Method for predicting and evaluating daily gas production rate of tight gas based on intelligent optimization algorithm and machine learning

Abstract

The invention relates to a method for predicting and evaluating the daily gas production rate of tight gas based on an intelligent optimization algorithm and machine learning, and belongs to the technical field of information and the field of in-well gas production. The problem that a traditional tight gas yield prediction method is low in precision and high in prediction difficulty due to complex data is solved. According to the technical scheme, the method comprises the steps that a tight gas yield normalizing process is established to collect the daily gas production rate, the accumulated gas production rate and tight gas well yield main control factors of all tight gas wells in the tight gas reservoir in the production cycle, and the accumulated gas production rate in the production cycle serves as a grading measurement index of a tight gas well yield grade evaluation and prediction model; taking the daily gas production rate in the production cycle as a label of a tight gas well daily gas production rate prediction model; establishing a tight gas yield grade evaluation and prediction model based on a K nearest neighbor algorithm, and evaluating and predicting the yield grade of the tight gas well; and establishing a tight gas well daily output prediction model based on a random forest algorithm optimized by an Archimedes optimization algorithm, and predicting the yield of the tight gas well.

Description

Technical Field

[0001] The invention relates to a method for predicting and evaluating daily gas production of tight gas based on intelligent optimization algorithm and machine learning, and belongs to the field of information technology and gas production from wells. Technical Background

[0002] As an important strategic resource, oil and gas are highly valued by all countries. Since the 1960s, my country's oil has been booming, but at present, the main old oil fields such as Daqing Oilfield, Changqing Oilfield and Shengli Oilfield have generally entered the late development stage of high water content or ultra-high water content. The remaining recoverable reserves are difficult to recover due to technical limitations, and the oil and gas production has decreased. In the long-term exploitation of major oil fields, a large amount of production data has been retained. By analyzing these data, it can be found that the recoverable reserves are directly related to factors such as geological parameters and drainage parameters. This is of great significance for more reasonable selection of high-reserve and easy-to-recover oil and gas fields and optimization of oil and gas field exploitation plans, thereby increasing oil and gas production. However, with the continuous increase in production data, the excessively large and complicated number makes data analysis more difficult, and traditional manual experience analysis methods are not enough to meet the needs of efficient development of oil and gas fields. Therefore, it has become an important task to choose a new, efficient and accurate method for data analysis to further better establish the development direction and support development decisions.

[0003] Tight gas is an unconventional natural gas resource with widely distributed reserves. The United States is the world's leading producer of tight gas, and Canada, China, Argentina and other countries have also made some progress in the exploration and production of tight gas. my country has carried out exploration, development and production of tight gas in the Ordos Basin and other places. The commercial development of tight gas has changed the pattern of the global natural gas market. The increase in tight gas production has led to an increase in natural gas supply, which has also had a certain impact on natural gas prices, and even affected the import and export patterns of natural gas in some regions. With the increasing contradiction between natural gas supply and demand, it is imperative to accelerate the development of tight gas. Accurate prediction and evaluation of gas well production capacity is of great significance for gas reservoir development. With the rapid increase in proven reserves and production of tight gas, how to improve the recovery rate of tight gas and improve the development level of tight gas reservoirs more reasonably and efficiently is a major research issue to be solved at present.

[0004] Intelligent optimization algorithm is a method that uses biological evolution, physical laws or swarm intelligence to conduct global search. It can be applied to solve challenges in the oil field, such as oil and gas production prediction and tight gas reservoir development. In the oil field, intelligent optimization algorithms can be used to optimize oil and gas field development plans, predict oil and gas production, and improve the recovery rate of tight gas reservoirs. By analyzing production data and geological parameters, combined with the global search capabilities of intelligent optimization algorithms, oil and gas resource reserves and geological conditions can be more accurately evaluated to guide the formulation of development decisions. In addition, these algorithms can also optimize drilling plans and improve oil and gas exploration efficiency, thereby promoting the sustainable development of the oil industry.

[0005] As an important technology in the field of machine learning, the advantage of the random forest algorithm is that it can process large-scale, high-dimensional data and improve the accuracy and stability of predictions by building multiple decision trees and combining their results. This makes it an ideal choice for solving complex problems in the oil and gas industry, bringing more efficient data analysis and decision support to the industry. The application of the random forest algorithm in the oil and gas industry is expected to solve challenges such as exploration and production optimization, improve the industry's intelligence level and competitiveness, and promote the industry to develop in a more intelligent, efficient and sustainable direction.

