatory elements is extracted. The Pearson correlation coefficient algorithm is used to calculate the degree of linear correlation between the two, resulting in a quantified correlation. The correlation value ranges from 0 to 1, with a value closer to 1 indicating a stronger correlation. Simultaneously, a value judgment threshold is set based on domain experience. Generally, a correlation value < 0.2 indicates that the task element data class has limited explanatory or predictive value for the task regulatory objectives. When the calculated correlation value is lower than the value judgment threshold of 0.2, it indicates a weak correlation between the task element data class and the task regulatory elements, contributing little to supporting the achievement of regulatory objectives, and thus the task element data class is deemed to have low value.

[0051] When performing the second data dimension transformation, the core objective is to reduce data acquisition complexity. Minimum-order dimensionality reduction is applied to the first data class of each mapping group. For example, in the cable line aging risk assessment mapping group mentioned above, the first data class contains 144 multi-dimensional data points per day. By analyzing data correlations, the system finds that the hourly average temperature and the duration of continuous exceedances of thresholds can be derived from the instantaneous temperature every 10 minutes. Therefore, only the instantaneous temperature every 30 minutes is retained, reducing the daily data from 144 to 48. Derived dimensions are removed, forming the second data class for this group. The data volume is reduced, but the core acquisition information is retained. Similarly, for the bus voltage stability monitoring mapping group mentioned above, the first data class consists of two dimensions: instantaneous voltage value and harmonic content. The frequency dimension has been removed. After verification, the instantaneous value every 15 minutes can reflect the overall voltage trend. Therefore, the sampling interval is adjusted to 15 minutes, retaining the instantaneous voltage value. The harmonic content can be calculated and derived in the background, forming the second data class. This significantly reduces the acquisition load while ensuring monitoring requirements are met.

[0052] By flexibly upgrading or downgrading the dimensions based on data value to determine the first data class, and by using a lightweight approach to perform minimum-order dimensionality reduction to determine the second data class, the needs of power grid supervision for multi-dimensional, high-value data are met, while also achieving high efficiency in front-end data collection. This lays a precise and efficient foundation for subsequent data processing and control.

[0053] In one possible implementation, the micro-generator building module 20 further includes:

[0054] The second data layer determination unit is used to determine the first data layer by performing graph node mapping of the first data class and graph node mapping of the second data class, based on the task causal graph as the base layer; the task supervision architecture determination unit is used to cascade the base layer, the first data layer and the second data layer to determine the task supervision architecture.

[0055] Specifically, when deploying the micro-generator in the task supervision unit, 1000 sets of matching historical data are first selected from the power grid database. The second data class consists of low-dimensional sampled data, while the first data class consists of the corresponding high-dimensional analysis data. An adversarial neural network architecture is adopted, with the micro-generator's initial parameters randomly set. Using the second data class as input, it outputs a simulated high-dimensional first data class through a multilayer perceptron. The discriminator is also a multilayer perceptron, taking the generated data and the real first data class as input and outputting the probability that the data is the true value, with a probability range of 0-1. During training, the micro-generator aims to minimize the discriminator's recognition accuracy, while the discriminator aims to maximize its ability to distinguish generated data. The parameters of both are updated after each iteration. After multiple iterations, when the discriminator's misclassification rate for generated data stabilizes at 8%, i.e., the generated data accuracy reaches 92%, the first-order adversarial training stops. At this point, the micro-generator can accurately generate high-dimensional analysis data based on the low-dimensional sampled data, laying the foundation for the subsequent lightweight processing of the micro-generator.

