Intelligent calculation method for injection allocation amount of water injection well based on big data and artificial intelligence

Abstract

The invention discloses an intelligent calculation method for the injection allocation amount of a water injection well based on big data and artificial intelligence. The intelligent calculation method sequentially comprises the following steps that S1, data related to water injection are collected and preprocessed; s2, time sequence features are extracted and optimized; s3, establishing a supervised learning model and performing training; s4, optimizing the supervised learning model by adopting a genetic algorithm; s5, verifying and adjusting the supervised learning model; and S6, deploying the verified model in a production environment to calculate the injection allocation quantity of the water injection well. Through efficient data management, an intelligent algorithm and model analysis, the injection allocation amount of the water injection well can be calculated timely and accurately, and the water injection effect can be predicted, so that the oil field production efficiency is improved, the cost is reduced, and intelligentization and benefit maximization of oil field development are realized.

Description

Technical Field

[0001] The present invention relates to an injection amount calculation method for an oilfield water injection well, and in particular to an intelligent calculation method for the injection amount of a water injection well based on big data and artificial intelligence, belonging to the technical field of stratified water injection. Background Art

[0002] At present, most domestic oil fields are mainly developed by water drive and have achieved good results. Reasonable injection volume is an important guarantee for maintaining efficient development of oil fields, especially in the high water content period of oil fields. Maintaining formation pressure, stabilizing production, and controlling water content increase have become the focus of work. However, the calculation of reasonable injection volume has always been a difficult point in oil field injection work. The following is the calculation method currently used:

[0003] 1. Groundwater dynamics model: This is a mathematical model based on the principle of groundwater flow. By solving the groundwater flow equation, the injection volume of the injection well can be calculated. This method takes into account the complexity of groundwater flow and can accurately predict the effect of water injection. However, the establishment and solution of the groundwater dynamics model requires a large amount of geological and hydrological data, as well as professional numerical calculation technology support.

[0004] 2. Hydrogeological formula: This is a simplified calculation method based on hydrogeological theory and empirical formulas, which is often used in engineering practice. According to the formation parameters and wellbore parameters, the water injection capacity of the injection well is calculated by the hydrogeological formula to determine the injection volume. This method is simple to calculate and is suitable for general engineering projects, but the accuracy is relatively low.

[0005] 3. Empirical formula: For certain types of injection wells or specific geological conditions, there may be empirical formulas for reference. These empirical formulas are usually summarized based on actual engineering experience and test data, and have certain reliability and applicability. However, it should be noted that when using empirical formulas, they should be corrected and adjusted according to actual conditions.

[0006] 4. On-site test and adjustment: In actual projects, the injection amount is often determined through on-site test and adjustment. The optimal injection amount is determined by gradually adjusting the injection amount and observing the changes in groundwater level and injection effect.

[0007] The porous media structure, permeability distribution, oil-water interface morphology and other factors of the formation affect the water injection effect. These geological parameters usually need to be obtained through geological exploration and test data, and their accuracy is limited. Therefore, the complexity of oil field geological conditions causes a lot of workload and challenges to the calculation of injection volume.

[0008] In addition, traditional manual calculations are very labor-intensive and there are also human factors such as operational errors and improper parameter selection that may affect the accuracy of the calculation results of the injection quantity. Summary of the invention

[0009] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0010] In view of the above and / or the deficiencies in the prior art in data processing, fault prediction and system optimization, the present invention is proposed.

[0011] The purpose of the present invention is to overcome the problems existing in the prior art and provide an intelligent calculation method for the injection amount of water injection wells based on big data and artificial intelligence. Through efficient data management, intelligent algorithms and model analysis, the injection amount of water injection wells can be calculated in a timely and accurate manner, and the water injection effect can be predicted, thereby improving the efficiency of oil field production and reducing costs, realizing the intelligence of oil field development and maximizing benefits.