[0006] In general, in order to better establish the development direction for tight gas exploitation, support the formulation of development decisions, and thus improve the tight gas recovery rate, a reasonable prediction and evaluation method that is suitable for the complex characteristics of tight gas data sets and has high accuracy is needed. Summary of the invention

[0007] The purpose of the present invention is to solve the problem that there are few tight gas production prediction methods and insufficient accuracy at present. The present invention proposes a tight gas production prediction method based on intelligent optimization algorithm and machine learning. The technical solution of the present invention is as follows:

[0008] A method for predicting and evaluating daily gas production of tight gas based on intelligent optimization algorithm and machine learning, comprising the following steps:

[0009] S1: Establish a tight gas production regularization process to collect the daily gas production, cumulative gas production and main controlling factors of tight gas well production in the production cycle of each tight gas well in the tight gas reservoir, use the cumulative gas production in the production cycle as a grading measurement indicator for the tight gas well production grade evaluation and prediction model, and use the daily gas production in the production cycle as a label for the tight gas well daily gas production prediction model;

[0010] S2: Based on the K-nearest neighbor algorithm, a tight gas production grade evaluation and prediction model is established to evaluate and predict the production grade of tight gas wells;

[0011] S3: A daily gas production prediction model for tight gas wells is established based on the random forest algorithm optimized by the Archimedean optimization algorithm to predict the production of tight gas wells.

[0012] In the above step S1, the main controlling factors of the tight gas well production include geological factors, drainage parameters and construction parameters;

[0013] The above geological parameters include geological characteristic parameters around the production wells, such as porosity, permeability, gas saturation, reservoir thickness, reservoir depth, dynamic reserves, bottom hole temperature and bottom hole initial pressure;

[0014] The above-mentioned production parameters include initial production, average production in the first year, peak production, average production, maximum water production rate, average gas-water ratio, shut-in time, cumulative water production, bottom hole flowing pressure and daily water production;

[0015] The above-mentioned construction parameters include average sand ratio, construction pressure and fracturing fluid flowback rate;

[0016] In the above step S1, the daily gas production, cumulative gas production and main controlling factors of the tight gas well production in the tight gas reservoir during the production period are collected as a data set for subsequent prediction.

[0017] In the above data sets, different factors are normalized and unified in dimension to facilitate prediction;

[0018]

[0019] In the above dataset, interpolation method is used to complete the data with missing values;

[0020] In the above data set, noise data and outliers are cleared.

[0021] In the above step S2, the tight gas well production evaluation index is divided into three levels of high, medium and low according to the actual cumulative production, so as to improve the subsequent prediction accuracy.

[0022] In the above step S3, the optimization of the Archimedes optimization algorithm refers to using three strategies to insert the head, middle and tail of the original algorithm respectively to accelerate the convergence speed of the Archimedes optimization algorithm and improve its global search capability.

[0023] The steps of optimizing the Archimedean optimization algorithm are:

[0024] 1 Initialize the population, initialize individual positions, density and volume

[0025] 2 Calculate the fitness of each individual in the initial population and select the individual with the best fitness as the initial individual 3 Set the number of iterations t = 1

[0026] 4While t<tmax do

[0027] 5for each object ido6 uses strategy 1 to update the density and volume of each object

[0028] 7 Update the transfer factor TF and density factor d t+1

[0029] 8 if TF≤0.5then——go to the exploration stage

[0030] 9 Update the acceleration using strategy 2 and normalize the acceleration

[0031] 10 Update the location using strategy 3

[0032] 11 else——transfer to development stage

[0033] 12 Update acceleration using strategy 2 and normalize acceleration

[0034] 13 Update F

[0035] 14 Update the location using strategy 3

[0036] 15 end if

[0037] 16endfor

[0038] In the above algorithm steps, strategy 1 is the chaotic logistic mapping strategy.