[0056] Following this, a second-order lightweight distillation training process is performed. In this stage, the model obtained after the first-order adversarial training, capable of generating the first data class from the second data class, is used as the teacher model. The teacher model refers to a pre-trained model with rich knowledge. Based on this, a more streamlined student model is constructed. The student model is a lightweight model that learns the knowledge of the teacher model, i.e., a micro-generator. This is achieved by pruning redundant neurons and network layers in the teacher model, reducing its parameter size by 60%. Simultaneously, a feature mapping algorithm extracts 75% of the key parameters related to the core logic of data transformation in the teacher model, such as the low-dimensional to high-dimensional feature transformation matrix and non-linear activation weights, retaining its core transformation capabilities. During training, the probability distribution of the high-dimensional data generated by the teacher model is used as a soft label, combined with the real first data class as a hard label, to guide the student model's learning. After multiple iterations, the error between the sample student model's generated data and the real first data class only increases by 2%, but the parameter simplification and computational path optimization improve the response speed by 40%. Finally, based on the transformation requirements of the first and second data classes, these student models, i.e. micro-generators, are distributed and deployed to the corresponding nodes: for example, one micro-generator focused on load characteristic transformation is deployed on the line load data transformation node, and one micro-generator focused on temperature characteristic transformation is deployed on the transformer temperature data transformation node, so as to achieve targeted data transformation processing.

[0057] Next, using a task cause-effect graph as the base layer, which contains multiple sets of core element associations, such as line load → transformer temperature, switch status → regional power supply stability, etc., multiple task element data classes from the first data class, such as line load peak value, transformer temperature fluctuation range, etc., are mapped to the nodes of the cause-effect graph according to their corresponding task supervision elements, forming the first data layer. Each node is associated with 3-5 high-dimensional data items. Similarly, multiple low-dimensional data classes from the second data class, such as line load sampling values ​​every 15 minutes, transformer temperature sampling values ​​every 30 minutes, etc., are mapped to the corresponding nodes, forming the second data layer. Each node is associated with 1-2 low-dimensional sampling data items.

[0058] The task causal graph of the base layer, the first data layer, and the second data layer are then cascaded: the causal relationship of the base layer serves as the logical framework for data processing, guiding the data interaction between the first and second data layers; the first data layer establishes a transformation channel with the second data layer through a micro-generator to generate high-dimensional analysis data from low-dimensional sampling data; real-time data synchronization is achieved between layers through preset interfaces, ultimately forming a task supervision architecture that includes causal logic, high-dimensional analysis data, and low-dimensional sampling data.

[0059] By constructing and deploying a micro-generator through first-order adversarial training and second-order lightweight distillation, and combining graph node mapping and inter-layer cascading, a task supervision architecture that can efficiently realize data transformation and has clear logic is constructed, providing a structured data processing and control foundation for the supervision tasks of smart grids.

[0060] In one possible implementation, the micro-generator building module 20 further includes:

[0061] The system includes a plugin deployment subunit for generating and deploying a plugin based on the transformation between the first and second data dimensions, and a judgment logic for further deployment. A data generator determination subunit is used to cascade the generation plugin and the judgment plugin, and determine the data generator through adversarial supervised training. A task supervision unit determination subunit is used to perform transfer training on the data generator, and is distributed across the task supervision architecture to determine the task supervision unit, which is a built-in functional area of ​​the power grid management platform.

[0062] Specifically, after determining the task supervision architecture, the generation plugin and the judgment plugin are deployed first. The generation plugin is configured to generate data for the conversion between the high-dimensional first data dimension and the low-dimensional second data dimension. For example, the second data category is low-dimensional sampling data such as line current and voltage collected every 30 seconds, which is a single-dimensional sequence. The first data category is high-dimensional analysis data containing features such as fluctuation trends, peak duration, and three-phase imbalance, which is a multi-dimensional matrix. One generation plugin is deployed for each of the above five different line conversion scenarios, and each plugin has a preset corresponding data dimension mapping rule. The judgment plugin is deployed based on the generation judgment logic to evaluate the authenticity of the generated data. A deviation threshold of 5% is set, that is, when the difference between the generated high-dimensional data and the real first data category is within 5%, it is judged as valid. Each judgment plugin is associated with one generation plugin, forming a pairing relationship.

