Cloud-side collaborative substation intelligent decision analysis operator library construction method

Abstract

The invention discloses a cloud-side collaborative substation intelligent decision analysis operator library construction method, belongs to the technical field of power system automation and intelligent decision, and aims to solve the problems that a traditional substation decision analysis system is dispersed in operators, poor in reusability, insufficient in cloud-side resource adaptation, high in decision delay and difficult to maintain. Firstly, operator demand classification and specification definition are completed through substation business scene disassembly; secondly, an atomic operator and a composite operator are designed according to the function modules, and standardized interface packaging is adopted; unified management and index construction of operator metadata are realized through a cloud edge collaborative registration center; then dynamically configuring a cloud edge scheduling strategy based on resource awareness and service priority; and finally, iterative optimization of the operator library is realized through incremental updating and version management and control. Based on a cloud edge collaborative architecture, a unified, reusable and dynamically adaptive intelligent decision analysis operator library is constructed through a five-step core process of'demand modeling-operator design-packaging registration-collaborative scheduling-iterative maintenance '.

Description

Technical Field

[0001] This invention belongs to the field of power system automation and intelligent decision-making technology, specifically relating to a method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations. Background Technology

[0002] As the core hub of energy conversion and transmission in the power system, the real-time perception of the substation's operating status, accurate fault diagnosis, and dynamic optimization control have become crucial supports for ensuring the safe and stable operation of the power grid. Currently, substations are developing towards "unmanned operation, intelligent inspection, and online decision-making." The amount of monitoring data generated by field equipment (such as sensors, intelligent terminals, and measurement and control devices) is growing exponentially; the daily data volume generated by a single 220kV smart substation has reached the TB level, and the decision response latency requirements for core operations such as fault diagnosis and emergency control have been reduced to the hundreds of milliseconds level. Traditional substation decision analysis systems mostly adopt a centralized architecture, with all data processing and decision calculations completed in the cloud, which presents the following problems:

[0003] 1. The massive amounts of real-time data generated by substation field equipment need to be uploaded to the cloud for processing. Due to network bandwidth limitations, this can easily lead to delays in decision response and fail to meet real-time requirements such as fault diagnosis.

[0004] 2. Decision-making algorithms developed by different manufacturers are independent of each other, lack a unified interface and classification standard, making them difficult to reuse, resulting in redundant development and waste of resources;

[0005] 3. The characteristics of strong cloud computing power and edge nodes being close to the device were not fully considered. The operator deployment method was fixed and could not be dynamically adjusted according to the resource status, resulting in excessive cloud load or idle edge resources.

[0006] 4. The decision-making algorithm is deeply coupled with the business system. Upgrading or fixing a certain algorithm requires reconstructing the entire system, resulting in poor scalability and flexibility.

[0007] In existing technologies, some solutions attempt to build substation decision-making algorithm libraries, but without combining cloud-edge collaborative architecture, they still suffer from insufficient real-time performance and low resource utilization. Other cloud-edge collaborative technologies are applied to power systems, but they do not design dedicated operator classification, encapsulation, and scheduling mechanisms for substation intelligent decision-making scenarios, making it difficult to meet the personalized needs of substation operations.

[0008] Therefore, a cloud-edge collaborative method for constructing a substation intelligent decision analysis operator library is needed to solve the problems of dispersed operators, poor reusability, insufficient cloud-edge resource adaptation, high decision latency, and difficult maintenance in existing substation decision analysis systems. Summary of the Invention

[0009] The purpose of this invention is to provide a method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations, so as to solve the problems mentioned in the background art.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations, comprising the following steps:

[0011] Step S1: Based on the substation business scenario, decompose the core tasks of intelligent decision-making, classify the operators in multiple dimensions according to function type, resource requirements, and real-time requirements, and clarify the input and output specifications, computational complexity and dependencies of each type of operator.

[0012] Step S2: Design atomic operators and composite operators, and adopt standardized interfaces to unify the calling format, data interaction protocol and exception handling mechanism of operators;

[0013] Step S3: Build a cloud-edge collaborative registration center, upload operator metadata and build a multi-dimensional index to achieve fast operator retrieval and matching;

[0014] Step S4: Edge nodes collect resource status in real time and upload it to the cloud. The cloud combines business priorities to formulate dynamic scheduling strategies and decide the deployment and operation mode of operators on cloud-edge nodes.

