Electric power Internet of Things sensing equipment configuration method and system based on edge computing

Abstract

The invention relates to a power Internet of Things sensing equipment configuration method and system based on edge computing. The method comprises the steps of performing formalized description processing on an input business intention, generating a business intention description file, converting the business intention description file into quantifiable perception task requirements through semantic analysis, screening and optimizing sensor combinations from a resource library through dynamic matching and a multi-objective optimization technology, generating specific configuration parameters, and issuing and executing the specific configuration parameters. By adopting the method, a traditional sensor configuration process depending on artificial experience can be improved into a self-adaptive closed-loop system based on semantic comprehension and an optimization algorithm, the time and labor cost for converting service intention to equipment configuration are remarkably reduced, and the resource utilization efficiency is maximized while complex service requirements are met through dynamic optimization; and the response speed and the intelligent level of the electric power Internet of Things system under the scenes of fault diagnosis, load prediction and the like are improved.

Description

Technical Field

[0001] This invention belongs to the field of sensor device configuration technology, and in particular relates to a method and system for configuring power Internet of Things (IoT) sensor devices based on edge computing. Background Technology

[0002] With the deep integration of the power Internet of Things (IoT) and edge computing technologies, the demand for refined and intelligent management of massive amounts of sensing devices in smart grids is becoming increasingly prominent. In complex business scenarios such as substation condition monitoring, distribution network fault location, and transmission line inspection, how to efficiently coordinate and manage widely distributed sensor nodes to enable them to flexibly respond to high-level business needs has become a key focus of the industry. Currently, the configuration and management of sensing devices in power systems mainly rely on centralized processing based on preset rules or expert experience.

[0003] In traditional technologies, maintenance personnel typically manually configure sensor operating parameters such as sampling frequency, range, and trigger conditions based on business needs, or rely on pre-set fixed configuration templates to batch configure a certain type of sensor. For example, during routine inspection tasks, all relevant sensors collect data at a lower frequency according to a uniform template; when fault diagnosis is required, sensors must be manually re-selected and their configurations adjusted one by one to obtain higher-precision data. Throughout this process, there is a lack of an automated conversion mechanism between business needs and specific device parameters. Edge computing nodes mainly undertake data forwarding and simple filtering functions, and their potential in near-field resource scheduling and real-time decision-making has not been fully utilized.

[0004] However, the current approach has significant limitations: manual configuration is inefficient and difficult to adapt to scenarios with frequent or dynamic changes in business intent, such as switching from routine monitoring to emergency fault handling; static configuration templates lack flexibility and cannot automatically optimize sensor combinations and parameters based on real-time business needs, equipment status, and network conditions, which can easily lead to low resource utilization or substandard perception quality; more importantly, there is a huge gap between high-level business semantics and low-level sensor configuration parameters, making it difficult to translate business intents into actionable actions that can be performed by the equipment in a timely and accurate manner, thus hindering the improvement of the overall intelligence level of the power Internet of Things system. Summary of the Invention

[0005] Therefore, it is necessary to provide a method that can automatically and dynamically transform business requirements into sensor configurations and support closed-loop optimization to address the aforementioned technical issues.

[0006] In a first aspect, this application provides a method for configuring power Internet of Things (IoT) sensing devices based on edge computing, including:

[0007] The input high-level power business intent is formally described to generate a business intent description file; the business intent description file includes business objectives, quality constraints, and execution actions.

[0008] The business intent description file is semantically parsed to generate perception task requirements; the perception task requirements include the required data types, quality requirements, and spatiotemporal constraints.

[0009] Based on the requirements of the sensing task, the sensor resources managed by the edge computing platform are matched to obtain a set of candidate sensors;

[0010] Based on a multi-objective optimization algorithm, the candidate sensor set is optimized and selected to generate the optimal sensor combination.

[0011] Based on the specific indicators required for the sensing task, the configuration parameters of each sensor in the optimal sensor combination are calculated; and the configuration parameters are combined to obtain a detailed set of configuration parameters.

[0012] Furthermore, the method also includes:

[0013] After executing the detailed configuration parameter set, the raw data of each sensor in the optimal sensor combination is collected; and the raw data is processed in a structured manner at the edge node to obtain a structured dataset.

[0014] Based on a structured dataset, the degree to which the current business intent is satisfied is evaluated, and intent satisfaction evaluation results are generated.

[0015] Based on the intent satisfaction assessment results, determine whether to trigger the dynamic adjustment mechanism and obtain the judgment result;

[0016] If the judgment result is to trigger the dynamic adjustment mechanism, an adjustment instruction is generated based on the intent satisfaction assessment result. The adjustment instruction is used to instruct that the steps from generating the perception task requirements to generating the detailed configuration parameter set be re-executed according to the intent satisfaction assessment result, so as to obtain the updated detailed configuration parameter set.

[0017] Furthermore, the input high-level power business intent is formally described to generate a business intent description file, including:

[0018] Receive high-level power business intents input in natural language or structured forms; and extract key entities and relationships between key entities from the high-level power business intents.

[0019] Based on a pre-defined power business ontology library, the relationships between key entities are mapped to obtain standardized mapping results. The standardized mapping results include domain concepts, business objectives, quality constraints, and action instructions.

[0020] Based on the standardized mapping results, an intent description model is constructed; the intent description model includes a domain concept layer, a business goal layer, a quality constraint layer, and an action instruction layer.

[0021] The hierarchical intent description model is serialized into a standard JSON-LD format file to obtain the business intent description file.

[0022] Furthermore, semantic parsing is performed on the business intent description file to generate perception task requirements, including:

[0023] Perform syntactic and semantic analysis on the business intent description file to obtain core elements; the core elements include business scenario, target device, accuracy indicators and latency requirements.

[0024] Query the knowledge graph of the power industry to obtain domain rules and historical cases related to core elements;

[0025] Based on domain rules and historical cases, the abstract goals in the high-level power business intent are transformed into data collection requirements through a rule reasoning engine;

[0026] All data collection requirements are combined to generate a perception task requirement file; the perception task requirement file includes a list of data types, data quality indicators, collection time windows, and spatial ranges.

[0027] Furthermore, based on a multi-objective optimization algorithm, the candidate sensor set is optimized and selected to generate the optimal sensor combination, including:

[0028] Construct a multi-objective optimization function; the multi-objective optimization aims to minimize sensor resource overhead and maximize the satisfaction of business requirements.

[0029]

[0030]

[0031] in, For a multi-objective optimization function, For data quality functions; This is the total time delay function; This is the total energy consumption function; Here, X represents the constraint, m represents the total number of constraints, and j represents the constraint identifier.

[0032] Based on the candidate sensor set, initialize the population; each individual in the population represents a possible sensor combination scheme.

[0033] Based on a multi-objective optimization function, an improved genetic algorithm is used to iteratively optimize the population until the iteration process meets the convergence condition, and the optimal solution on the Pareto front is extracted from the final generation of the population.

