Control simulation method based on artificial intelligence

Patent metadata

Field	Value
Publication number	US-12373704-B2
Application number	US-202318506608-A
Country	US
Kind code	B2
Filing date	Nov 10, 2023
Priority date	Nov 14, 2022
Publication date	Jul 29, 2025
Grant date	Jul 29, 2025

Abstract

Disclosed is a method for control simulation based on artificial intelligence according to an exemplary embodiment of the present disclosure. Specifically, according to the present disclosure, a computing device obtains a first state information, a second state information, and a control information, and generates first output information based on the first state information, the second state information, and the control information by using an artificial neural network model including a sequential neural network. In this case, the first output information includes one or more output variables, and at least some of the one or more output variables correspond to variables included in the second state information, and the first output information is generated based on applying an attention mechanism to each of the one or more output variables.

First claim

What is claimed is:

1. A method performed by a computing device for control simulation of a control environment based on artificial intelligence, the method comprising:
obtaining a first state information, a second state information, and a control information, wherein the first state information includes variables related to an outside of the control environment and that are independent of the control information, and wherein the second state information includes variables related to an inside of the control environment; and
generating a first output information based on the first state information, the second state information, and the control information by using an artificial neural network model including a sequential neural network,
wherein the first output information includes a plurality of output variables,
at least some of the plurality of output variables correspond to variables included in the second state information, and
the first output information is generated based on applying an attention mechanism generating query vectors and value vectors of a number corresponding to the plurality of output variables from a hidden state.

2. The method of claim 1 , wherein the generating of the first output information including the prediction value for the second state information based on the first state information, the second state information, and the control information by using the artificial neural network model including the sequential neural network includes
generating a hidden state information at a control time point based on inputting the first state information, the second state information, and the control information into the sequential neural network,
calculating attention values at the control time point with respect to the plurality of output variables, respectively based on the hidden state information, and
generating the first output information based on the attention values.

3. The method of claim 2 , wherein the calculating of the attention values at the control time point with respect to the plurality of output variables, respectively based on the hidden state information includes
generating value vectors and query vectors based on the hidden state information, wherein the number of value vectors and the number of query vectors is equal to a number of output variables in the plurality of output variables,
calculating an attention score for each of the plurality of output variables based on the query vector,
calculating an attention distribution based on the attention score, and
calculating attention values with respect to the plurality of output variables, respectively based on the attention distribution and the value vector.

4. The method of claim 3 , wherein the generating of the query vectors and the value vectors of a number corresponding to the plurality of output variables based on the hidden state information includes
generating the value vectors and the query vectors of a number corresponding to the plurality of output variables based on inputting the hidden state information into a fully connected layer.

5. The method of claim 3 , wherein the generating of the query vectors and the value vectors of a number corresponding to the plurality of output variables based on the hidden state information includes
concatenating the control information to the hidden state information, and
generating the value vectors and the query vectors of a number corresponding to the plurality of output variables based on the concatenated information.

6. The method of claim 2 , further comprising:
generating a second output information based on the hidden state information; and
generating a final output information based on the first output information and the second output information.

7. The method of claim 6 , wherein the final output information is generated based on a result of a weighted sum operation of a value of the output variable included in the first output information and a value of the corresponding output variable included in the second output information.

8. The method of claim 7 , wherein a weight of the weighted sum operation may be set differently for each output variable.

9. A computer program stored in a non-transitory computer-readable storage medium, wherein the computer program performs operations for control simulation of a control environment based on artificial intelligence when executed by at least one processor included in a computing device, and the operations comprising:
an operation of obtaining a first state information, a second state information, and a control information, wherein the first state information includes variables related to an outside of the control environment and that are independent of the control information, and wherein the second state information includes variables related to an inside of the control environment; and
an operation of generating a first output information based on the first state information, the second state information, and the control information by using an artificial neural network model including a sequential neural network, and
wherein the first output information includes a plurality of output variables,
at least some of the plurality of output variables correspond to variables included in the second state information, and
the first output information is generated based on applying an attention mechanism generating query vectors and value vectors of a number corresponding to the plurality of output variables from a hidden state.

10. A computing device comprising:
at least one processor; and
a memory,
wherein the at least one processor is configured to:
obtain a first state information, a second state information, and a control information, wherein the first state information includes variables related to an outside of a control environment, and wherein the second state information includes variables related to an inside of the control environment and that are independent of the control information, and
generate a first output information based on the first state information, the second state information, and the control information by using an artificial neural network model including a sequential neural network, and
the first output information includes a plurality of output variables,
at least some of the plurality of output variables correspond to variables included in the second state information, and
the first output information is generated based on applying an attention mechanism generating query vectors and value vectors of a number corresponding to the plurality of output variables from a hidden state.

Assignees

• Makinarocks Co Ltd
• Hanon Systems

Inventors

• Park Hongje
• Choi Sanghyeok
• Park Jongwon
• Kim Minseop
• Lee Jeonghoon
• Kim Joongjae
• Ju Ikchan

Classifications

• G06N3/10 Primary

  Interfaces, programming languages or software development kits, e.g. for simulating neural networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title