# US 12,596,857 B2 · Method, apparatus and device for optimizing process parameter, and storage medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,596,857 B2
- Document ID: US12596857B2
- Status: Granted
- Kind: B2
- Application number: 19/175,628
- Filed: 2025-04-10
- Published: 2025-10-30
- Granted: 2026-04-07
- Earliest claimed: 2024-04-24
- Assignees: Cosmo Institute of Industrial Intelligence (Qingdao) Co., Ltd.; COSMOPlat IoT Technology, Co., Ltd
- Applicants: Cosmo Institute of Industrial Intelligence (Qingdao) Co., Ltd.; COSMOPlat IoT Technology Co., Ltd.
- Inventors: Lucheng Chen; Deyan Liu; Chao Wang; Chenggang Qin; Lin Wang; Qingze Tian; Xiang Li
- Web page: https://patentgrep.com/patent/US12596857B2

## CPC

- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

The present application pertains to the field of data processing technology, specifically relates to a method, apparatus and device for optimizing a process parameter and a storage medium, which includes: using a first weight parameter of a pre-trained support vector regression model for a manufacturing equipment as an iterative initial value, and calculating a second weight parameter of a support vector regression model using an online proximal gradient algorithm based on training data; obtaining an optimized support vector regression model by updating the first weight parameter of the pre-trained support vector regression model to the second weight parameter; inputting a first process parameter of the manufacturing equipment into the optimized support vector regression model to obtain a first detection parameter output by the optimized support vector regression model; and calculating, according to the first process parameter and the first detection parameter, a target process parameter for the manufacturing equipment.

## Claims

### Claim 1

A method for optimizing a process parameter, applied to a server, comprising:
using, by a hardware processor of the server, a first weight parameter of a pre-trained support vector regression model for a manufacturing equipment as an iterative initial value, and calculating a second weight parameter of a support vector regression model using an online proximal gradient algorithm based on training data, wherein the training data comprises: a historical process parameter of the manufacturing equipment, and a historical detection parameter corresponding to when the manufacturing equipment operates with the historical process parameter;
obtaining, by the hardware processor, an optimized support vector regression model by updating the first weight parameter of the pre-trained support vector regression model to the second weight parameter;
inputting, by the hardware processor, a first process parameter of the manufacturing equipment into the optimized support vector regression model to obtain a first detection parameter output by the optimized support vector regression model; and
calculating, by the hardware processor, according to the first process parameter and the first detection parameter, a target process parameter for the manufacturing equipment;
wherein the using, by the hardware processor, the first weight parameter of the pre-trained support vector regression model for the manufacturing equipment as the iterative initial value, and calculating the second weight parameter of the support vector regression model using the online proximal gradient algorithm based on the training data comprises:
calculating, by the hardware processor, the second weight parameter by using a linear iterative formula with the first weight parameter as the iterative initial value when the pre-trained support vector regression model is a linear support vector regression model, and the linear iterative formula is:
w
k
+
1
=
w
k
-
βλ
⁢
t
k
⁢
ϕ
⁡
(
x
k
)
⁢
sign
⁡
(
w
k
T
⁢
ϕ
⁡
(
x
k
)
-
y
k
)
⁢
max
⁡
(
0
,
❘
"\[LeftBracketingBar]"
w
k
T
⁢
ϕ
⁡
(
x
k
)
-
y
k
❘
"\[RightBracketingBar]"
-
ε
)
β
-
1
1
+
t
k
;
or
calculating, by the hardware processor, the second weight parameter by using a nonlinear iterative formula with the first weight parameter as the iterative initial value when the pre-trained support vector regression model is a nonlinear support vector regression model, and the nonlinear iterative formula is:
w
k
+
1
=
w
k
-
βλ
⁢
t
k
⁢
x
k
⁢
sign
⁡
(
w
k
T
⁢
x
k
-
y
k
)
⁢
max
⁡
(
0
,
❘
"\[LeftBracketingBar]"
w
k
T
⁢
x
k
-
y
k
❘
"\[RightBracketingBar]"
-
ε
)
β
-
1
1
+
t
k
;
wherein wk+1 represents a (k+1)-th second weight parameter, wk represents a k-th second weight parameter, an initial second weight parameter w0 is the first weight parameter, tk represents a step size for a k-th iteration, β represents an order of a penalty function, λ represents a penalty factor of the penalty function, xk represents the historical process parameter, yk represents the historical detection parameter corresponding to xk, ε represents an insensitive coefficient of the support vector regression model, and φ(xk) represents a kernel mapping function.

### Claim 2 (depends on 1)

The method according to claim 1, wherein calculating, by the hardware processor, according to the first process parameter and the first detection parameter, the target process parameter for the manufacturing equipment comprises:
calculating, by the hardware processor, according to the first process parameter and the first detection parameter, the target process parameter by using a swarm intelligence algorithm.

### Claim 3 (depends on 2)

The method according to claim 2, wherein calculating, by the hardware processor, according to the first process parameter and the first detection parameter, the target process parameter by using the swarm intelligence algorithm comprises:
calculating, by the hardware processor, according to the first process parameter and the first detection parameter, with a historical optimal process parameter of the manufacturing equipment as an initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm.