[0034] The optimal solution is decoded to obtain the optimal sensor combination.

[0035] Furthermore, based on the structured dataset, the degree to which the current business intent is satisfied is evaluated, generating intent satisfaction evaluation results, including:

[0036] Assign weight coefficients to the evaluation metrics for the current business intent; the evaluation metrics include data accuracy, data integrity, timeliness, and specific metrics based on business results; the weight coefficients are used to reflect the importance of the metrics to the business intent;

[0037] Calculate the actual values ​​of each evaluation indicator based on the structured dataset;

[0038] Based on the actual values ​​and weighting coefficients of each evaluation indicator, the intention satisfaction score is calculated.

[0039] Based on the intent satisfaction score and a preset threshold, an intent satisfaction assessment result is generated.

[0040] Secondly, this application also provides a power Internet of Things (IoT) sensor device configuration system based on edge computing, including:

[0041] The document generation module is used to perform formal description processing on the input high-level power business intent and generate a business intent description file; the business intent description file includes business objectives, quality constraints and execution actions;

[0042] The requirement generation module is used to perform semantic parsing on the business intent description file to generate perception task requirements; perception task requirements include the required data types, quality requirements, and time and space constraints.

[0043] The candidate sensor set generation module is used to match the sensor resources managed by the edge computing platform according to the requirements of the sensing task to obtain a candidate sensor set.

[0044] The sensor combination generation module is used to optimize and select the candidate sensor set based on a multi-objective optimization algorithm to generate the optimal sensor combination.

[0045] The configuration parameter set generation module is used to calculate the configuration parameters of each sensor in the optimal sensor combination based on the specific indicators required by the sensing task; and to combine the configuration parameters to obtain a detailed configuration parameter set.

[0046] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores at least one instruction, at least one program, a code set, or an instruction set, and the at least one instruction, the at least one program, the code set, or the instruction set is loaded and executed by the processor to implement any of the edge computing-based power Internet of Things sensing device configuration methods described in the embodiments of this application.

[0047] Fourthly, this application also provides a computer-readable storage medium storing at least one piece of program code, which is loaded and executed by a processor to implement the edge computing-based power Internet of Things sensing device configuration method described in any of the embodiments of this application.

[0048] The aforementioned edge computing-based power IoT sensor configuration method and system formally describes the input business intent, generates a business intent description file, and transforms it into quantifiable sensing task requirements through semantic parsing. It then uses dynamic matching and multi-objective optimization techniques to select and optimize sensor combinations from a resource library, generating specific configuration parameters and issuing them for execution. This transforms the traditional sensor configuration process, which relies on manual experience, into an adaptive closed-loop system based on semantic understanding and optimization algorithms. This significantly reduces the conversion time and labor costs from business intent to device configuration. Furthermore, dynamic optimization ensures that resource utilization efficiency is maximized while meeting complex business needs, improving the response speed and intelligence level of the power IoT system in scenarios such as fault diagnosis and load forecasting. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating a power Internet of Things (IoT) sensor device configuration method based on edge computing in one embodiment.

[0051] Figure 2 This is a flowchart illustrating the steps of formally describing the input high-level power service intent and generating a service intent description file in one embodiment.

[0052] Figure 3 This is a schematic diagram of the configuration system for a power Internet of Things (IoT) sensing device based on edge computing in one embodiment. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In one embodiment, a method for configuring power Internet of Things (IoT) sensing devices based on edge computing is provided. This embodiment illustrates the application of this method to a terminal, but it is understood that the method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. Figure 1 As shown, in this embodiment, the method includes the following steps:

[0055] Step S101: Perform formal description processing on the input high-level power business intent to generate a business intent description file; the business intent description file includes business objectives, quality constraints, and execution actions.

[0056] Formal description processing refers to transforming vague, natural language information such as concepts, rules, and problems into precise, unambiguous, and symbolic expressions that conform to specific logical or mathematical norms.

[0057] For example, high-level power business intentions are typically input in the form of natural language or structured forms, and can be used to generate business intention description files through formal description processing.

[0058] Step S102: Perform semantic parsing on the business intent description file to generate perception task requirements; perception task requirements include the required data types, quality requirements, and spatiotemporal constraints.

[0059] Semantic parsing processing refers to transforming human-understandable business objective descriptions into structured semantic representations of structured information that machines can understand and execute, enabling computers not only to "understand text" but also to "understand the meaning, logical relationships, and intentions behind the text."

[0060] For example, although the business intent description file is already structured, it needs to be further transformed into executable sensor-level instructions. Semantic parsing is used to generate perception task requirements that include information such as the required data type, quality requirements, and spatiotemporal constraints.

[0061] Step S103: Based on the requirements of the sensing task, the sensor resources managed by the edge computing platform are matched to obtain a set of candidate sensors.

[0062] Among them, edge computing platform management is the core link for unified coordination, monitoring and optimization of hardware resources, software services, data flow and application deployment in the edge computing architecture. It can solve the management complexity problem caused by the "scattered distribution, large scale and complex scenarios" of edge nodes, and ensure that the edge computing capabilities can efficiently and stably support business needs such as industrial control and smart cities. The edge computing platform maintains a dynamic sensor resource library. Each sensor in the library is described by a physical model, which describes its static capabilities such as measurement type, range and accuracy range, as well as its dynamic status such as battery power and network latency. The dynamic sensor resource library is a collection system built around sensor resources and has dynamic management and adaptation capabilities.

[0063] For example, based on the data type (e.g., "current") in the sensing task requirements, sensors with basic matching are filtered from a dynamic sensor resource library. For instance, all sensors supporting current measurement are selected. Then, the capability matching degree of each sensor is calculated; for example, if the task requires a sampling frequency of 1kHz, only sensors supporting a frequency range including 1kHz are selected. Simultaneously, the real-time status of the sensors is evaluated, such as excluding devices with battery levels below 20% or communication latency exceeding 100ms, to ensure stable task execution. A weighted scoring method can be used to comprehensively consider capability matching degree, real-time status health, and geographical proximity, outputting a candidate sensor set. Sensors in this set all meet the basic task requirements and are sorted by matching degree. The weighted scoring method refers to a quantitative analysis tool that assigns weights to different indicators—coefficients reflecting the importance of the indicators—and combines the actual scores of each indicator to calculate a comprehensive result.

[0064] Step S104: Based on the multi-objective optimization algorithm, the candidate sensor set is optimized and selected to generate the optimal sensor combination.

[0065] Among them, optimization selection processing refers to eliminating non-optimal options from multiple options and locking in the solution that best meets the objective through systematic analysis and decision-making; multi-objective optimization algorithm is a class of algorithms that solve optimization problems with multiple conflicting objectives. In scenarios where there is no absolutely optimal solution, it finds a set of feasible optimal solutions with trade-offs, rather than pursuing a unique optimal value like single-objective optimization.

[0066] For example, selecting the optimal sensor combination from the candidate set requires balancing multiple conflicting objectives such as data quality, latency, and energy consumption. An optimal sensor combination can be generated by optimizing the candidate sensor set using a multi-objective optimization algorithm.