[0015] Step S5: Upgrade operator functions through version management and incremental update mechanisms, and optimize operator classification and scheduling strategies based on operation monitoring data.

[0016] It should be noted in the solution that, in step S1, the substation business scenarios include status monitoring, fault diagnosis, load forecasting, optimization control, and safety assessment. The corresponding operator function types are divided into data preprocessing operators, feature extraction operators, model inference operators, decision output operators, and result verification operators. The resource requirements include CPU or GPU utilization and memory or disk usage. The real-time requirements are divided into: high priority is response latency ≤ 100ms, medium priority is 100ms < response latency ≤ 500ms, and low priority is response latency > 500ms.

[0017] It is worth noting that in step S2, the standardized interface adopts the RESTful API or gRPC protocol, and the encapsulated content includes the input parameter list, output parameter format, execution timeout threshold, and exception return code definition; the composite operator is configured with combination logic through a visual interface, and supports custom execution order, branch conditions, and data flow path of atomic operators.

[0018] Furthermore, it should be noted that in step S3, the operator metadata includes operator identifier, function description, resource requirement threshold, applicable scenario tag, and version number; the multi-dimensional index includes function tag index, resource requirement index, and real-time priority index, and a hierarchical index structure is adopted to improve retrieval efficiency.

[0019] As a preferred implementation, the retrieval matching algorithm of the multi-dimensional index adopts a weighted scoring formula:

[0020]

[0021] in, The score is the matching score for the functional tags, ranging from 0 to 100. The resource requirement matching score ranges from 0 to 100. The score is the real-time priority matching score, with a value range of 0-100; To match weights, satisfy Returns rating during search A list of operators.

[0022] In a preferred implementation, in step S4, the dynamic scheduling strategy employs an adaptive scheduling algorithm based on reinforcement learning, with the objective function being a multi-objective optimization formula:

[0023]

[0024] in, To delay operator decision, This represents the resource utilization rate of cloud-edge nodes (value range 0-1). Energy consumption for operator operation; For the weighting coefficients, satisfying Furthermore, it is dynamically adjusted according to business scenarios, especially in high real-time scenarios. .

[0025] In a preferred implementation, in step S5, the incremental update mechanism is implemented using an operator difference algorithm, and the amount of updated data is calculated as follows:

[0026]

[0027] in, This represents the total data volume for the new version of the operator. This represents the complete data volume for the old version of the operator. The actual incremental data to be transmitted includes only code differences and metadata updates, and the transmission overhead is further reduced through data compression algorithms.

[0028] As a preferred implementation, it also includes: step S6, real-time calculation of operator call frequency, execution success rate, response time, and resource utilization, and generation of a running status report; when the operator execution success rate is lower than 95% or the response time exceeds a preset threshold, an alarm is triggered and an exception log is automatically recorded to provide data support for operator iteration and scheduling optimization.

[0029] Compared with existing technologies, the cloud-edge collaborative substation intelligent decision analysis operator library construction method provided by this invention has at least the following beneficial effects:

[0030] (1) This invention uses a dynamic scheduling strategy of cloud-edge collaboration to deploy high-priority operators (such as fault diagnosis) on edge nodes first. Data can be processed without being uploaded to the cloud. The decision latency is reduced from 300ms-1s in the traditional centralized architecture to less than 100ms, which fully meets the real-time requirements of fault diagnosis, emergency control and other services. Furthermore, through a unified classification system, standardized interface encapsulation and cloud-edge collaborative registration center, operators are modularized and standardized. Operators developed by different manufacturers can be seamlessly connected and reused, avoiding redundant development.

[0031] (2) The present invention is based on an adaptive scheduling algorithm of reinforcement learning, which dynamically allocates operators according to the resource status of cloud and edge nodes and business priorities, avoiding excessive cloud load and idle edge resources; at the same time, the operator library supports incremental updates and version management. When adding or upgrading operators, there is no need to reconstruct the entire system. The update can be completed by pushing incremental data through the registration center, which reduces the update time by more than 80% compared with the full update; and the operator classification system can be dynamically optimized according to changes in business scenarios to adapt to the needs of adding new devices and expanding business, improving system scalability by 60% and reducing maintenance costs by 45%.

[0032] (3) This invention uses a result verification operator and a full life cycle monitoring mechanism to monitor the operator's operating status in real time, promptly detect decision anomalies and operator faults, trigger alarms and record logs, and provide support for troubleshooting. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating the method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to the present invention. Detailed Implementation

[0034] The present invention will be further described below with reference to embodiments.

