CN-122018859-A - Algorithm supermarket system based on new energy scene

CN122018859ACN 122018859 ACN122018859 ACN 122018859ACN-122018859-A

Abstract

The algorithm supermarket system based on the new energy scene mainly comprises a user layer, an interface layer, a scheduling layer, an adaptation layer, an algorithm resource layer and a supportability module, wherein the user layer is used for providing algorithm demand submission, result receiving and feedback interaction for various users of new energy, the interface layer is used for providing a standardized call inlet, the scheduling layer is used for scheduling and distributing algorithm resources, the adaptation layer is used for dynamically matching user demands with the algorithm, the algorithm resource layer is used for storing and managing the algorithm, the supportability module is used for providing data support, safety guarantee and monitoring evaluation, and the algorithm is executed by the scheduling layer according to scheduling strategies after the adaptation layer is matched with the corresponding algorithm according to the user demands. The system can effectively realize efficient matching and calling of multiple types of new energy scenes and algorithm resources, and greatly improves the convenience of new energy users in acquiring adaptive algorithm resources.

Inventors

SHEN ZHONGMING
HU KUN
LI PEIHAN
DENG XINYU
WU XUEJIE

Assignees

中电建新能源集团股份有限公司

Dates

Publication Date: 20260512
Application Date: 20251216

Claims (9)

1. The algorithm supermarket system based on the new energy scene is characterized by sequentially comprising a user layer, an interface layer, a scheduling layer, an adaptation layer, an algorithm resource layer and a supporting module from top to bottom, wherein: the user layer is used for providing algorithm demand submission, result receiving and feedback interaction for various users of new energy; the interface layer is used for providing a standardized call entry and supporting multi-protocol adaptation; The scheduling layer is used for scheduling and distributing computing resources, sequencing priorities and balancing loads; The adaptation layer is used for dynamically matching the user requirements with the algorithm; the algorithm resource layer is used for algorithm storage, registration, auditing and version management; The support module is used for providing data support, safety guarantee, monitoring and evaluation; after the adaptation layer matches the corresponding algorithm according to the user requirement, the scheduling layer calls the computing power resource according to the scheduling strategy to execute the algorithm.
2. The system of claim 1, wherein the system employs a "cloud + edge" hybrid deployment architecture: The cloud node is used for deploying an algorithm resource layer, a scheduling layer and a supporting module and is responsible for centralized management of algorithms, large-scale calculation and data storage; The edge end node is deployed with a lightweight adaptation layer, an interface layer and a high-frequency calling algorithm, is deployed on the local of the new energy project, and meets the low-delay calling requirement; The cloud end and the edge end realize data synchronization through a communication network, wherein the data synchronization comprises algorithm version updating, operation log uploading and scene data synchronization.
3. The system according to claim 1, wherein the system realizes matching of new energy scenes required by a user and algorithm resources through setting of a label, wherein the label comprises: (1) The application scene label of the algorithm resource; (2) A new energy scene tag comprising: the scene basic attribute tag comprises one or more of item type, installed capacity, equipment model and geographic coordinates; The data tag comprises one or more of data type, data format and data sampling frequency; The requirement labels comprise one or more of functional requirements, performance requirements and optimization targets.
4. The system of claim 3, wherein the algorithm resource layer comprises an algorithm execution package, algorithm metadata, a test data set, a historical operating record; Wherein the algorithm execution package supports containerized deployment; the algorithm metadata comprises one or more of input and output parameters, applicable scene tags, performance indexes and dependent environments.
5. The system according to claim 4, wherein the specific method for implementing the dynamic matching of the user requirements and the algorithm resources by the adaptation layer comprises the following steps: After receiving the user demand label, carrying out similarity calculation with the applicable scene label of the algorithm metadata, and screening candidate algorithms with the adaptation degree larger than a set threshold value; data adaptation, namely automatically converting a data format input by a user into a format required by an algorithm, and complementing the missing data; Parameter calibration, namely automatically adjusting algorithm built-in parameters according to new energy scene parameters required by a user, wherein the parameters comprise feature weights of a prediction model and/or constraint conditions of an optimization algorithm; The compatibility verification is that the executable of the algorithm after the adaptation is verified through the lightweight test case, so that the running error is avoided; And outputting an adaptation result.
6. The system of claim 1, wherein the scheduling layer's scheduling allocation policy comprises: (1) Disassembling the user demand into a function demand, a performance demand and a resource demand; (2) Setting priority based on the type of the user and the emergency degree of the demand; the priority calculation is performed by weighted summation, and the formula is as follows: Wherein: 、、 Is a weight coefficient; a priority score, wherein the higher the score is, the higher the priority is; User type weight, wherein the power grid dispatching center > new energy operators > third party developers; e, a demand emergency degree weight, wherein the real-time scheduling class > the quasi real-time analysis class > the non-real-time planning class; S, scene importance weight, wherein the power grid safety correlation > the power generation efficiency correlation > the operation and maintenance management correlation; (3) Distributing computing resources from high to low according to the priority, wherein an operation node deployed at the edge end responds to a calling request of a local new energy project preferentially so as to reduce network delay; (4) Monitoring CPU and/or memory occupancy rate of the operation node in real time, and distributing a new calculation power calling request to the node with load less than or equal to a set load threshold value; With the aim of minimizing the load difference of each computing node, the following optimization model is established: wherein N is the number of available compute nodes; Real-time load rate of the ith node; The average load rate of all nodes is calculated, ; Constraint conditions: setting a load threshold value; (5) Preemptive scheduling triggers: When a high priority request arrives, the following conditions are met to trigger the preemption of a low priority task: new request priority scores; The priority score of the current running task; A priority differential threshold; idle computing resource duty ratio; Idle resource threshold.
7. The system of claim 1, wherein the standardized call portal provided by the interface layer comprises: A RESTful API for responding to non-real time requirements; the WebSocket interface is used for responding to the real-time requirement; The SDK toolkit is used for locally deploying requirements and supporting Java/Python/C++ multi-language calling.
8. The system of claim 1, wherein the interactive functions of the user layer settings include one or more of user-defined algorithm parameters, viewing an algorithm running log, feeding back algorithm effects, applying for algorithm customization modifications.
9. The system of any one of claims 1-8, wherein the supportive module comprises: The data support module integrates new energy real-time monitoring data, historical data, meteorological data and power grid data and provides data cleaning, fusion and query services; The security module is used for realizing data transmission encryption, algorithm calling authority control, user data isolation storage and encryption deployment of core algorithm codes; The monitoring evaluation module is used for monitoring the running state of the algorithm in real time and evaluating the effect of the algorithm, and when the algorithm runs abnormally, the monitoring evaluation module automatically triggers an alarm and switches to the standby algorithm.