[0067] Step S105: Calculate the configuration parameters of each sensor in the optimal sensor combination based on the specific indicators required by the sensing task; and combine the configuration parameters to obtain a detailed set of configuration parameters.

[0068] For example, the optimal sensor combination needs further parameter configuration to adapt to task requirements. Based on the quality indicators in the sensing task requirements, such as accuracy of 0.01% and response time of 5 seconds, the specific operating parameters of each sensor are calculated through a parameter mapping function. For example, the sampling frequency needs to satisfy the Nyquist theorem and is set to at least 2.5 times the highest signal frequency of the task; the measurement range is dynamically adjusted based on the fluctuation range of historical data to avoid saturation or insufficient resolution; the operating mode is selected as continuous sampling or event triggering based on latency requirements. After parameter calculation, a configuration parameter set containing fields such as sensor ID, sampling frequency, measurement range, and accuracy mode is generated and encrypted and sent to each sensor via the MQTT protocol. After receiving the parameters, the sensor returns an acknowledgment signal, and the edge computing platform records the configuration version. Among them, the parameter mapping function is a tool used to convert, associate, or adapt parameters to different contexts, and is used to solve problems such as parameter format mismatch, inconsistent meaning, and incompatible range; the Nyquist theorem is used to solve the key problem of how to sample from continuous analog signals to recover the original signal without distortion, and was developed by American engineer Harry... Nyquist proposed that two core conditions must be met: first, the sampling frequency must be greater than twice the highest frequency of the original analog signal; second, the original analog signal must be a band-limited signal, that is, the frequency components of the signal have a clear upper limit and there are no frequency components higher than the highest frequency of the original signal. MQTT (Message Queuing Telemetry Transport) is a lightweight, low-bandwidth IoT-specific communication protocol used to solve the problem of efficient data transmission in scenarios with limited device resources and unstable network environments.

[0069] In this embodiment, the input business intent is formally described to generate a business intent description file, which is then semantically parsed and transformed into quantifiable perception task requirements. Dynamic matching and multi-objective optimization techniques are used to select and optimize sensor combinations from a resource library, generating specific configuration parameters and issuing them for execution. This elevates the traditional sensor configuration process, which relies on manual experience, to an adaptive closed-loop system based on semantic understanding and optimization algorithms. This significantly reduces the conversion time and labor costs from business intent to device configuration. Furthermore, dynamic optimization ensures that resource utilization efficiency is maximized while meeting complex business needs, improving the response speed and intelligence level of the power IoT system in scenarios such as fault diagnosis and load forecasting.

[0070] In one exemplary embodiment, the method further includes:

[0071] Step S201: After executing the detailed configuration parameter set, collect the raw data of each sensor in the optimal sensor combination; and perform structured processing on the raw data at the edge node to obtain a structured dataset.

[0072] For example, the sensor begins collecting raw data based on the configured parameters such as sampling frequency, range, and accuracy. This data may include current waveforms, voltage amplitudes, temperature readings, etc., output in analog or digital signal form. Since raw data is typically heterogeneous, high-noise, and low-value-density, at edge nodes, a Z-score-based anomaly detection algorithm can be used for data cleaning to identify and remove outliers. Simultaneously, linear interpolation methods can be used to fill in missing data to ensure data quality. A time synchronization protocol can be used to calibrate the timestamps of all cleaned sensor data to ensure data consistency along the timeline. Principal component analysis can be used to reduce the dimensionality of the data and extract key features to reduce the data volume. Finally, a structured dataset is generated, stored in a standardized format such as JSON (JavaScript Object Notation), containing fields such as timestamps, sensor IDs, and feature values. Among these, time synchronization protocols ensure that the local clocks of multiple network devices (such as computers and sensors) remain consistent, resolving time discrepancies caused by differences in the precision of device hardware clocks; Principal component analysis transforms high-dimensional data into low-dimensional data while preserving as much key information (such as variance) as possible, simplifying analytical complexity; Z-score-based anomaly detection algorithms are classic anomaly detection methods based on the assumption of statistical normality. They determine whether a data point is an outlier by calculating the degree of deviation of a data point from the overall distribution of the dataset; the formula for calculating the Z-score (standardized score) is:

[0073]

[0074] Z represents the standardized score, and X represents the single data point to be tested. σ is the mean of the entire dataset, reflecting the central location of the data; σ is the standard deviation of the entire dataset, reflecting the dispersion of the data.

[0075] Step S202: Based on the structured dataset, evaluate the degree of satisfaction of the current business intent and generate intent satisfaction evaluation results.

[0076] For example, multidimensional evaluation metrics are defined, including data accuracy, data integrity, timeliness, and specific metrics based on business performance (such as fault diagnosis accuracy or power flow prediction error). Weighting coefficients are assigned to each metric, and the actual values ​​of each metric are obtained from a structured dataset. Based on the assigned weighting coefficients and the actual values, an intent satisfaction evaluation result is generated, output in numerical or grading form, such as satisfied or partially satisfied.

[0077] Step S203: Based on the intention satisfaction assessment results, determine whether to trigger the dynamic adjustment mechanism and obtain the judgment result.

[0078] For example, based on the intent satisfaction assessment result and preset trigger conditions, it is determined whether to trigger the dynamic adjustment mechanism. Simultaneously, a rule engine is used to identify the main factors causing a decrease in satisfaction, such as sensor performance degradation, network congestion, or changes in business intent, to obtain the judgment result. This judgment result is output as a binary flag or enumerated value, indicating whether it is triggered or not, and may include de-jitter processing to prevent erroneous adjustments caused by transient noise. Furthermore, the judgment mechanism considers business priorities; for example, trigger sensitivity is higher in critical fault diagnosis scenarios to ensure timely response. The preset trigger conditions can be static thresholds, such as triggering when the intent satisfaction score is below 0.8; or they can be dynamically adaptive thresholds calculated using an exponentially weighted moving average method to adapt to environmental fluctuations. De-jitter processing means that the dynamic adjustment mechanism is only confirmed to be triggered after multiple consecutive assessments below the threshold.

[0079] Step S204: If the judgment result is to trigger the dynamic adjustment mechanism, an adjustment instruction is generated based on the intent satisfaction evaluation result. The adjustment instruction is used to instruct the steps from generating the perception task requirements to generating the detailed configuration parameter set to be re-executed according to the intent satisfaction evaluation result, so as to obtain the updated detailed configuration parameter set.

[0080] For example, when the judgment result indicates that the dynamic adjustment mechanism is triggered, an adjustment instruction is generated based on the specific indicator deviation in the intent satisfaction evaluation result. For instance, if the data accuracy is insufficient, the instruction may indicate increasing the sampling frequency or switching to a higher-precision sensor; if the timeliness is not up to standard, the instruction may optimize sensor selection to reduce transmission path latency. This adjustment instruction can adopt a structured format (such as JSON), including the adjustment type, adjustment range, target sensor list, and priority information. After the instruction is generated, it is fed back to the intent parsing stage (i.e., step S102) to restart the workflow: the business intent is re-parsed, generating new perception task requirements, which in turn triggers sensor re-matching, combination optimization, and parameter calculation to obtain an updated detailed configuration parameter set.

