Academic Journal of Science and Technology 
ISSN: 2771-3032 | Vol. 7, No. 2, 2023 

 

166 

Gas	Well	Annulus	Pressure	Time	Sequence	Predictive	
Algorithm	Research	
Miaomiao Chen, Man Pu, Xin Yi, Jialong Xie, Ziqi Xi, Xueyi Wu, Xiaohang Hua, Yuxin Mao 

Southwest Petroleum University, Chengdu, 610500, China 

 

Abstract: The abnormality of gas well annulus pressure is one of the main risks that threaten the safety of gas wells and affect their 
production efficiency. In order to further improve the level of Gas Well Annulus Pressure management, this study introduced the time sequence 
prediction method. Through the design of multiple variable gray prediction algorithms and neural network prediction algorithm design, the 
effectiveness of the model was verified by the comparison of the prediction results and the actual measurement data. The research results verify 
the feasibility of the time sequence predictive algorithm on the dynamic prediction of the gas well annulus pressure, which provides theoretical 
support for the early diagnosis and active prevention of the gas well annulus pressure. 

Keywords: Gas Well Annulus Pressure, Multi -variable gray prediction, Neural network prediction, Production safety. 
 

1. Introduce 
As my country has accelerated the development of oil and 

natural gas, the phenomenon of gas in the gas well is 
becoming more and more prominent. In order to reduce the 
impact of gas wells on production on production, there must 
be predictive research on gas wells[1].The time sequence 
prediction method is a statistical method that reveals the 
system's dynamic structure and laws based on dynamic data. 
At present, several common time sequence predictions are: 
regression analysis prediction method, gray system prediction 
method, and artificial neural network prediction method. The 
risk -free gas well that has been judged to be risky must be 
pre -judged as much as possible, and the time sequence 
prediction method is introduced from the perspective of 
informatization[2]. 

In short, according to the characteristics of the gas well 
circular system, according to the existing knowledge, the time 
sequence is used to predict the circular pressure. Early 
diagnosis and active prevention of abnormal circular pressure, 
providing new research ideas and methods for the 
management of gas well annulus pressure[3]. 

2. Gray System Prediction 
The gray system prediction is a study of information 

modeling. It combines the random process and the gray 
process, and uses the statistical law of the gray amount to 
generate data generation The number of new information 
priority is one of the basic principles of the theory of the gray 
system[4]. It emphasizes the importance of new information 
in related analysis, which means that the new information is 
more influencing the current operation of the system than the 
old information. "It continuously adds new information and 
deletes irrelevant outdated information, which can 
dynamically reflect the latest development trend of the 
system[5]. Based on the principle of gray -associated analysis 
and new information priority, a gas well circular pressure 
multi -variable gray predictive algorithm with metabolic 
functions with metabolic functions is designed. Obtain the 
latest measured data over time, update the modeling time 
sequence, dynamically predict the gas well circular pressure 
pressure The change trend, the calculation process is shown 
in Figure 1: 

Start

Obtain the sequence values of annular 
pressure and related factors from time 
t=i to time t=j.

Set time parametersi,j,k.
(i<j<k,j-i>3,k-j>1)

Calculate the gray correlation degree ρ 
of the annular pressure and the sequence 
values of related factors from time t=i to 
time t=j.

ρ≥0.5？

c

Establish multivariate grey prediction 
model.

Calculate the predicted value of annular 
pressure from time t=j+1 to time t=k.

End to predict？

i+1,j+1,k+1

End

No

Yes

Yes

 
Figure 1. Gas well annulus pressure,multi -variable gray 

predictive algorithm flowchart 

3. Neural Network Prediction 
Algorithm Design 

Neural network is a mathematical model established by the 
micro structure and functions of the human nervous system[6]. 
It is an important way to simulate human intelligence. It is 
generally composed of input layers, hidden layers, and output 
layers. According to the structure and basic functions of 
biological neurons, the expression is: 

 

,
1

( )
n

w b i i
i

h x f w x b


 
  

 
                (1) 

 
First, build a three -layer neural network. The cyclopenic 



 

167 

pressure sensitivity factors analyzed as gray as a related 
variable input layer are used as the output layer as the output 
layer. The time period of a large adjustment of the output of 
gas well at a certain period of time is used as the training set, 
and the time period of the production of large adjustment in 
another period of time is used as the test set. The setting of 
related parameters is as follows: 

(1) Three factors of oil pressure, temperature, and instant 
yield are used as the input layer of the neural network, and the 
historical data of the well -empty pressure of the well as the 

output layer of the neurons, and 10 neurons are selected as 
hidden function test nodes to form 3 -10-1 neural network; 

(2) Set the learning rate to 0.01, and the number of network 
iterations is set to 1000; 

(3) Set the transmission function of the hidden layer and 
the transport layer and the network training function. 