[0063] Subsequently, a cascaded generation plugin and a judgment plugin are trained under supervised supervision using an adversarial neural network architecture. The generation plugin acts as a generator, taking a low-dimensional second data class as input, such as current sampling values ​​collected every 3 minutes for 24 consecutive hours on a certain line, totaling 480 data points, to generate a simulated high-dimensional first data class, i.e., current analysis data containing multiple features. The judgment plugin acts as a discriminator, taking the generated data and the real first data class as input, and outputting the probability that it is real data. It is trained using 10,000 sets of historical matching samples. The generator aims to achieve a misclassification rate ≥90% for the discriminator, and the discriminator aims for an accuracy ≥95%. After multiple iterations, when the authenticity of the generated data stabilizes at 90%, training stops. The trained generation plugin is retained, while the judgment plugin serves as an auxiliary inspection component for subsequent sampling and quality checks of the generated data, forming the data generator.

[0064] Finally, the data generator is transferred and trained and distributed on the task supervision architecture. The multi-threaded cascading relationship is determined by the inter-layer cascading of the first data layer and the second data layer. The data generator is then transferred and each micro-generator is constructed by lightweight distillation training according to the thread cascading relationship. Finally, it is distributedly embedded and deployed on the task supervision architecture based on the inter-layer cascading, thereby determining the task supervision unit. The specific steps are explained in detail in the section on determining the micro-unit from the multi-threaded cascading relationship to the distributed embedding and deployment of the micro-unit.

[0065] By deploying generation and judgment plugins and conducting adversarial supervised training, the deterministic data generator can efficiently transform low-dimensional to high-dimensional data, converting single-dimensional sampled data into multi-dimensional analysis data. This provides a core component for the subsequent targeted deployment of distributed micro-generators, meeting the high-speed and targeted processing requirements under complex and interleaved data.

[0066] In one possible implementation, the task monitoring unit further includes the following subunit:

[0067] A multi-threaded cascading relationship determination micro-unit is used to determine the multi-threaded cascading relationship based on the inter-layer cascading of the first data layer and the second data layer, wherein each thread cascading relationship represents a set of inter-layer cascading data; a first micro-generator determination micro-unit is used to migrate the data generator and perform lightweight distillation training based on the first thread cascading relationship to determine the first micro-generator; an Nth micro-generator construction micro-unit is used to traverse the multi-threaded cascading relationship to complete the construction of the Nth micro-generator; and a distributed embedding deployment execution micro-unit is used to perform distributed embedding deployment based on inter-layer cascading on the first micro-generator up to the Nth micro-generator.

[0068] Specifically, when performing transfer training and distributed deployment of the data generator, the multi-threaded cascading relationship is first determined by the inter-layer cascading of the first and second data layers. For example, the first data layer contains 5 sets of high-dimensional analysis data, such as line load peak characteristics and transformer temperature fluctuation trends, while the second data layer corresponds to 5 sets of low-dimensional sampling data, such as line load sampling values ​​every 15 minutes and transformer temperature sampling values ​​every 30 minutes. By matching the correspondence between the inter-layer data, 5 sets of thread cascading relationships are determined. Each relationship represents the transformation link from low-dimensional sampling data to high-dimensional analysis data for a line. For example, thread 1 corresponds to the load data transformation of line A, thread 2 corresponds to the load data transformation of line B, and so on.

[0069] Subsequently, the data generator, which had already undergone adversarial training, was migrated and subjected to lightweight distillation training for the first thread cascade relationship. Using 1000 sets of historical data from thread 1 (low-dimensional sampled data and corresponding high-dimensional analysis data of line A) as the training set, the original data generator with 800 neurons was used as the teacher model. A student model with 320 neurons and a 60% reduction in parameter size was constructed. The core weights related to the data transformation of line A in the teacher model were extracted using the knowledge distillation algorithm, retaining 85% of the key transformation logic. After multiple rounds of iterative training, when the error between the high-dimensional data generated by the student model and the real data stabilized within 2.5%, the response speed was improved compared to the original generator, and it was determined to be the first micro-generator.