[0081] In this embodiment, a high-quality structured dataset is obtained by collecting and processing raw sensor data. Based on this dataset, the degree of satisfaction with business intentions is quantitatively assessed. An assessment result is generated according to the satisfaction level, and adjustment needs are intelligently determined. When the adjustment mechanism is triggered, a precise adjustment command is generated and the configuration process is re-executed. This allows for real-time perception of changes in business needs and environmental fluctuations, dynamic optimization of sensor configuration, and a significant improvement in the adaptability and robustness of the power Internet of Things system.

[0082] In one embodiment, such as Figure 2As shown, the input high-level power service intent is formally described to generate a service intent description file, including:

[0083] Step S301: Receive high-level power business intent input in natural language or structured form; and extract the relationships between key entities from the high-level power business intent.

[0084] For example, the system receives user-inputted business intent, such as a natural language description like "Fault diagnosis of XX transmission line, location must be completed within 5 seconds with an accuracy of no less than 50 meters," or fields such as business type, target equipment, and accuracy requirements filled in a structured form. Natural language can be processed using named entity recognition technology based on a pre-trained language model. Key entities in the text, such as "transmission line," "fault diagnosis," "5 seconds," and "50 meters," are identified through word embedding and context encoding. Dependency parsing is then used to extract semantic relationships between entities, such as "accuracy no less than 50 meters" modifying "fault diagnosis." Structured form input can directly parse key-value pairs of predefined fields, mapping them to standardized entities and relationships, ultimately outputting structured entity-relation pairs. Named Entity Recognition (NER) refers to the process of automatically identifying and classifying named entities with specific meanings from unstructured text, essentially labeling key information in the text. Pre-trained language model-based NER technology involves enabling the model to learn a general language before refining the entity recognition task. Essentially, it leverages the powerful language understanding capabilities of pre-trained models to address the issues of insufficient contextual dependence and strong data dependence in NER. Pre-trained language models learn deep language rules through tasks such as bidirectional masking of large-scale unlabeled text (e.g., books), randomly masking some characters and allowing the model to predict the original characters. This learning includes word collocation, grammatical structure, and contextual semantic relationships. For example, "apple" refers to fruit in "eat apples" but a brand in "Apple Inc." The model ultimately outputs a general language vector that represents semantics. Dependency parsing reveals the grammatical roles and logical connections of sentence components by identifying dependency relationships between words, rather than relying on traditional fixed syntactic frameworks such as subject, verb, object, modifier, and complement.

[0085] Step S302: Based on the preset power business ontology library, the relationships between key entities are mapped to obtain standardized mapping results; the standardized mapping results include domain concepts, business objectives, quality constraints, and action instructions.

[0086] Among them, the power business ontology library refers to a pre-built database that has defined the hierarchical structure, attributes and constraints of concepts such as equipment types, business scenarios and performance indicators in the power field. For example, "fault diagnosis" is a subclass of "business scenario" and "accuracy" is a data attribute of "quality constraint".

[0087] For example, similarity calculation and rule-based reasoning can be used to map the extracted entity-relation pairs to concepts in a pre-defined power business ontology library, resulting in standardized mapping results. For instance, "transmission line" in natural language can be mapped to an instance of the "TransmissionLine" class in the ontology, and "accuracy not less than 50 meters" can be mapped to the "AccuracyConstraint" attribute with the value "50 meters". The standardized mapping results are output as machine-understandable semantic triples, covering domain concepts (such as equipment objects), business objectives (such as diagnostic tasks), quality constraints (such as accuracy thresholds), and action instructions (such as data acquisition operations). Similarity calculation is a technique for measuring the degree of similarity between two or more objects (such as text), transforming the abstract concept of similarity into a quantifiable numerical value, typically between 0 and 1, or between -1 and 1. The closer the value is to the upper limit, the more similar the objects are. Rule-based reasoning is a reasoning method for decision-making or problem-solving based on pre-defined logical rules. It is an "IF-THEN" rule base and the process of matching facts and deriving conclusions using rules.

[0088] Step S303: Construct an intent description model based on the standardized mapping results; the intent description model includes a domain concept layer, a business goal layer, a quality constraint layer, and an action instruction layer.

[0089] The domain concept layer includes entities and their attributes instantiated based on ontology mapping results, such as target device ID and spatial location; the business objective layer is used to define task types (such as fault diagnosis) and their priorities; the quality constraint layer is used to quantify performance indicators such as maximum latency and minimum accuracy; and the action instruction layer is used to define specific operations, such as collecting current data.

[0090] For example, an object-oriented approach can be used to encapsulate each layer as an independent object and associate elements of each layer by reference (e.g., associating a business target object with its corresponding quality constraint object) to obtain a defined model framework. Based on the triples in the mapping result and the defined model framework, an intent description model is generated. This model structure follows the principle of layered design, with lower layers providing basic semantic support to upper layers, such as business entities defined in the domain concept layer being referenced by the business target layer. The object-oriented approach is a programming / design philosophy that first clearly defines the architectural layers, with each layer undertaking a single responsibility. Each layer is no longer a collection of fragmented functions / methods but is encapsulated into objects or classes with clearly defined attributes and behaviors. Following the principle of layered design decomposes a complex system into multiple functionally independent and clearly defined layers. Each layer focuses only on its own core task and interacts with adjacent layers through standardized interfaces, avoiding direct dependencies across layers.

[0091] Step S304: Serialize the hierarchical intent description model into a standard JSON-LD format file to obtain the business intent description file.

[0092] Serialization refers to the process of converting structured model data in memory into a storable and transmissible text format. The serialization process utilizes the context mechanism of JSON-LD (JavaScript Object Notation for Linked Data) to link to the semantic definition of the power business ontology library through the @context field, ensuring that the file is both machine-readable and semantically clear. The standard JSON-LD format is a structured data format based on JSON, which allows machines to more clearly understand the semantics (i.e., the meaning and association of content) of web pages, documents, etc., rather than being only human-readable. @context is a vocabulary and rules for defining data, used to tell the machine the meaning of fields.

[0093] For example, the objects at each layer of the intent description model are converted into JSON key-value pairs to obtain the business intent description file: the domain concept layer outputs device IDs and a list of attributes, the business goal layer outputs task types and parameters, the quality constraint layer outputs indicator key-value pairs, such as "maxDelay" (maximum delay): "5s", and the action instruction layer outputs a list of operation commands. The nested structure of JSON-LD maintains the hierarchical relationship, such as the business goal object embedding an associated quality constraint object. The generated business intent description file also contains metadata such as intent ID and creation time, and its integrity is ensured through digital signatures.

