# US 2026/0099656 A1 · Pikan-based seepage simulation method for heterogeneous reservoirs

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

- Number: US 2026/0099656 A1
- Document ID: US20260099656A1
- Status: Published application
- Kind: A1
- Application number: 19/322,667
- Filed: 2025-09-08
- Published: 2026-04-09
- Earliest claimed: 2024-10-21
- Assignees: Yangtze University; Western Research Institute of Yangtze University
- Applicants: Yangtze University; Western Research Institute of Yangtze University
- Inventors: Yuhui Zhou; Xiang Rao; Hui Zhao; Guanglong Sheng; Ruiqi Zong; Fankun Meng; Zongfa Li; Shaoyang Geng; Yunfeng Xu; Wei Liu; Wentao Zhan; Liang Yan; Penghua Luo; Yuchen Pei
- Web page: https://patentgrep.com/patent/US20260099656A1

## CPC

- G06F 30/28: using fluid dynamics, e.g. using Navier-Stokes equations or computational fluid dynamics [CFD]
- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06F 2113/08: Fluids
- Y02A 10/40: Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- Y02A: TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE

## Abstract

The present disclosure provides a PIKAN-based heterogeneous reservoir seepage simulation method, which belongs to the field of heterogeneous reservoir development technology, including the following steps: S1, using three independent KAN networks to approximate the PIKAN of pressure p, velocity component vx and velocity component vy respectively; S2, constructing a loss function of the mixed pressure-velocity formula including the residual of the mass conservation formula, the residual of Darcy's law and the residual of the relevant boundary conditions, and optimizing the loss function. The present disclosure adopts the above-mentioned PIKAN-based heterogeneous reservoir seepage simulation method, introduces KAN into the flow simulation of porous media for the first time, and provides a preliminary reference for the development of general seepage simulation tools based on PIKAN.

## Claims

### Claim 1

A heterogeneous reservoir seepage simulation method based on PIKAN, comprising the following steps:
S1, using three independent KAN networks to approximate a PIKAN of pressure p, velocity component vx and velocity component vy, respectively;
S2, constructing a loss function of a mixed pressure-velocity formula comprising a residual of a mass conservation formula, a residual of Darcy's law, and a residual of relevant boundary conditions, and optimizing the loss function.

### Claim 2 (depends on 1)

The heterogeneous reservoir seepage simulation method based on PIKAN according to claim 1, wherein in S1, three independent KAN networks are used to approximate the PIKAN of p, velocity component vx and velocity component vy, respectively, the specific operation is as follows:
for a smooth ƒ: [0,1]n→°, there is:
f
⁡
(
x
)
=
f
⁡
(
x
1
,
L
,
x
n
)
=
∑
q
=
1
2
⁢
n
+
1
Φ
q
(
∑
p
=
1
n
∅
q
,
p
(
x
p
)
)
(
1
)
where φq,p: [0,1]→°; Φq:°→°; x denotes a vector composed of n independent variables xi, i=1, L, n; ƒ(x) denotes a function of xi for n independent variables; xp denotes the p-th component of vector x; a univariate function φq,p processes the p-th component of vector x and contributes an item to a sum of the q-th external function, there are 2n+1 external functions, and each external function Φq is a one-variable function;
wherein a Chebyshev polynomial function is used to replace the B-spline function;
wherein, for a fully connected KAN with L+1 layers, the input layer is marked as the 0-th layer, nl denotes a count of neurons in the l-th layer, (l,j) denotes the l-th neurons in layer l, and
x
j
l
denotes an activation value of the neuron, there will be nlnl+1 parameterized univariate functions between the l-th layer and the l+1-th layer, where the univariate function connecting (l,j) and (l+1,i) is denoted as
T
i
,
j
l
,
and the activation value of neuron (l+1,i) is calculated as:
x
j
l
+
1
=
∑
i
=
1
n
l
T
i
,
j
l
⁢
(
x
i
l
)
(
2
)
where
⁢
x
j
l
+
1
denotes a value of the j-th neuron in the l+1-th layer,
x
i
l
value of the i-th neuron in the l-th layer; nl denotes a count of neurons in the l-th layer;
wherein, according to an activation value xl of the l+1-th layer of neurons, a matrix expression of the vector xl+1 composed of the activation values of the xl+1-th layer of neurons is as follows:
x
l
+
1
∘
=
T
l
⁢
x
l
(
3
)
where Tl denotes a function matrix corresponding to the l-th KAN layer,
T
i
,
j
l
is in the Tl-th row and the j-th column of Tl, and
T
i
,
j
l
is a Chebyshev polynomial function;
wherein all the learning parameters of the Chebyshev polynomial function
T
i
,
j
l
in KAN are recorded as θ, l=0, 1, L, L−1, j=1, 2, L, nl, i=1, 2, L, nl+1, it is obtained that when the data of the input layer is x0, a neuron value of the output layer of the fully connected KAN with L+1 layers is calculated as follows:
K
⁢
A
⁢
N
⁡
(
x
,
θ
)
=
(
T
L
-
1
∘
T
L
-
2
∘
L
∘
T
1
∘
T
0
)
⁢
x
0
(
4
)
where o denotes a composition between functions.