Description

Algorithm supermarket system based on new energy scene Technical Field The invention relates to the technical field of algorithm sharing and on-demand calling in a new energy scene, in particular to an algorithm supermarket system based on the new energy scene. Background With the rapid development of new energy industries, the demands for various data processing and analysis are increasing. In a new energy scene, different applications often need specific algorithms to realize data processing and analysis, but at present, a specialized algorithm management system for the new energy scene is still in a starting stage, the existing algorithm resources are dispersed, a unified management and sharing platform is lacking, so that development efficiency is low, resource waste is serious, and meanwhile, a new energy enterprise is difficult to find an algorithm meeting own requirements quickly when searching for a proper algorithm. The intelligent upgrading method is mainly characterized in that in the prior art, an algorithm service platform focuses on a single field (such as photovoltaic prediction only) or a general computer field, an adaptation mechanism is not designed aiming at the specificity of a new energy scene, the core requirements of 'cross-scene compatibility, dynamic adaptation, low-delay calling and safety and controllability' cannot be met, and the intelligent upgrading efficiency of a new energy system is restricted. Disclosure of Invention The algorithm supermarket system suitable for the new energy scene is suitable for an algorithm service architecture of the whole life cycle (planning design, operation monitoring, optimal scheduling and operation and maintenance management) of the new energy project, and can realize efficient matching, dynamic adaptation and safe calling of multiple types of new energy scenes and algorithm resources. The system sequentially comprises a user layer, an interface layer, a scheduling layer, an adaptation layer, an algorithm resource layer and a supporting module from top to bottom, wherein: the user layer is used for providing algorithm demand submission, result receiving and feedback interaction for various users of new energy; the interface layer is used for providing a standardized call entry and supporting multi-protocol adaptation; The scheduling layer is used for scheduling and distributing computing resources, sequencing priorities and balancing loads; The adaptation layer is used for dynamically matching the user requirements with the algorithm; the algorithm resource layer is used for algorithm storage, registration, auditing and version management; The support module is used for providing data support, safety guarantee, monitoring and evaluation; after the adaptation layer matches the corresponding algorithm according to the user requirement, the scheduling layer calls the computing power resource according to the scheduling strategy to execute the algorithm. Further, the system adopts a 'cloud end and edge end' hybrid deployment architecture: The cloud node is used for deploying an algorithm resource layer, a scheduling layer and a supporting module and is responsible for centralized management of algorithms, large-scale calculation and data storage; The edge end node is deployed with a lightweight adaptation layer, an interface layer and a high-frequency calling algorithm, is deployed on the local of the new energy project, and meets the low-delay calling requirement; The cloud end and the edge end realize data synchronization through a communication network, wherein the data synchronization comprises algorithm version updating, operation log uploading and scene data synchronization. Further, the system realizes the matching of the new energy scene required by the user and the algorithm resource through the setting of the label, wherein the label comprises: (1) The application scene label of the algorithm resource; (2) A new energy scene tag comprising: the scene basic attribute tag comprises one or more of item type, installed capacity, equipment model and geographic coordinates; The data tag comprises one or more of data type, data format and data sampling frequency; The requirement labels comprise one or more of functional requirements, performance requirements and optimization targets. Further, the storage content of the algorithm resource layer comprises an algorithm execution packet, algorithm metadata, a test data set and a history operation record; Wherein the algorithm execution package supports containerized deployment; the algorithm metadata comprises one or more of input and output parameters, applicable scene tags, performance indexes and dependent environments. Further, the specific method for realizing dynamic matching of the user demand and the algorithm resource by the adaptation layer comprises the following steps: After receiving the user demand label, carrying out similarity calculation with the applicable scene label of the algori