US 11,768,976 B2 · Subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network

Abstract

The present disclosure belongs to the field of petroleum engineering, and specifically relates to a subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network. The subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network includes three steps: digital twin model establishment, degradation process re-prediction model establishment, and remaining useful life calculation model establishment. The subsea Christmas tree re-prediction system integrating Kalman filter and Bayesian network includes a subsea distribution unit information acquisition subsystem mounted on an subsea distribution unit, a subsea control module information acquisition subsystem mounted on a subsea control module, a subsea valve bank information acquisition subsystem mounted on a subsea valve bank, a wellhead mechanical module information acquisition subsystem mounted on a wellhead mechanical module, a subsea environmental information acquisition module mounted on a subsea control module, and a subsea Christmas tree digital twin subsystem mounted in an overwater control station.

A subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network, comprising three steps:

digital twin model establishment,
degradation process re-prediction model establishment, and
remaining useful life calculation model establishment,
wherein the specific steps for digital twin model establishment are as follows:
for a physical structure of a subsea Christmas tree, establishing a digital twin geometric size model, wherein the digital twin geometric size model comprises an electronic structure geometric size model, a hydraulic structure geometric size model, a mechanical structure geometric size model, and the three structure geometric size models reflect a geometric size and an assembly relationship of a physical system;
for a marine environment of the subsea Christmas tree, establishing a digital twin production environment model, wherein the digital twin production environment model comprises real-time dynamic data composed of marine environmental monitoring data comprising typhoon, internal wave current, sea water temperature and pressure;
for a process parameter of the subsea Christmas tree, establishing a digital twin production process model, wherein the digital twin production process model comprises oil and gas production process data comprising conventional oil recovery, chemical injection, and paraffin removal;
for a monitoring parameter of the subsea Christmas tree, establishing a digital twin production state model, wherein the digital twin production state model comprises multi-source sensor system state data of a mechanical structure, a hydraulic structure, and an electronic structure;
the specific steps for degradation process re-prediction model establishment are as follows:
reading system state data of the subsea Christmas tree, calculating degradation amount of each assembly over time, wherein voltage information of the electronic structure, pressure of the hydraulic structure, flow information and stress-strain information of the mechanical structure of a historical process of the subsea Christmas tree are read, and historical degradation amount is determined by using a subsea Christmas tree failure mode;
estimating a Wiener process parameter based on degradation data, wherein for the subsea Christmas tree, a degradation model of each structure conforms to a Wiener process:
X(t)=X(0)+λt+σBB(t),
where λ is a drift coefficient, σB is a diffusion coefficient, and B(t) is a standard Brownian motion, and t is a sampling time and n represents a number of sets of degradation data and i represents a number of monitoring points in each set of degradation data;
for n sets of degradation data, each set of degradation data has i monitoring points, the degradation amount is recorded as X, the time is recorded as T, and the Wiener process parameter is estimated by using a maximum likelihood estimation method:
ln
⁢
L
⁡
(
Θ
|
X
)
=
-
1
2
⁢
ln
⁢
2
⁢
π
⁢
ni
-
1
2
⁢
ln
⁢
σ
B
2
⁢
n
⁢
i
-
1
2
⁢
∑
n
=
1
i
ln
⁢
❘
"\[LeftBracketingBar]"
Φ
n
❘
"\[RightBracketingBar]"
-
1
2
⁢
σ
B
2
⁢
∑
n
=
1
i
(
X
n
-
λ
n
⁢
X
n
)
′
⁢
❘
"\[LeftBracketingBar]"
Φ
n
❘
"\[RightBracketingBar]"
-
1
⁢
(
X
n
-
λ
n
⁢
X
n
)
,
where Φn=diag(ΔTn,t1, ΔTn,t2, . . . , ΔTn,ti);
