IP Library Granted Patent US 12663764
Granted Patent B2
US 12663764 · App. 17/109,870 · Granted Jun 23, 2026

Inference apparatus, training apparatus, and inference method

Inventors: Kento Kawaharazuka (Tokyo-to, JP); Toru Ogawa (Tokyo-to, JP)
Assignee: Preferred Networks, Inc.
G05B13/027G06N3/044G06N3/08
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12663764
App. No.
17/109,870
Granted
Jun 23, 2026
Kind
B2
Abstract

An inference apparatus includes one or more memories; and one or more processors. The one or more processors configured to acquire a latent state from input data regarding a control target; acquire a future latent state from the latent state and control data; infer, from the future latent state, a time series of a task to be executed by the control target to be controlled based on the control data; calculate a loss between the time series of the task and data indicating a target state; and update the control data based on the loss.

Claims (46)

1 . An inference method, comprising:

acquiring data on a state of a control target;

inferring, based on at least (i) the data on the state of the control target and (ii) control data for controlling the control target, a time series of a task to be executed by the control target to be controlled based on the control data;

calculating at least a loss between the time series of the task and data indicating a target state; and

updating at least a part of the control data based on at least the loss, the updated control data being utilized to control the control target, wherein

the control target is an industrial plant to produce a predetermined chemical compound,

the data on the state of the control target is observable data in the industrial plant,

the control data is related to a valve opening to control the industrial plant,

the time series of the task is related to an actual yield of the predetermined chemical compound produced in the industrial plant, and

the data indicating the target state is related to a production amount of the predetermined chemical compound.

2 . An inference apparatus, comprising:

one or more memories; and

one or more processors configured to:

acquire data on a state of a control target;

infer, based on at least (i) the data on the state of the control target and (ii) control data for controlling the control target, a time series of a task to be executed by the control target to be controlled based on the control data;

calculate at least a loss between the time series of the task and data indicating a target state; and

update at least a part of the control data based on at least the loss, the updated control data being utilized to control the control target, wherein

the control target is an industrial plant to produce a predetermined chemical compound,

the data on the state of the control target is observable data in the industrial plant,

the control data is related to a valve opening to control the industrial plant,

the time series of the task is related to an actual yield of the predetermined chemical compound produced in the industrial plant, and

the data indicating the target state is related to a production amount of the predetermined chemical compound.

3 . An inference apparatus, comprising:

one or more memories; and

one or more processors configured to:

acquire data on a state of a control target;

infer, based on at least (i) the data on the state of the control target and (ii) control data for controlling the control target, a time series of a task to be executed by the control target to be controlled based on the control data;

calculate at least a loss between the time series of the task and data indicating a target state; and

update at least a part of the control data based on at least the loss, the updated control data being utilized to control the control target, wherein

the control target is an industrial plant having one or more valves adjusting at least one of a temperature, a flow, a pressure, a level, or a density to produce a predetermined chemical compound,

the data on the state of the control target is observed data of the industrial plant,

the control data includes data on a valve opening of a valve in the industrial plant,

the time series of the task is related to an actual yield of the predetermined chemical compound produced in the industrial plant, and

the data indicating the target state is related to a production amount of the predetermined chemical compound.

4 . A non-transitory computer readable medium storing a program for executing a method, the method comprising:

acquiring observable data on a state of an industrial plant producing a predetermined chemical compound;

inferring, based on at least (i) the observable data on the state of the industrial plant and (ii) control data related to a valve opening for controlling the industrial plant, data related to an amount of the predetermined chemical compound produced if the industrial plant is controlled based on the control data;

calculating at least a loss between the inferred data related to the amount of the predetermined chemical compound and target data related to an amount of the predetermined chemical compound; and

updating at least a part of the control data based on at least the loss, the updated control data being utilized to control the industrial plant.

5 . A system comprising:

one or more memories; and

one or more processors configured to:

acquire observable data on a state of an industrial plant producing a predetermined chemical compound;

infer, based on at least (i) the observable data on the state of the industrial plant and (ii) control data related to a valve opening for controlling the industrial plant, data related to an amount of the predetermined chemical compound produced if the industrial plant is controlled based on the control data;

calculate at least a loss between the inferred data related to the amount of the predetermined chemical compound and target data related to an amount of the predetermined chemical compound; and

update at least a part of the control data based on at least the loss, the updated control data being utilized to control the industrial plant.