### Claim 4 (depends on 3)

The method according to claim 3, wherein calculating, by the hardware processor, according to the first process parameter and the first detection parameter, with the historical optimal process parameter of the manufacturing equipment as the initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm comprises:
determining, by the hardware processor, according to a target optimization type of the first detection parameter, a target historical optimal process parameter corresponding to the target optimization type in the historical optimal process parameter of the manufacturing equipment; and
calculating, by the hardware processor, according to the first process parameter and the first detection parameter, with the target historical optimal process parameter as the initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm.

### Claim 5 (depends on 4)

The method according to claim 4, wherein calculating, by the hardware processor, according to the first process parameter and the first detection parameter, with the target historical optimal process parameter as the initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm comprises:
calculating, by the hardware processor, according to the first process parameter and the first detection parameter, with the target historical optimal process parameter as an initial position parameter for a particle swarm optimization algorithm, the target process parameter through the particle swarm optimization algorithm.

### Claim 6

A device for optimizing a process parameter, comprising:
a processor and a memory;
the memory is configured to store a computer program; and
the processor is configured to execute the computer program stored in the memory to:
use a first weight parameter of a pre-trained support vector regression model for a manufacturing equipment as an iterative initial value, and calculate a second weight parameter of a support vector regression model using an online proximal gradient algorithm based on training data, wherein the training data comprises: a historical process parameter of the manufacturing equipment, and a historical detection parameter corresponding to when the manufacturing equipment operates with the historical process parameter;
obtain an optimized support vector regression model by updating the first weight parameter of the pre-trained support vector regression model to the second weight parameter;
input a first process parameter of the manufacturing equipment into the optimized support vector regression model to obtain a first detection parameter output by the optimized support vector regression model; and
calculate, according to the first process parameter and the first detection parameter, a target process parameter for the manufacturing equipment;
wherein the processor is further configured to:
calculate the second weight parameter by using a linear iterative formula with the first weight parameter as the iterative initial value when the pre-trained support vector regression model is a linear support vector regression model, and the linear iterative formula is:
w
k
+
1
=
w
k
-
βλ
⁢
t
k
⁢
ϕ
⁡
(
x
k
)
⁢
sign
⁡
(
w
k
T
⁢
ϕ
⁡
(
x
k
)
-
y
k
)
⁢
max
⁡
(
0
,
❘
"\[LeftBracketingBar]"
w
k
T
⁢
ϕ
⁡
(
x
k
)
-
y
k
❘
"\[RightBracketingBar]"
-
ε
)
β
-
1
1
+
t
k
;
or
calculate the second weight parameter by using a nonlinear iterative formula with the first weight parameter as the iterative initial value when the pre-trained support vector regression model is a nonlinear support vector regression model, and the nonlinear iterative formula is:
w
k
+
1
=
w
k
-
βλ
⁢
t
k
⁢
x
k
⁢
sign
⁡
(
w
k
T
⁢
x
k
-
y
k
)
⁢
max
⁡
(
0
,
❘
"\[LeftBracketingBar]"
w
k
T
⁢
x
k
-
y
k
❘
"\[RightBracketingBar]"
-
ε
)
β
-
1
1
+
t
k
;
wherein wk+1 represents a (k+1)-th second weight parameter, wk represents a k-th second weight parameter, an initial second weight parameter w0 is the first weight parameter, tk represents a step size for a k-th iteration, β represents an order of a penalty function, λ represents a penalty factor of the penalty function, xk represents the historical process parameter, yk represents the historical detection parameter corresponding to xk, ε represents an insensitive coefficient of the support vector regression model, and φ(xk) represents a kernel mapping function.

### Claim 7 (depends on 6)

The device according to claim 6, wherein the processor is further configured to:
calculate, according to the first process parameter and the first detection parameter, the target process parameter by using a swarm intelligence algorithm.

### Claim 8 (depends on 7)

The device according to claim 7, the processor is further configured to:
calculate, according to the first process parameter and the first detection parameter, with a historical optimal process parameter of the manufacturing equipment as an initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm.

### Claim 9 (depends on 8)

The device according to claim 8, wherein the processor is further configured to:
determine, according to a target optimization type of the first detection parameter, a target historical optimal process parameter corresponding to the target optimization type in the historical optimal process parameter of the manufacturing equipment; and
calculate, according to the first process parameter and the first detection parameter, with the target historical optimal process parameter as the initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm.

### Claim 10 (depends on 9)

The device according to claim 9, wherein the processor is further configured to:
calculate, according to the first process parameter and the first detection parameter, with the target historical optimal process parameter as an initial position parameter for a particle swarm optimization algorithm, the target process parameter through the particle swarm optimization algorithm.