and
the obtained Wiener process parameter is substituted into the Wiener process to obtain the degradation amount of the electronic structure, the hydraulic structure and the mechanical structure following the Wiener process respectively;
establishing an initial degradation prediction model for the subsea Christmas tree, and initially predicting degradation of the subsea Christmas tree in the established digital twin model by using Kalman filter, wherein the Kalman filter realizes optimal selection of a state estimation value and a state measurement value to obtain an optimal predicted value, and the Kalman filter comprises five steps:
{circle around (1)} state prediction:
{circumflex over (x)}t−=Ft{circumflex over (x)}t-1+Btut,
where Ft is a state transition matrix, Bt is a control matrix, representing the influence of a control variable ut representing a driven input vector on a current moment, {circumflex over (x)} represents an estimated value of x, and {circumflex over (x)}t− represents an estimated value of the current moment based on the state of a previous moment;
{circle around (2)} covariance matrix:
Pt−=FPt-1FT+Q,
where P is a covariance matrix and Q is the noise of the prediction model;
{circle around (3)} observed value:
zt=Hxt+ξ,
where zt represents an observed value, H represents an observed matrix, and is observed noise;
{circle around (4)} state update:
{circumflex over (x)}t={circumflex over (x)}t−+Kt(zt−H{circumflex over (x)}t−),
where Kt is a Kalman coefficient, which is expressed as:
Kt=Pt−HT(HPt−HT+R)−1,
where R is a covariance matrix of the observed noise;
{circle around (5)} covariance matrix update:
Pt=(I−KtH)Pt−,
wherein I represents an identity matrix;
to use Kalman filter to predict the state of the Wiener process, the Wiener process needs to be transformed into a state-space equation:
{
x
t
=
x
t
-
1
+
λ
t
-
1
⁢
t
+
σ
B
⁢
B
⁡
(
0
,
Δ
⁢
t
)
y
t
=
x
t
+
ξ
,
in the Wiener process, λt=λt-1 is set, then the constructed Kalman filter is:
{
[
x
t
λ
t
]
=
[
1
Δ
⁢
t
0
1
]
[
x
t
-
1
λ
t
-
1
]
+
[
σ
B
2
⁢
Δ
⁢
t
0
]
y
t
=
[
1
0
]
[
x
t
-
1
λ
t
-
1
]
+
ξ
,
the noise Q of the prediction model is expressed as
Q
=
[
σ
B
2
⁢
Δ
⁢
t
0
0
0
]
,
an initial state is expressed as:
X
0
=
[
0
μ
t
o
]
,
an initial covariance matrix is expressed as:
P
0
=
[
0
0
0
σ
t
0
2
]
,
with an Kalman filter algorithm constructed by using the Wiener process, combined with given initial conditions, the initial prediction of the degradation amount is realized through optimal estimation for the state measurement value and the state measurement value by Kalman filter;
establishing a degradation re-prediction model for the subsea Christmas tree, and establishing the Wiener process as a dynamic Bayesian network in the digital twin model by using the Bayesian network, wherein first, main parameters μλ, σλ2, σB2 and X of the Wiener process are nodes of a static Bayesian network, arrows between the nodes indicate a physical relationship between the nodes, after time expansion, a dynamic Bayesian network of t time slices is obtained, E represents an evidence node, and supplementary information in a physical entity is input to the dynamic Bayesian network to realize information update of the dynamic Bayesian network;
a calculation result of the dynamic Bayesian network after the information update is input as a new state measured value into the Kalman filter algorithm established in S203 for secondary calculation of the degradation amount;
the secondary calculation performs secondary optimization prediction based on an initial state estimated value and credibility of a measured value calculated by the dynamic Bayesian network after information supplementation, that is, the re-prediction of the degradation amount, which is expressed as:
{circumflex over (x)}t={circumflex over (x)}t−+Kt(zDTs−H{circumflex over (x)}t−),
where zDTs represents a new measured value calculated by the dynamic Bayesian network;
and
finally, a ratio of the degradation amount of the electronic structure, the hydraulic structure, and the mechanical structure to overall performance of the electronic structure, the hydraulic structure, and the mechanical structure is defined as the degradation rate of the electronic structure, the hydraulic structure, and the mechanical structure, then an overall degradation rate of the electronic structure containing n assemblies is:
