Search

CN-122019076-A - Agricultural intelligent model reasoning scheduling system and method

CN122019076ACN 122019076 ACN122019076 ACN 122019076ACN-122019076-A

Abstract

The application belongs to the technical field of artificial intelligence, and provides an agricultural intelligent model reasoning and scheduling system and method. The system comprises an application access layer, an inference scheduling center layer and a model running layer, wherein the application access layer comprises a plurality of agricultural service systems, the agricultural service systems initiate model calling requests through predefined fixed inference endpoints, the inference scheduling center layer is connected with the application access layer and is used for receiving the model calling requests and carrying out endpoint standardization processing on the model calling requests to obtain unified inference requests, and the model running layer is used for receiving the unified inference requests sent by the inference scheduling center layer, generating inference results and observation information according to the unified inference requests and feeding the inference results and the observation information back to the application access layer. The application can realize high availability, easy management and traceable unified management of cross-platform and cross-region model service, and remarkably improves the development efficiency and operation stability of agricultural intelligent application.

Inventors

  • YANG XINTING
  • LI YIBIN
  • SUN CHUANHENG
  • LUO NA
  • XING BIN

Assignees

  • 北京市农林科学院信息技术研究中心

Dates

Publication Date
20260512
Application Date
20251226

Claims (10)

  1. 1. An agricultural intelligent model inference scheduling system, the system comprising: An application access layer comprising a plurality of agricultural business systems, wherein the agricultural business systems initiate model call requests through predefined fixed reasoning endpoints; The reasoning scheduling center layer is connected with the application access layer and is used for receiving the model calling request and carrying out endpoint standardization processing on the model calling request so as to obtain a unified reasoning request; And the model operation layer is used for receiving the unified reasoning request sent by the reasoning dispatching center layer, generating a reasoning result and observation information according to the unified reasoning request, and feeding back the reasoning result and the observation information to the application access layer.
  2. 2. The system of claim 1, wherein the inferred dispatch hub layer comprises: The deployment mapping module is used for maintaining the mapping relation between the registration information of the model instance and the real-time running state and determining the instance positioning information of the model call request; The endpoint standardization module is used for converting the heterogeneous interface paths in the instance positioning information into unified endpoints consistent with the predefined fixed reasoning endpoints according to preset conversion rules; the request shaping module is used for carrying out verification, format conversion and standardization processing on the model call request so as to obtain the unified reasoning request; the unified scheduling module is used for distributing the unified reasoning request according to the self-adaptive fault-tolerant strategy; and the observation treatment module is used for collecting the system log, the performance index and the event information and generating dynamic configuration feedback information.
  3. 3. The system of claim 2, wherein the preset transformation rules include a matching pattern, transformation logic information, and weight information; The matching mode is used for identifying heterogeneous interface characteristics of the target model instance; The conversion logic information is used for converting the identified heterogeneous interfaces into the unified endpoint; the weight information is used to determine a conversion priority.
  4. 4. The system of claim 2, wherein the request shaping module is configured to perform the following process flow: abstracting the model call request into a basic request vector, wherein the basic request vector comprises model identification information, model version information, input data and reasoning parameters; creating request context information, wherein the request context information comprises request identification information, a target endpoint and a time stamp; converting the base request vector into a normalized vector by a shaping function, wherein the conversion process comprises at least one of: Checking the input field white list and filling a default value; Setting a configurable upper input volume limit; And (5) format conversion.
  5. 5. The system of claim 2, wherein the unified scheduling module is configured to perform the following process flows: When the request is overtime or the server returns preset error information, executing retry operation; Triggering fusing of the target model instance or switching to a standby instance according to the failure rate and the dynamic threshold value of the real-time monitoring; when the fusing or degrading condition is met, a caching result is returned or the standby instance is switched to.
  6. 6. The system of claim 2, wherein the observation and remediation module comprises: the log sub-module is used for generating a structured log and desensitizing sensitive data; The index submodule is used for outputting performance indexes including request quantity, success rate, delay distribution, retry and fusing state indexes; and the event sub-module is used for automatically triggering alarm notification and/or generating audit events and/or generating strategy adjustment suggestions.
  7. 7. An agricultural intelligent model reasoning scheduling method, characterized in that it is applied to the system according to any one of claims 1-6, said method comprising: Receiving a model call request, wherein the model call request is initiated based on a fixed endpoint; determining a target model instance according to the model identifier and version in the model call request; converting the model call request into a unified reasoning request through endpoint standardization processing; and generating an inference result and observation information according to the unified inference request and the target model instance.
  8. 8. An agricultural intelligent model reasoning and scheduling device is characterized by comprising: the receiving module is used for receiving a model call request, wherein the model call request is initiated based on a fixed endpoint; the determining module is used for determining a target model instance according to the model identifier and the version in the model calling request; the conversion module is used for converting the model call request into a unified reasoning request; and the generation module is used for generating an inference result and observation information according to the unified inference request and the target model instance.
  9. 9. An electronic device comprising a memory, a processor and a computer program stored on the memory and running on the processor, wherein the processor implements the agricultural intelligent model inference scheduling method of claim 7 when executing the computer program.
  10. 10. A non-transitory computer readable storage medium having stored thereon a computer program, which when executed by a processor implements the agricultural intelligent model inference scheduling method of claim 7.