[0039] x n+1 = r·x n ·(1-x n ) (2)

[0040] In the above algorithm steps, strategy 2 is the Gaussian random walk strategy:

[0041] X t+1 =X t +∈ t (4)

[0042] In the above algorithm steps, strategy 3 is the differential mutation perturbation strategy:

[0043] V i =X r1 +F×(X r2 -X r3 ) (5)

[0044] In the above step S3, a sliding window of size 15 is used to sample the daily gas production of the classified tight gas wells, and the samples are divided into a training set and a test set for subsequent prediction;

[0045] In the above step S3, the random forest algorithm is optimized using the optimized Archimedean optimization algorithm to obtain the best parameter combination;

[0046] In the above step S3, the random forest algorithm based on the optimized Archimedean optimization algorithm is used to perform training on the training set, and the accuracy of the algorithm is verified using the test set.

[0047] The beneficial effects of the present invention are:

[0048] 1. Combining intelligent optimization algorithms and machine learning and applying them to the oil field, it overcomes the complexity and difficulty of traditional prediction methods and provides a new means to solve oil and gas extraction problems;

[0049] 2. A tight gas well production classification prediction model was established to assist in the subsequent tight gas production prediction and obtain a more accurate production prediction model;

[0050] 3. The Archimedean optimization algorithm was improved by using chaotic Logisitic mapping, Gaussian random walk, and differential mutation perturbation strategies, which increased the convergence speed of the algorithm and enhanced the global search capability of the algorithm;

[0051] 4. Combining the improved algorithm with the random forest algorithm can more reasonably select the parameter combination of the random forest and improve the prediction accuracy compared with the manual selection of hyperparameter combinations. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0053] Figure 1 It is a flow chart of a method for predicting and evaluating daily production of tight gas based on intelligent optimization algorithm and machine learning according to the present invention;

[0054] Figure 2 Flowchart for optimizing Archimedes optimization algorithm;

[0055] Figure 3 This is a comparison chart of the tight gas daily production prediction and actual production data of the present invention Implementation

[0057] The present invention is further described below in conjunction with the embodiments and drawings.

[0058] The present invention provides a method for predicting and evaluating the daily gas production of tight gas based on intelligent optimization algorithm and machine learning. Figure 1 The technical roadmap of this method includes the following steps:

[0059] S1: Establish a tight gas production regularization process to collect the daily gas production, cumulative gas production and main controlling factors of tight gas well production in the production cycle of each tight gas well in the tight gas reservoir, use the cumulative gas production in the production cycle as a grading measurement indicator for the tight gas well production grade evaluation and prediction model, and use the daily gas production in the production cycle as a label for the tight gas well daily gas production prediction model;

[0060] S2: Based on the K-nearest neighbor algorithm, a tight gas production grade evaluation and prediction model is established to evaluate and predict the production grade of tight gas wells;

[0061] S3: A daily gas production prediction model for tight gas wells is established based on the random forest algorithm optimized by the Archimedean optimization algorithm to predict the production of tight gas wells.

[0062] In the above step S1, the main controlling factors of the tight gas well production include geological factors, drainage parameters and construction parameters;

[0063] The above geological parameters include geological characteristic parameters around the production wells, such as porosity, permeability, gas saturation, reservoir thickness, reservoir depth, dynamic reserves, bottom hole temperature and bottom hole initial pressure;

[0064] The above-mentioned production parameters include initial production, average production in the first year, peak production, average production, maximum water production rate, average gas-water ratio, shut-in time, cumulative water production, bottom hole flowing pressure and daily water production;

[0065] The above-mentioned construction parameters include average sand ratio, construction pressure and fracturing fluid flowback rate;

[0066] In the above step S1, the daily gas production, cumulative gas production and main controlling factors of the tight gas well production during the production period of each tight gas well in the tight gas reservoir are collected as a data set for subsequent prediction.

[0067] In the above data sets, different factors are normalized and unified in dimension to facilitate prediction;

[0068]

[0069] In the above dataset, interpolation method is used to complete the data with missing values;

[0070] In the above data set, noise data and outliers are cleared.

[0071] In the above step S2, the tight gas well production evaluation index is divided into three levels of high, medium and low according to the actual cumulative production, so as to improve the subsequent prediction accuracy.

[0072] In the above step S3, the optimization of the Archimedes optimization algorithm refers to using three strategies to insert the head, middle and tail of the original algorithm respectively to accelerate the convergence speed of the Archimedes optimization algorithm and improve its global search capability.