[0012] In order to solve the above technical problems, the intelligent calculation method of water injection well injection amount based on big data and artificial intelligence of the present invention comprises the following steps in sequence:

[0013] S1, collect and pre-process data related to water injection;

[0014] S2, extract and optimize time series features;

[0015] S3. Establish a supervised learning model and conduct training;

[0016] S4, using genetic algorithm to optimize supervised learning model;

[0017] S5. Verify and adjust the supervised learning model;

[0018] S6. Deploy the verified model to the production environment to calculate the injection volume of the injection well.

[0019] Furthermore, step S1 specifically includes the following steps:

[0020] S1.1. Collect data related to water injection;

[0021] S1.2. Clean the collected data;

[0022] S1.3, standardize and normalize the data;

[0023] S1.4. Visualize the data.

[0024] Furthermore, S1 also includes:

[0025] S1.5. Introduce data quality assessment indicators to monitor data quality.

[0026] Furthermore, in S2, seasonal adjustment techniques, trend fitting techniques, and periodic analysis techniques are used to capture the seasonal, trend, and periodic characteristics of time series data.

[0027] Furthermore, in S2, the sliding window technology is used to extract statistical features over a period of time in the past. The statistical features include but are not limited to the mean, variance, standard deviation, skewness and kurtosis, as well as the maximum and minimum values, reflecting the historical trend of the data.

[0028] Furthermore, in S2, the domain knowledge of oil field development is introduced to deeply explore various factors affecting the water injection effect, including but not limited to oil field geological characteristics, bottom hole conditions and formation pressure; through the knowledge of expert systems or domain experts, new features are constructed, including derivative features of formation permeability, interactive features of water injection volume and oil production, to enhance the model's ability to describe data relationships.

[0029] Furthermore, in S2, advanced feature selection techniques are used to screen out the feature subset that contributes most to the model performance. The advanced feature selection techniques include stability selection, tree model-based selection, and clustering-based selection.

[0030] Furthermore, in S2, when the feature space is relatively high, a dimensionality reduction technique is used to reduce the dimension of the data, thereby reducing the complexity and computational cost of the model; the dimensionality reduction technique includes principal component analysis technique and linear discriminant analysis technique.

[0031] Furthermore, in S2, after feature selection is completed, feature importance analysis is performed to evaluate the impact of each feature on model performance; the feature importance analysis includes but is not limited to calculating the importance score of the feature, and displaying the importance of the feature through visualization technology to provide a basis for the interpretation of the model.

[0032] Furthermore, in S3, a gradient boosting machine is used as a supervised learning model to process and predict data related to the water injection effect. A strong learner is formed by gradually constructing and combining multiple weak learners to achieve accurate prediction of the data.

[0033] Furthermore, S3 specifically includes the following sub-steps:

[0034] S3.1, select the gradient boosting machine as the supervised learning model;

[0035] S3.2. Use XGBoo step t and LightGBM library to train the gradient boosting machine model;

[0036] S3.3, k-fold cross validation was used to evaluate the performance of the model;

[0037] S3.4. Optimize the parameters of the model.

[0038] Furthermore, in step S3.2, when the model is trained using XGBoo step t:

[0039] In the training data set D = {x, y i )}, xi is the feature of the i-th sample, yi is the corresponding label; the goal is to minimize the loss function in is the model's prediction for the i-th sample;

[0040]

[0041] Among them, l is the loss function; Ω is the regularization term used to control the complexity of the model; f k is the model of the kth tree;

[0042] In each round of iteration, the loss function is approximated using a second-order Taylor expansion:

[0043]

[0044] Among them, g i and h i are the first and second order derivatives of the loss function;

[0045] Then, minimizing the approximate value of the objective function yields:

[0046]

[0047] Where T is the structure of the tree, γ and λ are regularization parameters, is the weight of the leaf node; by solving this optimization problem, we can get the weight of each tree Thus constructing the final model.

[0048] Furthermore, in step S3.2, LightGBM uses a histogram algorithm to accelerate tree construction and uses gradient-based one-side sampling to reduce the size of training data; in each round of iteration, only samples with larger gradients are retained to reduce the amount of calculation, computing time, and memory usage while maintaining the prediction accuracy of the model.