[0094] In this embodiment, key entities and their relationships are extracted from high-level power business intents, and semantic standardization is performed using an ontology library to obtain standardized mapping results. Based on these standardized mapping results, an intent description model is constructed, and standardized files are generated using JSON-LD serialization. This effectively bridges the gap between business semantics and underlying configuration, improving the accuracy of intent descriptions and system interoperability.

[0095] In one embodiment, the business intent description file is semantically parsed to generate perceived task requirements, including:

[0096] Step S401: Perform syntactic and semantic analysis on the business intent description file to obtain core elements; the core elements include business scenario, target device, accuracy indicators and latency requirements.

[0097] Among them, syntax analysis and semantic analysis refer to parsing the language structure from different dimensions to jointly judge the compliance of the text form and the accuracy of the meaning. Syntax analysis refers to performing lexical and syntactic analysis on the file to verify whether it conforms to the syntax specification of JSON-LD, including checking the matching of brackets, the format of key-value pairs, and the correctness of the @context field. Semantic analysis is based on the context link in JSON-LD to the preset power business ontology library, and identifies core elements such as business scenarios (such as fault diagnosis), target equipment (such as specific transmission lines), accuracy indicators (such as 0.5% error tolerance) and latency requirements (such as response time less than 5 seconds) through semantic reasoning.

[0098] For example, the business intent description file is subjected to syntactic and semantic analysis to obtain core elements such as business scenario, target device, accuracy indicators and latency requirements.

[0099] Step S402: Query the knowledge graph of the power industry to obtain the domain rules and historical cases related to the core elements.

[0100] Among them, the power domain knowledge graph is a structured knowledge collection built for the business scenarios of the power industry. It is stored using a graph database and includes entities such as equipment types, business scenarios, and performance constraints, as well as their relationships. The domain rules in the power domain knowledge graph are stored in the form of "IF-THEN" production rules. For example, IF business scenario is fault diagnosis, THEN requires the collection of current and voltage data. Historical cases include sensor configuration parameters and execution effect data in similar scenarios in the past. The graph database is a non-relational database designed based on graph theory models. It uses nodes and relationships, also called edges, to store, organize, and query data, emphasizing the relationships between data, rather than the structured storage of tables, rows, and columns as the core of traditional databases.

[0101] For example, the extracted core elements are converted into graph query statements. For instance, the business scenario "fault diagnosis" is mapped to a corresponding node in the graph, and its associated rule nodes (such as those requiring three-phase current data collection) and case nodes (such as historical fault records) are traversed. Using the graph database's indexing and path query algorithms, the domain rules and historical cases with the highest semantic similarity to the current core element are selected. The path query algorithm is used to find the optimal path or all possible paths between the "starting node" and the "target node" in the graph structure.

[0102] Step S403: Based on domain rules and historical cases, the abstract goals in the high-level power business intent are converted into data collection requirements through the rule reasoning engine.

[0103] The rule-based reasoning engine is an AI component that automates reasoning based on predefined logical rules. Essentially, it transforms human-understandable "IF-THEN" logic into a machine-executable decision-making process. It includes a rule base, a fact base, and an inference engine. The rule base stores all predefined reasoning rules, typically following an "IF-THEN" structure. The fact base stores the input data or known facts for the current reasoning, i.e., the specific information to be judged. The inference engine connects the rule base and the fact base, executing the core reasoning logic. By scanning the rule base, it finds all rules where the "condition part matches the information in the fact base." For example, it matches the information of "user k" in the fact base to a rule that prioritizes basic medical services. For the matched rule, it executes its conclusion part, such as generating medical recommendations. If the new conclusion can be used as the basis for subsequent reasoning, it feeds back into the fact base, supporting multi-step chained reasoning. For example, after recommending basic medical services, it further matches rules on which items are included in the basic medical services.

[0104] For example, based on the retrieved domain rules, the rule reasoning engine matches the abstract goal in the business intent (such as "achieving accurate fault location") with the rule conditions, triggering the corresponding rule execution and generating specific data collection instructions. For instance, when the rule "If the business goal is fault location and high accuracy is required, the sampling frequency must be no less than 1kHz" is triggered, the engine will output the specific numerical requirement for the sampling frequency. Simultaneously, historical case data is used for similarity-based case reasoning to assist decision-making, calculating the similarity between the current business scenario and historical cases (such as Euclidean distance), and adapting the parameter settings from the cases to the current requirements. The final reasoning result includes quantifiable data collection requirements such as data type (such as current waveform), sampling frequency (such as 2kHz), and accuracy level. The similarity-based case reasoning-assisted decision-making leverages past experience to solve new problems, simulating the human thought process of recalling similar experiences and adjusting coping strategies when facing unknown problems.

[0105] Step S404: Combine all data acquisition requirements to generate a perception task requirement file; the perception task requirement file includes a list of data types, data quality indicators, acquisition time windows, and spatial ranges.

[0106] For example, based on the template filling mechanism, the various collection requirements output are mapped to predefined task template fields, and consistency checks and arbitrations are performed on conflicting items (such as different sampling frequencies derived from multiple rules) (e.g., taking the highest frequency requirement). The file also contains metadata, such as task ID, generation timestamp, and version number, ultimately generating a perception task requirement file. This file can be in structured JSON format, containing a list of data types (such as voltage, current, and temperature), data quality indicators (such as signal-to-noise ratio greater than 40dB), collection time windows (such as start time and duration), and spatial ranges (such as device coordinates or area identifiers). The template filling mechanism is a basic logic for generating or processing information by dynamically supplementing personalized content through predefined task templates.

[0107] In this embodiment, key business elements are extracted from the business intent description file through syntactic and semantic analysis, domain knowledge and historical experience are acquired using knowledge graphs, and abstract goals are transformed into specific data collection requirements through rule-based reasoning. These requirements are then integrated to generate a standardized task file. This approach eliminates the semantic gap between business intent and sensor commands, enabling automated mapping from business requirements to device configuration and improving configuration efficiency and accuracy.

[0108] In one embodiment, an optimization selection process is performed on the candidate sensor set based on a multi-objective optimization algorithm to generate an optimal sensor combination, including:

[0109] Step S501: Construct a multi-objective optimization function; the multi-objective optimization aims to minimize sensor resource overhead and maximize the satisfaction of business requirements.

[0110]

[0111]

[0112] in, For a multi-objective optimization function, For data quality functions; This is the total time delay function; This is the total energy consumption function; Let X be the constraint, m be the decision variable, m be the total number of constraints, and j be the constraint identifier.

[0113] Among them, the data quality function The value is positively correlated with the sensor's measurement accuracy, sampling frequency, and data integrity; the total delay function Covering the end-to-end latency of data acquisition, transmission, and edge processing; total energy consumption function This refers to calculating the total power consumption of all selected sensors; constraints. This includes resource boundary constraints, such as the maximum allowable latency threshold, minimum data accuracy requirements, and sensor workload limits; the decision variable X is a binary vector, and its components... Indicates sensor Whether it is selected, 1 means selected, 0 means not selected.