[0073] The steps of optimizing the Archimedean optimization algorithm are:

[0074] 17 Initialize the population, initialize individual positions, density and volume

[0075] 18 Calculate the fitness of each individual in the initial population and select the individual with the best fitness as the initial individual 19 Set the number of iterations t = 1

[0076] 20While t<t max do

[0077] 21for each object ido22Use strategy 1 to update the density and volume of each object

[0078] 23 Update the transfer factor TF and density factor d t+1

[0079] 24if TF≤0.5then——go to the exploration stage

[0080] 25 Update acceleration using strategy 2 and normalize acceleration

[0081] 26 Update the position using strategy 3

[0082] 27 else——transfer to development stage

[0083] 28 Update acceleration using strategy 2 and normalize acceleration

[0084] 297 Update F30 Update position using strategy 3

[0085] 31 end if

[0086] 32endfor

[0087] In the above algorithm steps, strategy 1 is the chaotic logistic mapping strategy.

[0088] x n+1 = r·x n ·(1-x n ) (2)

[0089] In the above algorithm steps, strategy 2 is the Gaussian random walk strategy:

[0090] X t+1 =X t +∈ t (4)

[0091] In the above algorithm steps, strategy 3 is the differential mutation perturbation strategy:

[0092] V i =X r1 +F×(X r2 -X r3 ) (5)

[0093] In the above step S3, a sliding window of size 15 is used to sample the daily gas production of the classified tight gas wells, and the samples are divided into a training set and a test set for subsequent prediction;

[0094] In the above step S3, the random forest algorithm is optimized using the optimized Archimedean optimization algorithm to obtain the best parameter combination;

[0095] In the above step S3, the random forest algorithm based on the optimized Archimedean optimization algorithm is used to perform training on the training set, and the accuracy of the algorithm is verified using the test set.

[0096] The beneficial effects of the present invention are:

[0097] 1. Combining intelligent optimization algorithms and machine learning and applying them to the oil field, it overcomes the complexity and difficulty of traditional prediction methods and provides a new means to solve oil and gas extraction problems;

[0098] 2. A tight gas well production classification prediction model was established to assist in the subsequent tight gas production prediction and obtain a more accurate production prediction model;

[0099] 3. The Archimedean optimization algorithm was improved by using chaotic Logisitic mapping, Gaussian random walk, and differential mutation perturbation strategies, which increased the convergence speed of the algorithm and enhanced the global search capability of the algorithm;

[0100] 4. Combining the improved algorithm with the random forest algorithm can more reasonably select the parameter combination of the random forest and improve the prediction accuracy compared with the manual selection of hyperparameter combinations.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the present invention can still be modified or replaced by equivalents. Any modification or partial replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A method for predicting and evaluating daily gas production of tight gas based on intelligent optimization algorithm and machine learning, characterized in that: The method comprises the following steps: S1: Establish a tight gas production regularization process to collect the daily gas production, cumulative gas production and main controlling factors of tight gas well production in the production cycle of each tight gas well in the tight gas reservoir, use the cumulative gas production in the production cycle as a grading measurement indicator for the tight gas well production grade evaluation and prediction model, and use the daily gas production in the production cycle as a label for the tight gas well daily gas production prediction model; S2: Based on the K-nearest neighbor algorithm, a tight gas production grade evaluation and prediction model is established to evaluate and predict the production grade of tight gas wells; S3: A daily gas production prediction model for tight gas wells is established based on the random forest algorithm optimized by the Archimedean optimization algorithm to predict the production of tight gas wells.

2. According to claim 1, a method for predicting and evaluating daily gas production of tight gas based on intelligent optimization algorithm and machine learning is characterized in that: The optimized Archimedean optimization algorithm refers to using three strategies to insert into the head, middle and tail of the original Archimedean optimization algorithm respectively, so as to accelerate the convergence speed and improve the global search capability.

3. A method for predicting and evaluating daily gas production of tight gas based on intelligent optimization algorithm and machine learning according to claim 1, characterized in that: The random forest algorithm optimized based on the optimized Archimedean optimization algorithm refers to using the optimized Archimedean optimization algorithm to optimize the hyperparameters of the random forest to obtain the optimal parameter combination.

Images