[0070] Next, the remaining four thread cascade relationships are traversed, and the second to fifth micro-generators are constructed in the same way. Each thread uses 1000 sets of historical data corresponding to the corresponding line. The parameter size of the trained micro-generators is controlled at 320-350 neurons, the generation data error is ≤3%, and the response speed is kept within 0.1 seconds, thus completing the construction of all micro-generators.

[0071] Finally, these micro-generators are deployed in a distributed embedded manner based on inter-layer cascading: the first micro-generator is embedded in the low-dimensional and high-dimensional data transformation node of line A, the second micro-generator is embedded in the corresponding node of line B, and so on, so that each group of thread cascading relationships has a dedicated micro-generator responsible for data transformation, realizing distributed parallel processing.

[0072] By defining the multi-threaded cascading relationship, performing transfer training and traversing to build micro-generators, and deploying them in a distributed manner, the targeted deployment of multiple lightweight micro-components was achieved. Each micro-generator can specifically handle the low-dimensional to high-dimensional data transformation of the corresponding line, thereby improving the response speed and meeting the high-speed and targeted processing requirements under complex interleaved data.

[0073] In one possible implementation, the target power grid management module 30 further includes:

[0074] A power grid sampling instruction generation unit is used to generate power grid sampling instructions upon the initiation of the power grid monitoring task; a multi-channel sampling instruction determination unit is used to activate the task monitoring unit according to the power grid sampling instructions, and determine multi-channel sampling instructions based on the second data layer, wherein the multi-channel sampling instructions are execution instructions collected by the driver layer data nodes; a distributed write execution unit is used to perform power grid-side sampling and backhaul based on the second data class according to the multi-channel sampling instructions, and perform distributed writing in the second data layer.

[0075] Specifically, when a power grid monitoring task is initiated, taking the regional power grid load fluctuation monitoring task as an example, the system first triggers the power grid sampling instruction generation process. By extracting the core monitoring scope of the task, such as 10 distribution lines and 5 transformers; the sampling period, such as once every 5 minutes; and key monitoring items, such as current, voltage, and equipment temperature, this information is integrated into a standardized power grid sampling instruction. The instruction clearly includes key parameters such as task identifier, sampling object code, and time window to ensure that subsequent data collection actions are consistent with the task objectives.

[0076] Next, based on the generated power grid sampling instructions, the system wakes up the task monitoring unit through a preset activation interface. Once activated, the task monitoring unit calls upon the metadata information of the second data layer. This layer records the sampling point distribution, data types, and communication protocols of each power grid node, and decomposes the power grid sampling instructions into multiple sampling instructions adapted to different physical nodes. For example, for 10 distribution lines, 20 sampling instructions are generated based on two monitoring points for each line, namely the beginning and end. Each instruction corresponds to a sensor at a specific node, specifying the second data type to be collected (such as instantaneous current value and effective voltage value) and the feedback format, ensuring that the instructions can directly drive the underlying data nodes to perform the acquisition operation.

[0077] Then, each power grid-side data node, such as line sensors and transformer monitoring modules, receives the corresponding multi-channel sampling instructions and initiates data acquisition according to the instructions: current sensors acquire instantaneous values, voltage sensors record valid values, and temperature sensors acquire equipment surface temperatures. All acquired data is encapsulated in the low-dimensional format of the second data class, i.e., single-precision floating-point type, fixed-length byte stream. After acquisition, the data is transmitted back to the system via a dedicated power communication network. During the transmission, a timestamp synchronization mechanism is used to ensure that the data of different nodes are aligned in the time dimension. After receiving the transmitted data, the system, based on the node mapping relationship of the second data layer (i.e., each physical node corresponds to one logical node within the layer), distributes the data and writes it into the storage area of ​​the corresponding logical node, achieving precise matching between the data and the structure within the layer.