ye=1−(1−ye1)·(1−ye2)⋅ . . . ⋅(1−yen),
an overall degradation rate of the hydraulic structure containing n assemblies is:
yh=1−(1−yh1)·(1−yh2)⋅ . . . ⋅(1−yhn),
an overall degradation rate of the mechanical structure containing n assemblies is:
ym=1−(1−ym1)·(1−ym2)⋅ . . . ⋅(1−ymn), and
then an overall degradation rate of the system is:
y=1−(1−ye)·(1−yh)·(1−ym);
wherein e represents an initial of the electronic structure, h represents an initial of the hydraulic structure, m represents an initial of the mechanical structure;
and
updating physical entity information of the subsea Christmas tree to the digital twin model, and inputting environmental information and working state information as supplementary information into the evidence node of the dynamic Bayesian network model of the digital twin model to realize real-time degradation amount calculation of the digital twin model, wherein the degradation amount re-predicted by the digital twin model is fed back to the physical entity of the subsea Christmas tree at the same time, and the Wiener process parameter is continuously corrected through comparison with actual degradation amount to realize the correction of the digital twin model;
the specific steps for remaining useful life calculation model establishment are as follows:
calculating optimized remaining useful life based on the degradation data re-predicted by the digital twin model and a failure threshold, and determining a system failure threshold according to system failure data, wherein a moment when the degradation amount reaches the failure threshold is referred to as failure time, and the remaining useful life is defined as a time period from detection time to the failure time;
calculating a probability distribution model of the remaining useful life of the subsea Christmas tree over time, wherein according to the characteristics of the degradation of the Wiener process, the probability distribution of the remaining useful life is defined as:
f
⁡
(
t
)
=
(
x
t
⁢
h
-
x
t
)
2
2
⁢
π
⁢
σ
B
2
⁢
t
3
⁢
exp
⁢
(
-
(
x
t
⁢
h
-
x
t
-
λ
⁢
t
)
2
2
⁢
σ
B
2
⁢
t
)
,
where xth is the failure threshold;
and
feeding life information back to the digital twin model to realize information update of the digital twin model;
and
the subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network being applied to a subsea Christmas tree re-prediction system integrating Kalman filter and Bayesian network, wherein the subsea Christmas tree re-prediction system comprises a subsea distribution unit information acquisition subsystem mounted, a subsea control module information acquisition subsystem, a subsea valve bank information acquisition subsystem, a wellhead mechanical module information acquisition subsystem, a subsea environmental information acquisition module, and a subsea Christmas tree digital twin subsystem,
the subsea distribution unit information acquisition subsystem comprises a subsea distribution unit hydraulic information acquisition module, a subsea distribution unit degradation rate calculation module, and a subsea distribution unit electronic information acquisition module;
the subsea control module information acquisition subsystem comprises a subsea control module hydraulic information acquisition module, a subsea control module degradation rate calculation module, and a subsea control module electronic information acquisition module;
the subsea valve bank information acquisition subsystem comprises a subsea valve bank hydraulic information acquisition module and a subsea valve bank degradation rate calculation module;
the wellhead mechanical module information acquisition subsystem comprises a wellhead mechanical module information acquisition module and a wellhead mechanical module degradation rate calculation module;
and
the subsea Christmas tree digital twin subsystem comprises a system degradation rate calculation module, an initial degradation prediction module, a degradation process re-prediction module, a remaining useful life calculation module, and an information update module.

350 more elements

Subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network

Life of this patent

Subsea Christmas tree re-prediction method integrating Kalman filter and Bayesian network

Life of this patent

Abstract

Claims7