[0114] For example, a multi-objective optimization function is constructed, which transforms the multi-objective optimization problem into a solvable form through a weighted comprehensive index, where the negative sign indicates that the minimum requirements for delay and energy consumption need to be transformed into negative optimization under the maximum framework.

[0115] Step S502: Initialize the population based on the candidate sensor set; each individual in the population represents a possible sensor combination scheme.

[0116] For example, a binary encoding strategy can be adopted, mapping each sensor in the candidate sensor set to a gene bit. If a sensor is selected, the gene bit value is 1; otherwise, it is 0. The population size can be dynamically determined based on the number of candidate sensors and computational resources. For example, the initial population can be generated through a combination of random generation and heuristic strategies: some individuals are randomly generated to ensure diversity, while some high-quality individuals are constructed based on sensor performance scores (such as accuracy) to accelerate convergence. Each individual needs to undergo feasibility verification, eliminating schemes that violate hard constraints (such as the maximum number of sensors limit) to ensure that the initial state of the population meets the basic constraints of the optimization problem. Here, the binary encoding strategy refers to the rules and methods for converting non-binary information such as values, characters, and instructions into binary (consisting only of 0s and 1s); random generation refers to the process of generating data, content, results, etc., from a specific range or set without preset rules or fixed order; and the heuristic strategy refers to replacing comprehensive and complex logical reasoning with empirical and intuitive judgments, helping people quickly find approximate solutions when information is incomplete, time is limited, or the problem is too complex, rather than pursuing the absolute optimal solution.

[0117] Step S503: Based on the multi-objective optimization function, the population is iteratively optimized using an improved genetic algorithm until the iteration process meets the convergence condition, and the optimal solution on the Pareto front is extracted from the final generation of the population.

[0118] Genetic algorithms include selection, crossover, and mutation operations: selection uses a tournament selection mechanism to select individuals with higher fitness from the current population; crossover generates new offspring by exchanging gene segments of individuals through single-point or uniform crossover; mutation randomly flips gene positions of individuals with a small probability, introducing new features; the tournament selection mechanism is a commonly used selection strategy in genetic algorithms, analogous to the survival-of-the-fittest rules in sports tournaments, with the aim of selecting individuals with stronger fitness from the population; the improvement of the genetic algorithm lies in the introduction of an adaptive crossover and mutation probability adjustment mechanism, which dynamically adjusts the intensity of operations according to the diversity of the population; the convergence condition can be set as the improvement rate of the optimal solution being lower than a threshold or reaching the maximum number of iterations for several consecutive generations; Pareto refers to the Pareto principle (80 / 20 rule), that is, about 80% of the results are determined by 20% of the key factors, which is essentially a thinking tool for hierarchically ranking things according to their importance / contribution.

[0119] For example, for each individual in the population, its multi-objective optimization function value is calculated. This result is used to evaluate the individual's quality and is called fitness assessment. Based on the multi-objective optimization function values ​​of all individuals, a fast non-dominated sorting is performed, dividing the entire population into Pareto levels with different priorities. For individuals within each Pareto level, their crowding distance is calculated to maintain the distribution of the solution set. A tournament selection mechanism based on level and crowding distance is used for selection. For example, several individuals are randomly selected, and the individual with the highest Pareto level (smallest number) is chosen; if the levels are the same, the individual with the larger crowding distance is chosen. After the algorithm iteration terminates, all individuals with the highest Pareto level are directly extracted from the last generation of the population. This set is the Pareto optimal solution set, i.e., the non-dominated solution set. Fast non-dominated sorting involves progressively filtering solutions based on non-dominated relationships, ultimately classifying all solutions into different ranks according to their quality. The lower the rank, the better the solution. The fast non-dominated sorting process is as follows: First round: Traverse the entire population and find all individuals not dominated by any other individual. These individuals are called non-dominated solutions and are assigned to the first Pareto rank (Rank 1). These solutions constitute the optimal boundary (Pareto front) that the current population can find. Second round: Temporarily remove individuals from the first rank from the population. Among the remaining individuals, search for non-dominated solutions again. These found individuals are assigned to the second Pareto rank (Rank 2). They are dominated by individuals in the first rank but dominate individuals of lower ranks. Repeat the process: Repeat the above steps until all individuals in the population are assigned to a Pareto rank (Rank 1, Rank 2, Rank 3). (k3,...) The smaller the rank number, the better the individual. Domination refers to the following condition: if two individuals V and U are both satisfied, then V is said to dominate U: 1) V is no worse than U in all objectives: that is, for every component of the objective function F(X), the value of V is greater than or equal to the value of U; 2) V is strictly better than U in at least one objective: that is, there is at least one objective component where the value of V is strictly greater than the value of U. Non-dominated solution refers to an individual that cannot be dominated by any other individual in the entire population. Crowding distance refers to the process of ranking all individuals within the same Pareto rank for each objective function, calculating the difference between the values ​​of the left and right adjacent individuals in the objective function for each individual, normalizing this difference, and then summing the differences calculated for the individual in all objective functions.

[0120] Step S504: Decode the optimal solution to obtain the optimal sensor combination.

[0121] Decoding refers to mapping the binary vector output by the genetic algorithm back to the actual sensor identifier, for example, including sensors with a gene bit value of 1 in the final combination.

[0122] For example, the optimal solution is decoded. In cases where there are multiple non-dominated solutions in the Pareto solution set, the comprehensive score of each solution can be calculated based on the quality, latency, and energy consumption weights explicitly stated in the business intent. The highest decomposition is then selected as the optimal combination. If no weights are specified in the business intent, the solution with the closest Euclidean distance to the ideal point (i.e., the theoretical point composed of the optimal values ​​of all objective functions) in the Pareto front is selected by default. After decoding, a list of sensor IDs and their theoretical performance indicators are output. Euclidean distance is one of the most commonly used distance metrics (such as similarity), used to describe the straight-line distance between two points in multidimensional space, i.e., the square root of the sum of the squares of the differences in the x-coordinates and y-coordinates of the two points, corresponding to the Pythagorean theorem: the length of the hypotenuse of a right triangle.

[0123] In this embodiment, a multi-objective function is constructed to define the optimization direction, generating diverse initial solution sets. An improved genetic algorithm is then used to select the optimal solution, which is decoded into a combination of physical sensors. This effectively overcomes the limitations of traditional single-objective optimization and significantly reduces resource consumption while ensuring business requirements are met.

[0124] In one embodiment, based on a structured dataset, the degree of satisfaction of the current business intent is evaluated, and an intent satisfaction evaluation result is generated, including:

[0125] Step S601: Assign weight coefficients to the evaluation indicators of the current business intent; the evaluation indicators include data accuracy, data integrity, timeliness and specific indicators based on business results; the weight coefficients are used to reflect the importance of the indicators to the business intent.