Description

Agricultural intelligent model reasoning scheduling system and method Technical Field The application relates to the technical field of artificial intelligence, in particular to an agricultural intelligent model reasoning and scheduling system and method. Background In the intelligent agriculture field, along with the deep application of the internet of things, remote sensing and edge computing technologies, various AI models are widely deployed in heterogeneous environments from the cloud to the edge so as to support intelligent services such as pest and disease identification, crop estimation, agricultural machinery scheduling and the like. However, the current lack of a unified model service governance framework results in the online reasoning services exhibiting a highly fragmented state-different teams use multiple reasoning frameworks (e.g., pyTorch Serving, tensorFlow Serving, etc.), exposed application programming interface (Application Programming Interface, API) endpoints (e.g./v 1/models #)Predict,/predictions) are mutually incompatible, model examples are dispersed in private cloud, public cloud and field edge nodes, and unified discovery and scheduling are difficult. Under the mode, model reasoning and scheduling depend on manual adaptation at respective application ends or simple forwarding by means of a basic gateway, so that huge development and operation and maintenance burden can be caused, model version iteration is slow, cross-region collaboration is difficult, an effective fault tolerance and observable mechanism is lacking in a complex and changeable agricultural network environment, the rigid requirements of high service availability and safety audit cannot be met, and the large-scale expansion of an agricultural intelligent system is severely restricted. Disclosure of Invention The application provides an agricultural intelligent model reasoning scheduling method, device, equipment and storage medium, which are used for solving the problems that an agricultural intelligent model reasoning service interface is not uniform, cross-environment deployment is difficult to schedule, and reasoning service stability and observability are insufficient in the prior art. The application provides an agricultural intelligent model reasoning and scheduling system, which comprises: The application access layer comprises a plurality of agricultural service systems, and the agricultural service systems initiate model calling requests through predefined fixed reasoning endpoints; The reasoning dispatching center layer is connected with the application access layer and is used for receiving the model calling request and carrying out endpoint standardization processing on the model calling request so as to obtain a unified reasoning request; The model operation layer is used for receiving the unified reasoning request sent by the reasoning dispatching center layer, generating a reasoning result and observation information according to the unified reasoning request, and feeding the reasoning result and the observation information back to the application access layer. In one embodiment, the inferred dispatch hub layer includes: The deployment mapping module is used for maintaining the mapping relation between the registration information of the model instance and the real-time running state and determining the instance positioning information of the model call request; The endpoint standardization module is used for converting heterogeneous interface paths in the instance positioning information into unified endpoints consistent with the predefined fixed reasoning endpoints according to preset conversion rules; the request shaping module is used for checking, converting and standardizing the model call request to obtain a unified reasoning request; The unified scheduling module is used for distributing the unified reasoning request according to the self-adaptive fault-tolerant strategy; and the observation treatment module is used for collecting the system log, the performance index and the event information and generating dynamic configuration feedback information. In one embodiment, the preset conversion rule comprises a matching mode, conversion logic information and weight information; The matching mode is used for identifying heterogeneous interface characteristics of the target model instance; The conversion logic information is used for converting the identified heterogeneous interfaces into unified endpoints; the weight information is used to determine the conversion priority. In one embodiment, the request shaping module is configured to perform the following process flow: abstracting a model call request into a basic request vector, wherein the basic request vector comprises model identification information, model version information, input data and reasoning parameters; creating request context information, wherein the request context information comprises request identification information, a t