### Claim 3 (depends on 2)

The heterogeneous reservoir seepage simulation method based on PIKAN according to claim 2, wherein in S2, a loss function of the mixing pressure-velocity formula is:
L
⁡
(
θ
)
=
ω
1
⁢
L
PDE
1
+
ω
2
⁢
L
PDE
2
+
ω
3
⁢
L
DB
+
ω
4
⁢
L
NB
(
5
)
where L(θ) denotes a loss function; ω1, ω2, ω3, ω4 are all weighted coefficients; LPDE1 denotes a residual of a mass conservation formula; LPDE2 denotes a residual of Darcy's law; LDB denotes a residual of a Dirichlet boundary condition; LNB denotes a residual of a Neumann boundary condition.

### Claim 4 (depends on 3)

The heterogeneous reservoir seepage simulation method based on PIKAN according to claim 3, wherein the expression of the residual LPDE1 of the mass conservation formula is as follows:
L
PDE
1
=
1
N
in
⁢
∑
i
=
1
N
in
(
∂
v
x
∂
x
+
∂
v
y
∂
y
)
2
❘
(
x
i
,
y
i
)
(
6
)
where Nin denotes a count of collocation points within a computational domain; (xi,yi) denotes a coordinate of a collocation point; x, y and are spatial independent variables of the function.

### Claim 5 (depends on 3)

The heterogeneous reservoir seepage simulation method based on PIKAN according to claim 3, wherein the expression of the residual LPDE2 of Darcy's law is as follows:
L
PDE
2
=
1
N
in
⁢
∑
i
=
1
N
in
(
v
x
-
k
⁢
∂
p
∂
x
)
2
❘
(
x
i
,
y
i
)
+
1
N
i
⁢
n
⁢
∑
i
=
1
N
in
(
v
y
-
k
⁢
∂
p
∂
x
)
2
❘
(
x
i
,
y
i
)
(
7
)
where k denotes permeability.

### Claim 6 (depends on 3)

The heterogeneous reservoir seepage simulation method based on PIKAN according to claim 3, wherein the expression of the residual of the Dirichlet boundary condition is as follows:
L
DB
=
1
N
DB
⁢
∑
i
=
1
N
DB
(
p
-
g
)
2
❘
(
x
i
,
y
i
)
(
8
)
where NDB denotes a count of points on the Dirichlet boundary; p denotes a predicted value of PIKAN at collocation point (xi,yi); g denotes a Dirichlet boundary condition value.

### Claim 7 (depends on 3)

The heterogeneous reservoir seepage simulation method based on PIKAN according to claim 3, wherein the expression of the residual of the Neumann boundary condition is as follows:
L
NB
=
1
N
NB
⁢
∑
i
=
1
N
NB
(
v
x
-
h
x
)
2
❘
(
x
i
,
y
i
)
+
1
N
NB
⁢
∑
i
=
1
N
NB
(
v
y
-
h
y
)
2
❘
(
x
i
,
y
i
)
(
9
)
where NNB denotes a count of points on the Neumann boundary; hx denotes a x-direction velocity value given by the Neumann boundary condition; hy denotes a y-direction velocity value given by the Neumann boundary condition.