### Claim 11

A non-transitory readable storage medium, wherein the readable storage medium stores a computer program; the computer program is used for implementing the following steps:
using a first weight parameter of a pre-trained support vector regression model for a manufacturing equipment as an iterative initial value, and calculating a second weight parameter of a support vector regression model using an online proximal gradient algorithm based on training data, wherein the training data comprises: a historical process parameter of the manufacturing equipment, and a historical detection parameter corresponding to when the manufacturing equipment operates with the historical process parameter;
obtaining an optimized support vector regression model by updating the first weight parameter of the pre-trained support vector regression model to the second weight parameter;
inputting a first process parameter of the manufacturing equipment into the optimized support vector regression model to obtain a first detection parameter output by the optimized support vector regression model; and
calculating, according to the first process parameter and the first detection parameter, a target process parameter for the manufacturing equipment;
wherein using the first weight parameter of the pre-trained support vector regression model for the manufacturing equipment as the iterative initial value, and calculating the second weight parameter of the support vector regression model using the online proximal gradient algorithm based on the training data comprises:
calculating, by the hardware processor, the second weight parameter by using a linear iterative formula with the first weight parameter as the iterative initial value when the pre-trained support vector regression model is a linear support vector regression model, and the linear iterative formula is:
w
k
+
1
=
w
k
-
βλ
⁢
t
k
⁢
ϕ
⁡
(
x
k
)
⁢
sign
⁡
(
w
k
T
⁢
ϕ
⁡
(
x
k
)
-
y
k
)
⁢
max
⁡
(
0
,
❘
"\[LeftBracketingBar]"
w
k
T
⁢
ϕ
⁡
(
x
k
)
-
y
k
❘
"\[RightBracketingBar]"
-
ε
)
β
-
1
1
+
t
k
;
or
calculating, by the hardware processor, the second weight parameter by using a nonlinear iterative formula with the first weight parameter as the iterative initial value when the pre-trained support vector regression model is a nonlinear support vector regression model, and the nonlinear iterative formula is:
w
k
+
1
=
w
k
-
β
⁢
λ
⁢
t
k
⁢
x
k
⁢
sign
⁢
(
w
k
T
⁢
x
k
-
y
k
)
⁢
max
⁢
(
0
,
❘
"\[LeftBracketingBar]"
w
k
T
⁢
x
k
-
y
k
❘
"\[RightBracketingBar]"
-
ε
)
β
-
1
1
+
t
k
;
wherein wk+1 represents a (k+1)-th second weight parameter, wk represents a k-th second weight parameter, an initial second weight parameter w0 is the first weight parameter, tk represents a step size for a k-th iteration, β represents an order of a penalty function, λ represents a penalty factor of the penalty function, xk represents the historical process parameter, yk represents the historical detection parameter corresponding to xk, ε represents an insensitive coefficient of the support vector regression model, and φ(xk) represents a kernel mapping function.

### Claim 12 (depends on 11)

The non-transitory readable storage medium according to claim 11, wherein the computer program is further used for implementing the following step:
calculating, according to the first process parameter and the first detection parameter, the target process parameter by using a swarm intelligence algorithm.

### Claim 13 (depends on 12)

The non-transitory readable storage medium according to claim 12, wherein the computer program is further used for implementing the following step:
calculating, according to the first process parameter and the first detection parameter, with a historical optimal process parameter of the manufacturing equipment as an initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm.

### Claim 14 (depends on 13)

The non-transitory readable storage medium according to claim 13, wherein the computer program is further used for implementing the following steps:
determining, according to a target optimization type of the first detection parameter, a target historical optimal process parameter corresponding to the target optimization type in the historical optimal process parameter of the manufacturing equipment; and
calculating, according to the first process parameter and the first detection parameter, with the target historical optimal process parameter as the initial position parameter for the swarm intelligence algorithm, the target process parameter through the swarm intelligence algorithm.

### Claim 15 (depends on 1)

The method according to claim 1, wherein before using, by the hardware processor of the server, the first weight parameter of the pre-trained support vector regression model for a manufacturing equipment as the iterative initial value, and calculating the second weight parameter of the support vector regression model using the online proximal gradient algorithm based on the training data, the method further comprises:
performing, by the hardware processor, data screening, feature selection, and feature normalization on the training data.

### Claim 16 (depends on 6)

The device according to claim 6, wherein the processor is further configured to:
perform data screening, feature selection, and feature normalization on the training data before using the first weight parameter of the pre-trained support vector regression model for the manufacturing equipment as the iterative initial value, and calculating the second weight parameter of the support vector regression model using the online proximal gradient algorithm based on the training data.

### Claim 17 (depends on 11)

The non-transitory readable storage medium according to claim 11, wherein before using the first weight parameter of the pre-trained support vector regression model for the manufacturing equipment as the iterative initial value, and calculating the second weight parameter of the support vector regression model using the online proximal gradient algorithm based on the training data, the computer program is further used for implementing the following step:
performing data screening, feature selection, and feature normalization on the training data.

### Claim 18 (depends on 1)

The method according to claim 1, wherein the method further comprises:
enabling, by the hardware processor, the manufacturing equipment to produce according to the target process parameter.

### Claim 19 (depends on 6)

The device according to claim 6, wherein the processor is further configured to:
enable the manufacturing equipment to produce according to the target process parameter.

### Claim 20 (depends on 11)

The non-transitory readable storage medium according to claim 11, the computer program is further used for implementing the following step:
enabling the manufacturing equipment to produce according to the target process parameter.