[0078] By generating sampling instructions through task-driven processes, decomposing multiple instructions based on the second data layer, and linking nodes to execute sampling feedback and distributed writing, accurate and efficient collection and storage of the second data class at the power grid end is achieved, providing reliable front-end data support for subsequent data processing and regulatory decisions.

[0079] In one possible implementation, the target power grid management module 30 further includes:

[0080] The first data layer update unit is used to perform data reconstruction generation from the second data layer to the first data layer according to the distributed micro-generator embedded between the layers, and update the first data layer; the task supervision result determination unit is used to perform power grid supervision decision based on the mapping between the first data layer and the base layer and the updated data in the first data layer and the task causal graph, and to standardize and determine the task supervision result; the task supervision result display unit is used to display the task supervision result in a pop-up window on the display port of the power grid management platform.

[0081] Specifically, after the data acquisition and control steps at the power grid end, the distributed micro-generators embedded between layers begin to operate. These micro-generators correspond to different power grid data conversion needs, such as micro-generators responsible for line load data conversion and micro-generators focused on transformer temperature data conversion. When the second data layer receives and stores new low-dimensional sampled data (such as line current values ​​and transformer temperature values ​​updated every 5 minutes), each micro-generator reconstructs the second data class of its respective domain in real time according to the preset conversion logic: converting single-dimensional current sampled values ​​into high-dimensional feature data containing peak value, fluctuation frequency, and duration, and converting temperature sampled values ​​into high-dimensional analysis data containing temperature rise rate and correlation with load. The generated high-dimensional data is written to the corresponding node of the first data layer, completing the update of the first data layer.

[0082] After the first data layer is updated, the system invokes the mapping relationship between the first data layer and the base layer, which is the task causal graph. Each data node in the first data layer corresponds to a specific causal node in the base layer. For example, the peak load node for line A in the first data layer corresponds to the causal relationship of line A load → transformer B temperature in the base layer. Based on this, the system matches the updated data from the first data layer with the regulatory rules in the task causal graph, such as a line load peak exceeding a threshold for 10 minutes triggering a transformer cooling warning. It then performs multi-dimensional decision-making: determining whether each element's data is within the normal range, whether the correlation between elements conforms to causal logic, and whether there are potential risks. By integrating these judgment results, a task regulatory result containing abnormal elements, risk levels, and associated impacts is generated.

[0083] Finally, once the power grid management platform's display port receives the compiled task monitoring results, it immediately triggers a pop-up display mechanism. The pop-up displays the core information of the monitoring results: for example, the specific name of the abnormal element, such as abnormal load on line C; key parameters, comparing current values ​​with thresholds; related causal relationships, such as correlation with abnormal temperature on transformer D; and suggested actions, such as reducing the load on line C. The pop-up uses a hierarchical display strategy, as shown in Table 1, to ensure that managers can quickly identify priorities.

[0084] Table 1: Pop-up Hierarchical Display Strategy Table

[0085]

[0086] By using a distributed micro-generator to achieve real-time conversion and updating of low-dimensional to high-dimensional data, combining task causal graphs for correlation decision-making, and then presenting the results in a timely manner through pop-up windows, precise and efficient control of target power grid monitoring tasks is achieved, providing real-time assurance for the stability of power grid operation.

[0087] In summary, the data acquisition and control system for smart grids provided in this application has the following technical effects:

[0088] This application acquires high-dimensional analysis data of the first data class by collecting low-dimensional sampling data of the second data class at the power grid end, and then processing it through first-order adversarial training and second-order lightweight distillation training of a distributed micro-generator. The application calculates the state of power grid elements and the monitoring results, and makes adjustments based on the base layer mapping and calculation results of the task causal graph, thereby accurately controlling the operating state of the target power grid and making the execution results of the smart grid monitoring task more accurate and reliable.

[0089] Example 2, as Figure 2 As shown, based on the same inventive concept as in Embodiment 1 above, this application provides a data acquisition and control method for a smart grid, the method comprising:

[0090] Step A100: Upload the power grid monitoring task to the power grid management platform. By performing task interpretation and data sampling dimension transformation, determine the first data class and the second data class. The task interpretation is accompanied by the output of a task cause-effect graph.