The neural network structure adopted in this section is 3-
10-1, and the schematic diagram of artificial neural network 
training is shown in Figure 2. 

 

Initialize ownership coefficient, 
learning rate,and number of iterations

Normalization of training set sample 
values

Input samples and provide regular output

Calculate the coefficients of each node 
in the hidden layer

Calculate the coefficients and output 
errors of each node in the output layer

Correction and update of hidden layer 
and output layer coefficients

Does the error meet the 
design requirements?

Substitute all samples?

Output trained neural network structure

Iterations
+1

是

是

NO

否

Begin

end

 
Figure 2. Gas well annulus pressure neural network prediction algorithm flow chart 

 

4. Example Analysis 

4.1. Example Analysis of Gray System 
Prediction Algorithm 

First of all, the GM (1, 1) model was established to 
establish the GM (1, 1) model at 00: 00 to 06: 00, and the GM 
(1, 1) model was established, and a precise test was 
performed[7]. Get 00: 00-06: 00 The oil pressure and the 

temperature gray GM (1,1) prediction of the well-exit 
temperature in the period of the time period of the average 
relative percentage error accuracy level reaches the level, and 
the gas production gray GM (1,1) forecast an average 
relatively percentage percentage percentage percentage The 
error accuracy level reaches secondary, so it can be used for 
short -term prediction[8]. Through the predicted value of each 
factors from the establishment of the gray GM (1,1) model at 
07: 00-11: 00, the calculation result is shown in Table 1: 

 
Table 1. Each relevant factors 07: 00-11: 00 Forecast values 

Time Oil pressure(MPa) temperature(℃) Instant yield(m3) 
07:00 55.674 33.894 202095.417 
08:00 55.178 34.182 204973.519 
09:00 54.688 34.471 207892.608 
10:00 54.201 34.764 210853.269 
11:00 53.719 35.058 213856.094 



 

168 

 
Establish a multi -variable gray prediction model for the A 

ring pressure sequence and related factors. Among them, the 
data of 00: 00-06: 00 is used as the basis for the minimum of 
the model parameter. In the restoration of the time response, 

calculate the prediction value of the pressure of the circular 
wells at the time of the circular port at a time, and the 
comparison of the circular pressure prediction results of the 
07: 00-11: 00 is as shown in Table 2. 
 

Table 2. A ring gas pressure 07: 00-11: 00 prediction results 

Time 
gas well annulus pressure prediction 

value (MPa) 
gas well annulus pressure actual 

value (MPa) 
error(%) 

07:00 40.429 40.401 0.07% 
08:00 40.634 40.565 0.17% 
09:00 40.837 40.723 0.28% 
10:00 41.039 40.853 0.46% 
11:00 41.241 40.975 0.65% 

 

4.2. Analysis of Neural Network Prediction 
Algorithms 

Selected from 12:00 to 16:00 on October 2020 (10 × 104m3) 
at 12:00-16: 00 (about 10× 104m3), the oil pressure, well 

estate temperature, instant yield, and A-ring gas pressure 
actual measurement value (measured by hour) for training set, 
Each parameter is as shown in Table 3. It iterative calculation 
is 1,000 times, and the accuracy of the neural network model 
reaches 0.99625. 