[0049] Furthermore, the specific steps of step S4 are as follows:

[0050] S4.1. Select genetic algorithm as the optimization algorithm, which represents candidate solutions through individuals in the population and generates new solutions through iterative selection, crossover and mutation operations, which can effectively explore the solution space and find the global optimal or approximate optimal solution;

[0051] S4.2. Design an integrated framework in which the GBM model is used to predict the water injection effect, and the genetic algorithm is used to optimize the parameters of the GBM model to improve the accuracy of the prediction;

[0052] S4.3, design a fitness function that comprehensively considers the prediction accuracy of the GBM model and the practicality of the water injection scheme, and quantifies the pros and cons of individuals in solving the water injection effect prediction problem;

[0053] S4.4. The key parameters of the genetic algorithm include population size, crossover rate and mutation rate. The population size determines the number of candidate solutions considered simultaneously, the crossover rate controls the degree of mixing of genetic information, and the mutation rate determines the algorithm's ability to explore new solution space.

[0054] Furthermore, the specific steps of step S5 are as follows:

[0055] S5.1. Validate the performance of the ensemble model: The holdout method and time series cross-validation method are used to evaluate the performance of the ensemble model. The holdout method involves dividing the dataset into a training set and a test set; the training set is used to train the model, while the test set is used to evaluate the predictive performance of the model; the time series cross-validation method ensures the independence of the validation by splitting the data in chronological order;

[0056] S5.2. Adjust the parameters of GBM and GA based on the validation results to optimize the predictive ability of the model; the adjustment strategies include but are not limited to:

[0057] Modify GBM's learning rate, number and depth of trees, and maximum number of leaf nodes to improve the model's fitting ability;

[0058] Adjust the population size, crossover rate, and mutation rate of GA to improve the algorithm's search ability and convergence speed.

[0059] Furthermore, the specific steps of step S6 are as follows:

[0060] S6.1. Use web frameworks such as Flask and Django to create APIs and deploy trained models to production environments. Model deployment includes configuring model files and related dependent environments on the server and ensuring stable operation and efficient response of the model.

[0061] S6.2. In the actual application of the model, the user inputs the current water injection parameters, including wellhead pressure, water content, and oil production, through the API; the server receives these parameters and passes them to the deployed model for processing. The model calculates based on the input parameters, returns the recommended injection amount, and guides the user to perform water injection operations;

[0062] S6.3. After the model is deployed, the model should be continuously monitored and maintained.

[0063] Compared with the prior art, the present invention has achieved the following beneficial effects: 1. Improved data quality: Through comprehensive data preprocessing, including data collection, cleaning, conversion, visualization and quality monitoring, the high quality and availability of data are ensured, laying a solid foundation for subsequent analysis and modeling work;

[0064] 2. Comprehensive feature extraction: Through time series feature extraction and optimization, the present invention can more comprehensively capture and optimize various factors affecting water injection effects, and construct a prediction model that is both accurate and well interpretable;

[0065] 3. Accurate prediction model: Gradient boosting machine (GBM) is used as a supervised learning model to effectively capture the nonlinear relationship in the data and improve the prediction and generalization capabilities of the model;

[0066] 4. Model optimization: Genetic algorithms are used to optimize the GBM model to further improve the accuracy and practicality of water injection effect prediction, ensuring the excellent performance of the model in solving the problem of water injection effect prediction;

[0067] 5. Model deployment and application: The model is deployed to the production environment through the Web framework. Users can input parameters through the API to obtain prediction results, providing reliable decision support for oilfield development;

[0068] 6. Improved the accuracy and timeliness of the calculation of the injection volume of water injection wells in oil fields, and improved the water injection efficiency: Through accurate prediction models and optimized parameters, the present invention can guide oil fields to carry out water injection operations more effectively, improve water injection efficiency, and thus increase the output of oil reservoirs;

[0069] 7. Reduce production costs: By optimizing the water injection process, reducing unnecessary water injection volume and injection times, and reducing the number of calculations, it helps to reduce the production costs of the oil field and improve the overall economic benefits;