[0091] Step A200: Based on the first data class, the second data class, and the task causal graph, construct a task supervision unit, wherein micro generators are distributed between the first data class and the second data class, constructed by first-order adversarial training based on data class transformation and second-order lightweight distillation training based on directional transformation generation relationship.

[0092] Step A300: Upon task startup, the task monitoring unit embedded in the power grid management platform is activated. It uses the second data class to collect and control data on the power grid side, and uses the first data class based on the micro-generator to generate and assess the state of power grid elements based on the task causal graph, thereby controlling the target power grid based on the power grid monitoring task.

[0093] Furthermore, after performing task interpretation, step A100 of this application embodiment also includes:

[0094] Connecting to the power grid database, data mining is performed using power grid regulatory elements and element data classes to construct a power grid causal graph, where the power grid regulatory elements are the smallest element units. By separating spurious correlations, the power grid causal graph is pruned in its first stage to determine a first-order causal graph. Spurious correlation coefficients are determined by balancing data relationships and data values. Using task interpretation information, the first-order causal graph is pruned in its second stage to determine a task causal graph.

[0095] Furthermore, the data sampling dimension is transformed to determine the first data class and the second data class. Step A100 in this embodiment of the application also includes:

[0096] Based on the task cause-effect graph, identify task element data classes and task supervision elements; map the task element data classes and task supervision elements to determine multiple mapping groups; traverse the multiple mapping groups, perform a first data dimension transformation to determine a first data class, and perform a second data transformation to determine a second data class.

[0097] Furthermore, performing a first data dimension transformation to determine a first data class and performing a second data transformation to determine a second data class, step A100 in this embodiment of the application further includes:

[0098] For each mapping group, based on the data value orientation of the task supervision elements within the group, the task element data within the group is transformed into a first data dimension to determine a first data class, wherein the first data dimension is transformed into either an up-dimensionality or a down-dimensionality. For each mapping group, based on the lightweight collection orientation, the task element data within the group is transformed into a second data dimension to determine a second data class, wherein the second data dimension is transformed into a minimum-order data down-dimensionality process to determine a second data class.

[0099] Furthermore, based on the first data class, the second data class, and the task cause-effect graph, a task supervision unit is constructed. Step A200 in this embodiment further includes:

[0100] Using the task cause-effect graph as the base layer, graph node mapping of the first data class is performed to determine the first data layer, graph node mapping of the second data class is performed to determine the second data layer; the base layer, the first data layer and the second data layer are cascaded to determine the task supervision architecture.

[0101] Furthermore, after determining the task supervision architecture, step A200 of this application embodiment also includes:

[0102] A generation plugin is deployed based on the transformation between the first and second data dimensions. A judgment plugin is deployed to generate the judgment logic. The generation plugin and the judgment plugin are cascaded, and an adversarial supervised training is conducted to determine the data generator. The data generator is then subjected to transfer training and distributed deployment in the task supervision architecture to determine the task supervision unit, wherein the task supervision unit is a built-in functional area of ​​the power grid management platform.

[0103] Furthermore, the data generator undergoes transfer training and is deployed in a distributed manner on the task supervision architecture. Step A200 in this embodiment further includes:

[0104] The multi-threaded cascading relationship is determined by the inter-layer cascading of the first data layer and the second data layer, wherein each thread cascading relationship represents a set of inter-layer cascading data; the data generator is migrated and lightweight distillation training is performed with the first thread cascading relationship to determine the first micro-generator; the multi-threaded cascading relationship is traversed to complete the construction of the Nth micro-generator; the first micro-generator to the Nth micro-generator are deployed in a distributed embedding based on inter-layer cascading.