[0035] Please see Figure 1 This invention provides a method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations, comprising the following steps:

[0036] Step S1: Based on the substation business scenario, decompose the core tasks of intelligent decision-making, classify the operators in multiple dimensions according to function type, resource requirements, and real-time requirements, and clarify the input and output specifications, computational complexity and dependencies of each type of operator.

[0037] Specifically, a comprehensive review of the core business of substations is conducted, including five major scenarios: equipment condition monitoring (such as transformer temperature, circuit breaker mechanical characteristics, and insulation condition monitoring), fault diagnosis (such as inter-turn short circuit, partial discharge, and mechanical jamming fault identification), load forecasting (short-term / ultra-short-term load forecasting), optimized control (reactive power compensation, voltage regulation control, and energy storage charging and discharging control), and safety assessment (equipment health assessment and power grid operation risk assessment). The core decision-making tasks and technical indicators for each scenario are clearly defined.

[0038] The operators are divided into five categories: data preprocessing operators (responsible for data cleaning, normalization, missing value imputation, and data format conversion), feature extraction operators (responsible for extracting time-domain features, frequency-domain features, entropy features, and correlation features), model inference operators (responsible for inference calculations of models such as neural networks, decision trees, and support vector machines), decision output operators (responsible for fault type localization, control command generation, and evaluation result output), and result verification operators (responsible for decision confidence verification and result rationality verification). This ensures high cohesion and low coupling of operator functions.

[0039] The resource consumption of each operator was quantified through offline testing, including CPU / GPU utilization (in %), memory usage (in MB), and disk storage requirements (in KB), and divided into three levels: low resource requirement (CPU utilization < 20%, memory usage < 512MB), medium resource requirement (20% ≤ CPU utilization < 50%, 512MB ≤ memory usage < 2GB), and high resource requirement (CPU utilization ≥ 50% or requires GPU support, memory usage ≥ 2GB).

[0040] Based on the urgency of the business, operators are divided into three levels: high priority (response delay ≤ 100ms, such as fault diagnosis and emergency control operators), medium priority (100ms < response delay ≤ 500ms, such as load forecasting and status monitoring operators), and low priority (response delay > 500ms, such as security assessment and health assessment operators).

[0041] Operator specification definition: Clearly define the input and output specifications of various operators (input data format supports JSON / CSV / ProtocolBuffers, output results support numerical, boolean values, control instruction strings, matrices, etc.), computational complexity (time complexity O(n), O(nlogn), O(n²), etc.) and dependencies (e.g., a certain composite operator needs to rely on the output of a data preprocessing operator as input).

[0042] Step S2: Design atomic operators and composite operators, and adopt standardized interfaces to unify the calling format, data interaction protocol and exception handling mechanism of operators;

[0043] The standardized interface adopts the RESTful API or gRPC protocol, and the encapsulated content includes the input parameter list, output parameter format, execution timeout threshold, and exception return code definition; the composite operator is configured with combination logic through a visual interface, and supports custom execution order, branch conditions, and data flow path of atomic operators.

[0044] Specifically, atomic operators are indivisible, smallest decision units, developed for a single specific function, such as "3σ outlier removal operator", "Min-Max normalization operator", "FFT frequency domain feature extraction operator", "BP neural network fault diagnosis operator", "decision confidence verification operator", etc. Each atomic operator implements only one core function to ensure its reusability.

[0045] Composite operators are designed for complex decision-making tasks and consist of at least two atomic operators combined according to a preset logic. For example, the combination logic of the "transformer fault diagnosis composite operator" is "data cleaning operator → normalization operator → FFT feature extraction operator → BP neural network inference operator → fault location operator → confidence verification operator". The combination logic of composite operators is implemented through a visual configuration interface. Users can customize the execution order of atomic operators, branch conditions (such as switching to SVM inference operator when the feature value exceeds the threshold) and data flow path, and complete the configuration of complex operators without writing code.

[0046] Step S3: Build a cloud-edge collaborative registration center, upload operator metadata and build a multi-dimensional index to achieve fast operator retrieval and matching;

[0047] Specifically, the encapsulated operator metadata is uploaded to the registration center. The metadata includes the operator's unique identifier ID, function description, function tag, resource requirement threshold, real-time priority, applicable scenario tag, interface address, version number, development time, update log, and other information to ensure the integrity of the operator information.