[0070] 8. Improve the reliability of injection volume calculation: By introducing advanced models and algorithms, optimizing model parameters and algorithms, the performance and stability of the early warning model system are continuously improved, thereby improving the reliability of injection volume calculation for oilfield water injection wells;

[0071] 9. Enhance the scientific nature of oilfield management: The present invention provides a data-based decision support system to make oilfield management more scientific and help formulate more reasonable water injection strategies and oilfield development plans;

[0072] 10. Improve the safety of oilfield development: Through accurate water injection prediction and management, the present invention helps to reduce the safety risks caused by improper water injection and improve the safety of oilfield development;

[0073] 11. Promote sustainable development of oil fields: By optimizing water injection technology, the present invention helps to achieve sustainable development of oil field resources, extend the life of oil fields, and reduce environmental impact. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor. The accompanying drawings are only provided for reference and explanation, and are not used to limit the present invention.

[0075] in:

[0076] Figure 1 This is a flow chart of an intelligent calculation method for injection volume of water injection wells based on big data and artificial intelligence according to the present invention;

[0077] Figure 2 is a flow chart of step S1 in the present invention;

[0078] Figure 3 is a flow chart of step S2 in the present invention;

[0079] Figure 4 is a flow chart of step S3 in the present invention;

[0080] Figure 5 is a flow chart of step S4 in the present invention;

[0081] Figure 6 is a flow chart of step S5 in the present invention;

[0082] Figure 7 is a flow chart of step S6 in the present invention;

[0083] Figure 8 This is a real-time water injection interface diagram of the first embodiment of the present invention;

[0084] Fig. 9 This is a real-time water injection interface diagram of the second embodiment of the present invention. DETAILED DESCRIPTION

[0085] In the following description of the present invention, the directions or positional relationships indicated by terms such as "up", "down", "front", "back", "left", "right", "inside" and "outside" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description, and do not mean that the device must have a specific direction.

[0086] In order to make the technical means, creative features, objectives and effects of the present invention easier to understand, the present invention is further described below in conjunction with specific diagrams. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them.

[0087] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0088] like Figures 1 to 6 As shown, the intelligent calculation method of water injection well injection amount based on big data and artificial intelligence of the present invention comprises the following steps in sequence:

[0089] S1. Collect and pre-process data related to water injection

[0090] This step involves comprehensive preprocessing of the data to ensure the quality and availability of the data, providing a solid foundation for subsequent analysis and modeling. Data preprocessing includes but is not limited to key links such as data collection, cleaning, conversion, visualization, and quality monitoring.

[0091] S1.1. Collect data related to water injection

[0092] In the data collection stage, the method of the present invention involves collecting and integrating water injection data, geological information, downhole equipment information and surface topographic data, surface flow, formation water injection volume, geological data, and corresponding oil well data before and after water injection for the past 10 years. The geological data include layer position, layer number, formation depth, formation water absorption rate, oil-bearing area, water zone affected area, water zone controlled reserves and water zone mobilized reserves. The corresponding oil well data before and after water injection include oil well name, liquid production before water injection, water content before water injection, liquid production after water injection and water content after water injection.

[0093] In addition, in order to comprehensively evaluate the environmental factors and influencing factors of the water injection process, the acquisition of other relevant data was also considered, including but not limited to groundwater level, surface water conditions, meteorological data, etc.

[0094] S1.2. Clean the collected data

[0095] In the data cleaning stage, the present invention uses a variety of techniques to deal with missing values ​​and outliers. For missing values, the present invention not only uses interpolation methods to fill them, but also combines domain knowledge to perform reasonable data recovery. For outliers, the present invention tries a variety of correction methods to retain valid information in the data while ensuring the accuracy and reliability of the data.

[0096] S1.3. Standardize and normalize the data

[0097] The data conversion stage includes standardizing and normalizing the data to eliminate the effects of different dimensions and data scales. The present invention also considers using feature selection methods to reduce data dimensions, eliminate redundant information, and enhance the generalization ability of the model. In order to handle nonlinear relationships in the data, the present invention also uses polynomial feature transformation or kernel methods.