[0105] Furthermore, in the implementation of the second data category for data acquisition and control at the power grid end, step A300 of this application embodiment also includes:

[0106] Upon initiation of the power grid monitoring task, a power grid sampling instruction is generated. Based on the power grid sampling instruction, the task monitoring unit is activated, and based on the second data layer, multiple sampling instructions are determined, wherein the multiple sampling instructions are execution instructions collected by the data nodes of the driving layer. Based on the multiple sampling instructions, power grid-side sampling and feedback based on the second data class are performed, and distributed writing is performed in the second data layer.

[0107] Furthermore, based on the generation of the first data class based on the micro-generator and the assessment of the power grid element status based on the task causal graph, the target power grid is managed and controlled based on the power grid regulatory task. In this embodiment, step A300 further includes:

[0108] Based on the distributed micro-generator embedded between layers, data reconstruction and generation from the second data layer to the first data layer are performed, and the first data layer is updated; based on the mapping between the first data layer and the base layer, power grid supervision decisions are made based on the updated data and the task causal graph within the first data layer, and the task supervision results are standardized and determined; the task supervision results are displayed in a pop-up window on the display port of the power grid management platform.

[0109] Through the foregoing detailed description of the data acquisition and control system for smart grids, those skilled in the art can clearly understand the data acquisition and control method for smart grids in this embodiment. As for the method disclosed in Embodiment 2, since it corresponds to the system disclosed in Embodiment 1, it has corresponding execution steps and technical effects. For relevant details, please refer to the system section description.

[0110] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A data acquisition and control system for smart grids, characterized in that, The system includes: The data class determination module is used to upload power grid supervision tasks to the power grid management platform. By performing task interpretation and data sampling dimension transformation, it determines the first data class and the second data class. The task interpretation is accompanied by the output of a task cause-effect graph. The micro-generator construction module is used to construct a task supervision unit based on the first data class, the second data class and the task causal graph. The first data class and the second data class are distributed with micro-generators constructed by first-order adversarial training based on data class transformation and second-order lightweight distillation training based on directional transformation generation relationship. The target power grid control module is used to activate the task supervision unit embedded in the power grid management platform when the task starts. It uses the second data class to collect and control data on the power grid side, and uses the first data class based on the micro generator to generate and assess the state of power grid elements based on the task causal graph to control the target power grid based on the power grid supervision task. After task interpretation, it includes: A power grid cause-effect graph construction unit is used to connect to the power grid database, perform data mining based on power grid regulatory elements and element data classes, and construct a power grid cause-effect graph, wherein the power grid regulatory elements are the smallest element units; The false correlation coefficient determination unit is used to perform first-level pruning on the power grid causal graph by separating false correlations, determine the first-order causal graph, and determine the false correlation coefficient by balancing data relationships and data value. The task cause-effect graph determination unit is used to perform secondary pruning on the first-order cause-effect graph using task interpretation information to determine the task cause-effect graph. Perform dimensional transformation on the data sampling to determine the first and second data classes, including: The task supervision element identification unit is used to identify task element data classes and task supervision elements based on the task cause-effect graph. A mapping group determination unit is used to map the task element data class and the task supervision element, and determine multiple mapping groups; The first and second data class determination unit is used to traverse the plurality of mapping groups, perform a first data dimension transformation to determine a first data class, and perform a second data transformation to determine a second data class. Perform a first data dimension transformation to determine a first data class, and perform a second data transformation to determine a second data class, including: The first data class determination subunit is used to perform a first data dimension transformation on the task element data within each mapping group based on the data value orientation of the task supervision elements within the group, and determine the first data class, wherein the first data dimension is transformed into either an increased dimension or a decreased dimension. The second data class determination subunit is used to perform a second data dimension transformation on the task element data within each mapping group in a lightweight acquisition orientation to determine the second data class. The second data dimension is converted into the minimum order data dimensionality reduction process to determine the second data class.