[0048] Based on metadata, we construct functional tag indexes, resource requirement indexes, and real-time priority indexes, employing a hierarchical inverted index structure to improve operator retrieval efficiency. During index construction, a weighted scoring algorithm is used to achieve accurate operator matching; the scoring formula is as follows:

[0049]

[0050] in, The score is the matching score for the functional tags, ranging from 0 to 100. The resource requirement matching score ranges from 0 to 100. The score is the real-time priority matching score, with a value range of 0-100; To match weights, satisfy Returns rating during search A list of operators.

[0051] Step S4: Edge nodes collect resource status in real time and upload it to the cloud. The cloud combines business priorities to formulate dynamic scheduling strategies and decide the deployment and operation mode of operators on cloud-edge nodes.

[0052] Specifically, edge nodes (deployed at substation sites, including edge gateways and local servers) collect real-time data on their own computing resources (CPU / GPU utilization, computing power), storage resources (remaining memory, available disk space), and network resources (uplink / downlink bandwidth, transmission latency). The data collection frequency is once every 10 seconds, and the data is uploaded to the cloud resource management module via the lightweight communication protocol (MQTT). The cloud resource management module summarizes the resource status of all edge nodes, establishes a resource status ledger, and updates the resource load in real time.

[0053] The cloud-based adaptive scheduling algorithm, which combines substation business priorities, operator resource requirements, and cloud-edge node resource status, decides on the deployment and operation mode of operators in the cloud or at edge nodes. The scheduling objective function is a multi-objective optimization formula:

[0054]

[0055] in, To delay operator decision, This represents the resource utilization rate of cloud-edge nodes (value range 0-1). Energy consumption for operator operation; For the weighting coefficients, satisfying Furthermore, it is dynamically adjusted according to business scenarios, especially in high real-time scenarios. .

[0056] The scheduling and execution logic is as follows: for high-priority operators, if the edge node resources meet their operating requirements (CPU utilization < 70%, memory remaining > 50%), they will be deployed on the edge node first to directly process the data of the field equipment and reduce data transmission latency; if the edge node resources are insufficient, the cloud will temporarily allocate idle computing power, and the edge node will only be responsible for data collection and instruction issuance. After the cloud completes the calculation, it will feed the results back to the edge node.

[0057] Medium priority operator: Dynamically allocated according to the cloud-edge resource load balancing status. When the cloud load rate is <60%, it is deployed in the cloud; otherwise, it is deployed on the edge node to ensure cloud-edge resource load balancing.

[0058] Low-priority operators: uniformly deployed in the cloud, leveraging the powerful computing capabilities of the cloud to batch process computing tasks from multiple substations, reducing resource consumption at edge nodes.

[0059] Step S5: Upgrade operator functions through version management and incremental update mechanisms, and optimize operator classification and scheduling strategies based on runtime monitoring data;

[0060] Specifically, each operator is assigned a unique version number (in the format of "major version number.minor version number.revision number", such as V1.2.0), a version iteration log is recorded (including new features, bug fixes, performance optimizations, etc.), and a version rollback function is supported (when a new version operator malfunctions, it can be quickly rolled back to a historical stable version).

[0061] Incremental updates are implemented using an operator differential algorithm, transmitting only the code and metadata of the operator update portion to avoid bandwidth consumption caused by full data transmission; the update data volume is calculated as follows:

[0062]

[0063] in, This represents the total data volume for the new version of the operator. This represents the complete data volume for the old version of the operator. The actual incremental data to be transmitted includes only code differences and metadata updates, and the transmission overhead is further reduced through data compression algorithms.

[0064] Furthermore, it is worth noting that this also includes: step S6, which involves real-time calculation of the operator's call frequency, execution success rate, response time, and resource utilization, and generating a running status report; when the operator's execution success rate is lower than 95% or the response time exceeds a preset threshold, an alarm is triggered and an exception log is automatically recorded, providing data support for operator iteration and scheduling optimization.