[0126] For example, subjective weights for indicators such as data accuracy, data integrity, timeliness, and specific indicators based on business performance can be calculated using the Analytic Hierarchy Process (AHP). Simultaneously, objective weights can be calculated using the entropy weight method. The final weights are obtained by weighted fusion of subjective and objective weights. The AHP breaks down complex, multi-objective decision-making problems into an ordered hierarchical structure, quantifying the importance of each factor through a combination of subjective judgment and objective calculation, ultimately leading to the optimal decision. The entropy weight method objectively determines indicator weights based on the amount of data information. It measures the dispersion (i.e., discriminative power) of an indicator using entropy values; the higher the dispersion, the more effective information the indicator provides, and the greater its weight should be, and vice versa.

[0127] Step S602: Calculate the actual values ​​of each evaluation index based on the structured dataset.

[0128] For example, based on a structured dataset, the actual value of data accuracy indicators can be quantified by calculating the root mean square error or mean absolute percentage error between measured values ​​and standard reference values; the actual value of data integrity indicators can be determined by statistically analyzing the proportion of missing data and considering the severity of consecutive missing segments in the time series; timeliness indicators can be obtained by calculating the end-to-end latency from data acquisition to edge processing completion and normalizing it in conjunction with the maximum allowable latency required by the business; and the actual value of business performance indicators can be customized according to specific business types, for example, location accuracy is used in fault diagnosis scenarios, and prediction error rate is used in power flow prediction scenarios.

[0129] Step S603: Calculate the intention satisfaction score based on the actual values ​​and weighting coefficients of each evaluation indicator.

[0130] For example, the actual values ​​of each evaluation indicator are normalized, and the normalized actual values ​​are weighted and summed with the assigned weight coefficients to obtain the intent satisfaction score:

[0131]

[0132] in, The score represents the degree of satisfaction with the intention. These are the weighting coefficients. This represents the normalized actual value. For example, the weighted summation of the original score can be mapped to the [0,1] interval using the Sigmoid function, making the score distribution more consistent with actual perception and yielding the final intention satisfaction score. Normalization refers to mapping raw data of different magnitudes and units to a unified numerical range, typically [0,1] or [-1,1]. The Sigmoid function is used to map input values ​​of any range to the (0,1) interval; the most common is the Logistic Sigmoid function, with the formula:

[0133]

[0134] x is the input, which can be any real number; e is the natural constant, approximately 2.718; σ(x) is the output, which strictly falls between (0,1).

[0135] Step S604: Generate intent satisfaction evaluation results based on intent satisfaction score and preset threshold.

[0136] The preset threshold is set according to the business type. For example, a higher threshold, such as 0.9, is set for critical business (such as fault handling), while a lower threshold, such as 0.7, is set for non-critical business (such as environmental monitoring).

[0137] For example, an intent satisfaction assessment result is generated based on an intent satisfaction score and a preset threshold. This assessment result uses a three-level classification: "Satisfied" is output when the score is above the threshold; "Partially Satisfied" is output when the score is below the threshold but above 80% of the threshold; and "Not Satisfied" is output when the score is below 80% of the threshold. The assessment result includes a confidence level evaluation, quantified by calculating the variance of the most recent N assessment results; the smaller the variance, the higher the confidence level. The final output is a structured assessment report, including the satisfaction level, analysis of key shortcomings, and improvement suggestions.

[0138] In this embodiment, weights are allocated through a combination of subjective and objective assessments, actual performance is quantified based on multi-dimensional indicators, satisfaction is comprehensively calculated using a weighted model, and tiered evaluation results are generated based on a threshold mechanism. This approach transforms vague business requirements into quantifiable evaluation metrics.

[0139] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0140] Based on the same inventive concept, this application also provides an edge computing-based power IoT sensor device configuration system for implementing the aforementioned edge computing-based power IoT sensor device configuration method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more edge computing-based power IoT sensor device configuration system embodiments provided below can be found in the limitations of the edge computing-based power IoT sensor device configuration method described above, and will not be repeated here.

[0141] In one exemplary embodiment, such as Figure 3 As shown, a power Internet of Things (IoT) sensor device configuration system 300 based on edge computing is provided, including:

[0142] The document generation module 301 is used to perform formal description processing on the input high-level power business intent and generate a business intent description file; the business intent description file includes business objectives, quality constraints and execution actions;

[0143] The requirement generation module 302 is used to perform semantic parsing processing on the business intent description file to generate perception task requirements; the perception task requirements include the required data types, quality requirements, and time and space constraints.

[0144] The candidate sensor set generation module 303 is used to match the sensor resources managed by the edge computing platform according to the requirements of the sensing task to obtain a candidate sensor set.

[0145] The sensor combination generation module 304 is used to perform optimization and selection processing on the candidate sensor set based on a multi-objective optimization algorithm to generate the optimal sensor combination;

[0146] The configuration parameter set generation module 305 is used to calculate the configuration parameters of each sensor in the optimal sensor combination according to the specific indicators required by the sensing task; and to combine the configuration parameters to obtain a detailed configuration parameter set.

[0147] In one exemplary embodiment, the system further includes:

[0148] The data acquisition module is used to collect raw data from each sensor in the optimal sensor combination after executing the detailed configuration parameter set; and to perform structured processing on the raw data at the edge nodes to obtain a structured dataset.

[0149] The evaluation result generation module is used to evaluate the degree of satisfaction of the current business intent based on a structured dataset and generate intent satisfaction evaluation results.

[0150] The judgment result generation module is used to determine whether to trigger the dynamic adjustment mechanism based on the intent satisfaction evaluation result, and obtain the judgment result;

[0151] The instruction generation module is used to generate adjustment instructions based on the intent satisfaction evaluation results if the judgment result triggers the dynamic adjustment mechanism. The adjustment instructions are used to instruct the steps from generating the perception task requirements to generating the detailed configuration parameter set to be re-executed according to the intent satisfaction evaluation results, so as to obtain the updated detailed configuration parameter set.

[0152] In one embodiment, the file generation module 301 is further configured to:

[0153] Receive high-level power business intents input in natural language or structured forms; and extract key entities and relationships between key entities from the high-level power business intents.

[0154] Based on a pre-defined power business ontology library, the relationships between key entities are mapped to obtain standardized mapping results. The standardized mapping results include domain concepts, business objectives, quality constraints, and action instructions.

[0155] Based on the standardized mapping results, an intent description model is constructed; the intent description model includes a domain concept layer, a business goal layer, a quality constraint layer, and an action instruction layer.

[0156] The hierarchical intent description model is serialized into a standard JSON-LD format file to obtain the business intent description file.

[0157] In one embodiment, the demand generation module 302 is further configured to:

[0158] Perform syntactic and semantic analysis on the business intent description file to obtain core elements; the core elements include business scenario, target device, accuracy indicators and latency requirements.

[0159] Query the knowledge graph of the power industry to obtain domain rules and historical cases related to core elements;

[0160] Based on domain rules and historical cases, the abstract goals in the high-level power business intent are transformed into data collection requirements through a rule reasoning engine;

[0161] All data collection requirements are combined to generate a perception task requirement file; the perception task requirement file includes a list of data types, data quality indicators, collection time windows, and spatial ranges.