2. The data acquisition and control system for smart grids as described in claim 1, characterized in that, Based on the first data class, the second data class, and the task cause-effect graph, a task supervision unit is constructed, including: The second data layer determination unit is used to determine the first data layer by performing graph node mapping of the first data class and graph node mapping of the second data class, based on the task cause-effect graph as the base layer. The task supervision architecture determination unit is used to cascade the base layer, the first data layer and the second data layer to determine the task supervision architecture.

3. The data acquisition and control system for smart grids as described in claim 2, characterized in that, After determining the task oversight architecture, this includes: The decision plugin deployment subunit is used to generate and deploy the generation plugin by converting the first data dimension and the second data dimension, in order to generate the decision logic and deploy the decision plugin. The data generator determines the subunit, which is used to cascade the generation plugin and the judgment plugin, and determines the data generator through adversarial supervised training; The task supervision unit determines sub-units for performing migration training on the data generator. These sub-units are distributed and deployed within the task supervision architecture. The task supervision unit is a built-in functional area of ​​the power grid management platform.

4. The data acquisition and control system for smart grids as described in claim 3, characterized in that, The data generator is subjected to transfer training and distributed deployment within the task supervision architecture, including: The multi-threaded cascading relationship determination micro-unit is used to determine the multi-threaded cascading relationship by cascading the first data layer and the second data layer, wherein each thread cascading relationship represents a set of inter-layer cascading data; The first micro-generator determines the micro-unit, which is used to migrate the data generator, and performs lightweight distillation training with the first thread cascade relationship to determine the first micro-generator; The Nth micro-generator constructs micro-units to traverse the multi-threaded cascade relationship and complete the construction of the Nth micro-generator; A distributed embedded deployment execution micro-unit is used to perform distributed embedded deployment based on inter-layer cascading on the first micro-generator up to the Nth micro-generator.

5. The data acquisition and control system for smart grids as described in claim 4, characterized in that, Data collection and management at the power grid end are carried out using the second data category, including: A power grid sampling instruction generation unit is used to generate power grid sampling instructions upon the initiation of the power grid monitoring task; A multi-channel sampling instruction determination unit is used to activate the task monitoring unit according to the power grid sampling instruction, and determine the multi-channel sampling instruction based on the second data layer, wherein the multi-channel sampling instruction is an execution instruction collected by the data node of the driving layer; The distributed write execution unit is used to perform grid-side sampling and backhaul based on the second data class according to the multi-channel sampling instructions, and to perform distributed writing in the second data layer.

6. The data acquisition and control system for smart grids as described in claim 5, characterized in that, Based on the generation of the first data class using a micro-generator and the assessment of power grid element status using a task-based causal graph, the target power grid is managed and controlled according to power grid regulatory tasks, including: The first data layer update unit is used to perform data reconstruction and generation from the second data layer to the first data layer according to the distributed micro-generator embedded between layers, and update the first data layer. The task supervision result determination unit, based on the mapping between the first data layer and the base layer, executes power grid supervision decisions based on the updated data and task causal graph within the first data layer, and standardizes and determines the task supervision result; The task monitoring result display unit is used to display the task monitoring results in a pop-up window on the display port of the power grid management platform.

7. A data acquisition and control method for smart grids, characterized in that, The method is implemented by the data acquisition and control system for smart grids according to any one of claims 1-6, and the method includes: Upload the power grid monitoring task to the power grid management platform. By performing task interpretation and data sampling dimension transformation, determine the first data class and the second data class. The task interpretation is accompanied by the output of a task cause-effect graph. Based on the first data class, the second data class, and the task causal graph, a task supervision unit is constructed. Among them, micro generators are distributed between the first data class and the second data class, which are constructed by first-order adversarial training based on data class transformation and second-order lightweight distillation training based on directional transformation generation relationship. Upon task initiation, the task monitoring unit embedded in the power grid management platform is activated. It uses the second data type to collect and control data from the power grid, and uses the first data type based on the micro-generator to generate and assess the state of power grid elements based on the task causal graph, thereby controlling the target power grid based on the power grid monitoring task.

Images