[0065] In summary, the cloud-edge collaborative substation intelligent decision analysis operator library construction method proposed in this invention achieves modular and standardized operator management through a series of innovative technologies, including multi-dimensional operator classification, standardized interface design, cloud-edge collaborative registration center, adaptive scheduling algorithm, and version management and incremental update mechanism. This significantly improves the real-time performance of decision-making, system scalability, and resource utilization. This method not only solves the problems of high decision latency and uneven resource utilization in traditional centralized architectures but also lowers the operator development threshold through a visual configuration interface, supporting seamless integration and reuse of operators from different manufacturers, effectively avoiding redundant development. Furthermore, based on reinforcement learning dynamic... The scheduling strategy can intelligently allocate computing power based on the resource status of cloud and edge nodes and business priorities, ensuring that high-priority tasks are processed first on edge nodes to reduce data transmission latency, while low-priority tasks are uniformly deployed in the cloud to make full use of cloud computing resources. In addition, the version management and incremental update mechanism makes operator function upgrades more efficient, requiring only the transmission of updated code and metadata, significantly reducing bandwidth consumption and update time. Finally, by monitoring the operator's running status in real time and generating reports, anomalies can be detected and handled in a timely manner, providing strong data support for operator iteration and scheduling optimization, thereby ensuring the stable and reliable operation of the substation intelligent decision-making system.

[0066] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0067] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations, characterized by: Includes the following steps: Step S1: Based on the substation business scenario, decompose the core tasks of intelligent decision-making, classify the operators in multiple dimensions according to function type, resource requirements, and real-time requirements, and clarify the input and output specifications, computational complexity and dependencies of each type of operator. Step S2: Design atomic operators and composite operators, and adopt standardized interfaces to unify the calling format, data interaction protocol and exception handling mechanism of operators; Step S3: Build a cloud-edge collaborative registration center, upload operator metadata and build a multi-dimensional index to achieve fast operator retrieval and matching; Step S4: Edge nodes collect resource status in real time and upload it to the cloud. The cloud combines business priorities to formulate dynamic scheduling strategies and decide the deployment and operation mode of operators on cloud-edge nodes. Step S5: Upgrade operator functions through version management and incremental update mechanisms, and optimize operator classification and scheduling strategies based on operation monitoring data.

2. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 1, characterized in that: In step S1, the substation business scenarios include status monitoring, fault diagnosis, load forecasting, optimization control, and safety assessment. The corresponding operator function types are divided into data preprocessing operators, feature extraction operators, model inference operators, decision output operators, and result verification operators. The resource requirements include CPU or GPU utilization and memory or disk usage. The real-time requirements are divided into: high priority with response latency ≤ 100ms, medium priority with 100ms < response latency ≤ 500ms, and low priority with response latency > 500ms.

3. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 1, characterized in that: In step S2, the standardized interface adopts the RESTful API or gRPC protocol, and the encapsulated content includes the input parameter list, output parameter format, execution timeout threshold, and exception return code definition; the composite operator is configured with combination logic through a visual interface, and supports custom execution order, branch conditions, and data flow path of atomic operators.

4. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 1, characterized in that: In step S3, the operator metadata includes operator identifier, function description, resource requirement threshold, applicable scenario tag, and version number; the multi-dimensional index includes function tag index, resource requirement index, and real-time priority index, and a hierarchical index structure is adopted to improve retrieval efficiency.

5. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 4, characterized in that: The retrieval matching algorithm of the multi-dimensional index adopts a weighted scoring formula: in, The score is the matching score for the functional tags, ranging from 0 to 100. The resource requirement matching score ranges from 0 to 100. The score is the real-time priority matching score, with a value range of 0-100; To match weights, satisfy Returns rating during search A list of operators.

6. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 1, characterized in that: In step S4, the dynamic scheduling strategy employs an adaptive scheduling algorithm based on reinforcement learning, with the objective function being a multi-objective optimization formula: in, To delay operator decision, This represents the resource utilization rate of cloud-edge nodes (value range 0-1). Energy consumption for operator operation; For the weighting coefficients, satisfying Furthermore, it is dynamically adjusted according to business scenarios, especially in high real-time scenarios. .

7. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 1, characterized in that: In step S5, the incremental update mechanism is implemented using the operator difference algorithm, and the amount of updated data is calculated as follows: in, This represents the total data volume for the new version of the operator. This represents the complete data volume for the old version of the operator. The actual incremental data to be transmitted includes only code differences and metadata updates, and the transmission overhead is further reduced through data compression algorithms.

8. The method for constructing a cloud-edge collaborative intelligent decision analysis operator library for substations according to claim 1, characterized in that: Also includes: Step S6: Real-time statistics are performed on the call frequency, execution success rate, response time, and resource utilization of the sub-task, and a running status report is generated. When the operator execution success rate is below 95% or the response time exceeds a preset threshold, an alarm is triggered and an exception log is automatically recorded, providing data support for operator iteration and scheduling optimization.

Images