[0162] In one embodiment, the sensor combination generation module 304 is further configured to:

[0163] Construct a multi-objective optimization function; the multi-objective optimization aims to minimize sensor resource overhead and maximize the satisfaction of business requirements.

[0164]

[0165]

[0166] in, For a multi-objective optimization function, For data quality functions; This is the total time delay function; This is the total energy consumption function; Here, X represents the constraint, m represents the total number of constraints, and j represents the constraint identifier.

[0167] Based on the candidate sensor set, initialize the population; each individual in the population represents a possible sensor combination scheme.

[0168] Based on a multi-objective optimization function, an improved genetic algorithm is used to iteratively optimize the population until the iteration process meets the convergence condition, and the optimal solution on the Pareto front is extracted from the final generation of the population.

[0169] The optimal solution is decoded to obtain the optimal sensor combination.

[0170] In one embodiment, the evaluation result generation module is further configured to:

[0171] Assign weight coefficients to the evaluation metrics for the current business intent; the evaluation metrics include data accuracy, data integrity, timeliness, and specific metrics based on business results; the weight coefficients are used to reflect the importance of the metrics to the business intent;

[0172] Calculate the actual values ​​of each evaluation indicator based on the structured dataset;

[0173] Based on the actual values ​​and weighting coefficients of each evaluation indicator, the intention satisfaction score is calculated.

[0174] Based on the intent satisfaction score and a preset threshold, an intent satisfaction assessment result is generated.

[0175] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the power Internet of Things sensing device configuration method based on edge computing as described above.

[0176] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0177] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0178] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for configuring power Internet of Things (IoT) sensing devices based on edge computing, characterized in that, The method includes: The input high-level power business intent is formally described to generate a business intent description file; the business intent description file includes business objectives, quality constraints, and execution actions. The business intent description file is semantically parsed to generate perception task requirements; the perception task requirements include the required data types, quality requirements, and spatiotemporal constraints. Based on the requirements of the perception task, the sensor resources managed by the edge computing platform are matched to obtain a set of candidate sensors; Based on a multi-objective optimization algorithm, the candidate sensor set is optimized and selected to generate the optimal sensor combination; Based on the specific indicators required by the perception task, the configuration parameters of each sensor in the optimal sensor combination are calculated; and the configuration parameters are combined to obtain a detailed set of configuration parameters.

2. The method according to claim 1, characterized in that, The method further includes: After executing the detailed configuration parameter set, raw data from each sensor in the optimal sensor combination is collected; and the raw data is processed in a structured manner at the edge node to obtain a structured dataset. Based on the structured dataset, the degree of satisfaction of the current business intent is evaluated, and an intent satisfaction evaluation result is generated. Based on the intent satisfaction assessment results, determine whether to trigger the dynamic adjustment mechanism, and obtain the judgment result; If the judgment result triggers the dynamic adjustment mechanism, an adjustment instruction is generated based on the intent satisfaction evaluation result; the adjustment instruction is used to instruct the steps of generating the perception task requirements to generating the detailed configuration parameter set to be re-executed according to the intent satisfaction evaluation result, so as to obtain the updated detailed configuration parameter set.

3. The method according to claim 1, characterized in that, The formal description processing of the input high-level power service intent to generate a service intent description file includes: Receive high-level power service intents input in natural language or structured forms; and extract key entities and relationships between the key entities from the high-level power service intents. Based on a pre-defined power business ontology library, the relationships between the key entities are mapped to obtain standardized mapping results; the standardized mapping results include domain concepts, business objectives, quality constraints, and action instructions. Based on the standardized mapping results, an intent description model is constructed; the intent description model includes a domain concept layer, a business goal layer, a quality constraint layer, and an action instruction layer; The hierarchical intent description model is serialized into a standard JSON-LD format file to obtain the business intent description file.

4. The method according to claim 1, characterized in that, The step of semantically parsing the business intent description file to generate perception task requirements includes: The business intent description file is subjected to syntactic and semantic analysis to obtain core elements; the core elements include business scenario, target device, accuracy indicators and latency requirements. Query the knowledge graph of the power industry to obtain the domain rules and historical cases related to the core elements; Based on the domain rules and historical cases, the abstract goals in the high-level power business intent are transformed into data collection requirements through the rule reasoning engine; All the data acquisition requirements are combined to generate a perception task requirement file; the perception task requirement file includes a list of data types, data quality indicators, acquisition time windows, and spatial ranges.

5. The method according to claim 1, characterized in that, The step of optimizing and selecting the candidate sensor set based on a multi-objective optimization algorithm to generate the optimal sensor combination includes: Construct a multi-objective optimization function; the multi-objective optimization aims to minimize sensor resource overhead and maximize the satisfaction of business requirements. in, For a multi-objective optimization function, For data quality functions; This is the total time delay function; This is the total energy consumption function; Here, X represents the constraint, m represents the total number of constraints, and j represents the constraint identifier. Based on the candidate sensor set, a population is initialized; each individual in the population represents a possible sensor combination scheme. Based on the multi-objective optimization function, the population is iteratively optimized using an improved genetic algorithm until the iteration process meets the convergence condition, and the optimal solution on the Pareto front is extracted from the final generation of the population. The optimal solution is decoded to obtain the optimal sensor combination.

6. The method according to claim 2, characterized in that, The process of evaluating the degree of satisfaction of the current business intent based on the structured dataset and generating an intent satisfaction evaluation result includes: Assign weight coefficients to the evaluation indicators of the current business intent; the evaluation indicators include data accuracy, data integrity, timeliness, and specific indicators based on business results; the weight coefficients are used to reflect the importance of the indicators to the business intent; Based on the structured dataset, calculate the actual values ​​of each of the evaluation metrics; Based on the actual values ​​of each evaluation indicator and the weighting coefficients, the intention satisfaction score is calculated. Based on the intent satisfaction score and the preset threshold, an intent satisfaction evaluation result is generated.

7. A power Internet of Things (IoT) sensor device configuration system based on edge computing, characterized in that, The system includes: The document generation module is used to perform formal description processing on the input high-level power business intent and generate a business intent description file; the business intent description file includes business objectives, quality constraints and execution actions; The requirement generation module is used to perform semantic parsing processing on the business intent description file to generate perception task requirements; the perception task requirements include the required data types, quality requirements, and spatiotemporal constraints. The candidate sensor set generation module is used to match the sensor resources managed by the edge computing platform according to the requirements of the sensing task to obtain a candidate sensor set. The sensor combination generation module is used to perform optimization selection processing on the candidate sensor set based on a multi-objective optimization algorithm to generate the optimal sensor combination; The configuration parameter set generation module is used to calculate the configuration parameters of each sensor in the optimal sensor combination according to the specific indicators required by the sensing task; and combine the configuration parameters to obtain a detailed configuration parameter set.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images