阅读说明：本技术 一种用于在现场设备上部署和执行机器学习模型的方法 (Method for deploying and executing machine learning model on field device ) 是由 J.索勒加里多 C.克恩 I.科根 于 2018-03-01 设计创作，主要内容包括：一种用于在目标现场设备上部署机器学习模型(MLM)的方法,该方法包括以下步骤：基于机器学习模型自动生成源代码文件的集合,其中,源代码文件的集合适于根据目标现场设备的预定能力执行机器学习模型；使用特定于目标现场设备的工具链将生成的源代码文件转换为模型二进制文件；以及在目标现场设备的存储器中部署模型二进制文件。(A method for deploying a Machine Learning Model (MLM) on a target field device, the method comprising the steps of: automatically generating a set of source code files based on the machine learning model, wherein the set of source code files is adapted to execute the machine learning model according to a predetermined capability of the target field device; converting the generated source code file into a model binary file using a target field device-specific toolchain; and deploying the model binary file in a memory of the target field device.)

1. A method for deploying a Machine Learning Model (MLM) on a target field device (10), comprising the steps of:

(a) automatically generating (S1) a set of Source Code Files (SCF) based on the trained machine learning model,

wherein the set of Source Code Files (SCF) is adapted to execute a machine learning model according to a predetermined capability of the target field device (10), and wherein the Machine Learning Model (MLM) is represented by a serialization model comprising text or binary strings encoding a graph topology comprising nodes, operations of the nodes, interconnections of the nodes and parameters of the nodes;

(b) converting (S2) the generated Source Code File (SCF) into a model binary file (MB) using a tool chain (3) specific to the target field device (10); and

(c) a model binary file (MB) is deployed (S3) in a memory (12) of a target field device (10).

2. The method according to claim 1, wherein the machine learning model is parsed by a model parser (2) to automatically generate at least one Source Code File (SCF) and extract parameters of the machine learning model.

3. Method according to claim 2, wherein the extracted parameters of the Machine Learning Model (MLM) are included as constants and/or static variables into the source code of the at least one generated Source Code File (SCF).

4. The method of claim 2, wherein the extracted parameters of the Machine Learning Model (MLM) are included into a separate parameter binary file (PB) that is deployed in the memory (12) of the target field device (10) along with the model binary file (MB).

5. The method of claim 2, wherein the extracted parameters of the Machine Learning Model (MLM) are included into a separate Parameter File (PF), which is converted into a parameter binary file (PB) using a target field device (10) -specific tool chain (3), wherein the parameter binary file (PB) is combined with the model binary file (MB) for deployment in the memory (12) of the target field device (10).

6. The method of one of the preceding claims 1 to 5, wherein the Model Binary (MB) and/or Parameter Binary (PB) are deployed locally by copying them into the memory (12) of the target field device (10) or remotely by copying them into the memory (12) of the target field device (10) via the network and the network interface of the target field device (10).

7. The method of any of the preceding claims 1 to 6, wherein the Machine Learning Model (MLM) for generating the Source Code File (SCF) comprises a machine learning model trained with training data and/or tested with test data.

8. The method of any preceding claim 1 to 7, wherein the Machine Learning Model (MLM) is parsed by a model parser (2), the model parser (2) having access to a data source (4), the data source (4) comprising a device-compatible operational module and a set of libraries of target field devices (10).

9. The method of one of the preceding claims 1 to 8, wherein the model binary file (MB) and the parameter binary file (PB) are stored in a non-volatile memory (12) of the target field device (10) and are loaded into a main memory (13) of the target field device (10) for execution by at least one processing unit (14) of the target field device (10).

10. The method of claim 1, wherein the generated source code is integrated with other device source code prior to generating the model binary file (MB).

11. A deployment system for deploying a Machine Learning Model (MLM) on a target field device (10),

the deployment system (1) comprises:

-a model parser (2) configured to parse a Machine Learning Model (MLM) to generate at least one Source Code File (SCF) adapted to execute the machine learning model using available resources of a target field device (10); and

-a target field device (10) -specific toolchain (3) adapted to convert at least one Source Code File (SCF) into a model binary file (MB) for deployment in a memory (12) of the target field device (10).

12. The deployment system of claim 11, wherein the model parser (2) has access to a database (4), the database (4) including a device-compatible operational module and a collection of libraries of target field devices (10).

13. The deployment system of any of the preceding claims 11 or 12, wherein the model parser (2) is configured to extract parameters of the Machine Learning Model (MLM) and include them as constants and/or static variables into the source code of the at least one generated Source Code File (SCF).

14. The deployment system of any of the preceding claims 11 or 12, wherein the model parser (2) is configured to extract parameters of the Machine Learning Model (MLM) and include them into a separate parameter binary file (PB) which is deployed in the memory (12) of the target field device (10) together with the model binary file (MB).

15. The deployment system of any of the preceding claims 11 or 12, wherein the model parser (2) is configured to extract parameters of the Machine Learning Model (MLM) and include them into a separate Parameter File (PF) which is converted into a parameter binary file (PB) using a tool chain (3) specific to the target field device (10), wherein the parameter binary file (PB) is deployed in the memory (12) of the target field device (10) together with the model binary file (MB).