 
Table 3. Parameter Training Collection of Each Parameters 

Time 
Oil pressure 

(MPa) 
The well mouth temperature 

(℃) 
Instant output 

(m3) 
gas well annulus pressure 

(MPa) 
2020.10.20 12:00 58.846 51.608 356240.619 60.023 
2020.10.20 13:00 58.811 53.103 356542.987 60.025 
2020.10.20 14:00 58.780 52.914 352368.220 60.023 
2020.10.20 15:00 58.801 53.750 347558.443 60.026 
2020.10.20 16:00 58.818 51.774 340825.167 60.019 

 

Use the above parameters of the above parameter to verify 
the effect of the neural network prediction effect at 0: 00-6: 
00 on November 29, 2020 (about 5 × 104m3). As shown in 

Table 4, the error meets the prediction accuracy requirements 
and verified the artificial neural network. The validity of the 
circular pressure prediction when the yield is adjusted. 

 

Table 4. Check piece test set 

Time 
Instant output (m3) gas well annulus pressure 

prediction (MPa) 
gas well annulus pressure 

measurement (MPa) 
error(%) 

2020.11.29 0:00 257295.463 60.253 60.192 0.10% 
2020.11.29 1:00 251931.842 60.251 60.195 0.09% 
2020.11.29 2:00 251737.679 60.251 60.205 0.08% 
2020.11.29 3:00 250534.258 60.249 60.206 0.07% 
2020.11.29 4:00 252978.626 60.251 60.214 0.06% 
2020.11.29 5:00 248761.070 60.244 60.216 0.05% 

 

The accuracy of the prediction model of the neural network 
adjustment is shown in Figure 3. 

 

 
Figure 3. Neural network prediction model accuracy 

schematic diagram 

5. Result 
(1) For the daily management of known gas well annulus 

pressure, the time sequence prediction method is used to 
determine the future change trend. Based on the theory of 
multiple variable gray system theory, the gas wells that 
consider multiple correlation and specific metabolic functions 
have been studied. Pressing dynamic prediction methods. 
Taking a gas well in northwestern Sichuan as an example, 
research believes that oil pressure, well mouth temperature 
and instant yield are the sensitive factor of changing the 
changes in the gas in the gas well. The system theory is used 
for the effectiveness of the dynamic prediction of pressure 
dynamic prediction. 

(2) Aiming at the forecast of the gas well annulus pressure 
during the production of the gas well, artificial neural network 
prediction experiments are carried out. The wells are used as 



 

169 

a related variable input layer with the temperature, oil 
pressure, and instantaneous output of the well. Taking a 
certain gas well annulus pressure in northwestern Sichuan as 
an example, the time period with a large adjustment of the 
output in the test gas well in a certain period of time is used 
as the training set, and the time period of the output of the 
output in the other time is used as a test set. The effectiveness 
of artificial neural networks is used for the prediction of 
circular pressure prediction when the gas well annulus 
pressure is adjusted. 

Acknowledgment 
The study is support by 2022 Innovation and 

Entrepreneurship Training for College Students in Sichuan 
Province (S202210615216) 

References 
[1] Wilkins R P,Free D. A new approach to the prediction of gas 

flow after cementing[J]. SPE/IADC18622, 1989. 

[2] Duan S , Wojtanowicz A . A Method for Evaluation of Risk of 
Continuous Air Emissions From Sustained Casinghead 
Pressure[C].SPE/EPA/DOE Exploration and Production 
Environmental Conference,2005. 

[3] Jackson, P.B..Effect of Casing Pressure on Gas Flow Through 
a Sheath of Set Cement[A].SPE/IADC Drilling 
Conference[C],1993. 

[4] David W.Microannulus leaks repaired with pressure-activated 
sealant[J].SPE91399,2004. 

[5] Ravi Kris,Darbe Robert.Auto-sealing cement sheath prevents 
sustained casing pressure.[J].Offshore,2007,Vol.67(9): 92-94. 

[6] P.Oudeman,M.Kerem.Transient behavior of annular pressure 
build-up in HP/HT wel1s[J].SPE88735,2004. 

[7] Nygaard Runar,Salehi Saeed,Weideman Benjamin.Effect of 
Dynamic Loading on Wellbore Leakage for the Wabamun 
Area CO2-Sequestration Project[J].JOURNAL OF 
CANADIAN PETROLEUM TECHNOLOGY,2014,Vol.53(1). 

[8] Jonathan Bellarby.Annular pressure build-up analysis and 
methodology with examples from multifrac horizontal wells 
and HPHT reservoirs[A].SPE/IADC Drilling Conference and 
Exhibition 2013[C],2013.