[0098] S1.4. Visualize the data

[0099] Through data visualization, the present invention uses tools Matplotlib and Seaborn to intuitively display data to guide feature engineering and model selection. In addition, the present invention also uses interactive visualization tools to provide a richer and more dynamic data display method to enhance the interpretability of data.

[0100] S1.5. Monitor data quality

[0101] In the data quality monitoring stage, the present invention establishes a monitoring system and introduces data quality evaluation indicators, including data integrity, consistency, accuracy, etc. Through quantitative evaluation and real-time monitoring of data quality, potential problems are discovered in a timely manner and corresponding measures are taken to deal with them, ensuring the reliability and continuity of data.

[0102] Through the above-mentioned comprehensive data preprocessing steps, the present invention can significantly improve the quality and availability of data, and lay a solid foundation for subsequent analysis and modeling work.

[0103] S2. Extract time series features and optimize

[0104] In this step, the method of the present invention focuses on extracting key features from time series data and combines domain-specific knowledge with advanced feature selection techniques to optimize the feature engineering process, thereby improving the predictive power and interpretability of the model.

[0105] S2.1. Time series feature extraction

[0106] To capture key features such as seasonality, trend, and periodicity in time series data, this paper uses the following techniques:

[0107] Seasonal Adjustment: A seasonal decomposition technique, X-11 decomposition, is used to identify and isolate seasonal patterns.

[0108] Trend fitting: Estimate and remove long-term trends in data through time series analysis, exponential smoothing or autoregressive moving average models (ARMA).

[0109] Periodicity Analysis: Use Fourier transform or other frequency domain analysis methods to identify periodic fluctuations in data.

[0110] In addition, the present invention uses sliding window technology to extract statistical features over a period of time in the past, including but not limited to mean, variance, standard deviation, skewness and kurtosis, as well as maximum and minimum values, etc., to reflect the historical trend of the data.

[0111] S2.2. Introducing domain knowledge

[0112] Combined with the domain knowledge of oilfield development, the present invention deeply explores various factors that affect the water injection effect, including but not limited to oilfield geological characteristics, bottom hole conditions, formation pressure, etc. Through the knowledge of expert systems or domain experts, new features are constructed, such as the derivative features of formation permeability, the interactive features of water injection volume and oil production, etc., to enhance the model's ability to describe data relationships.

[0113] S2.3. Advanced feature selection techniques

[0114] In order to select the feature subset that contributes most to model performance, this paper uses a series of advanced feature selection techniques:

[0115] Stability Selection: Identify features that are stable across different data samples through stability selection methods, Ridge Regression and Lasso Regression.

[0116] Tree-based selection: Use models such as decision trees or random forests to assess feature importance based on their segmentation purity.

[0117] Cluster-based selection: Through cluster analysis, features with similarities are identified and representative features are selected from them.

[0118] S2.4. Data Dimensionality Reduction

[0119] In the case of high feature space, the present invention uses dimensionality reduction techniques, including principal component analysis (PCA) and linear discriminant analysis (LDA), to reduce the dimensionality of the data while retaining the most important information, thereby reducing the complexity and computational cost of the model.

[0120] S2.5. Feature Importance Analysis

[0121] After feature selection is completed, the present invention performs feature importance analysis to evaluate the impact of each feature on model performance. This includes but is not limited to calculating the importance score of the feature and displaying the importance of the feature through visualization technology to provide a basis for the interpretation of the model.

[0122] Through the above steps, the present invention can more comprehensively consider various factors affecting the water injection effect and construct a prediction model that is both accurate and well interpretable.

[0123] S3. Build a supervised learning model and train it

[0124] In this step, the present invention uses Gradient Boosting Machine (GBM) as a supervised learning model to process and predict data related to water injection effects. GBM is a powerful ensemble learning algorithm that forms a strong learner by gradually building and combining multiple weak learners (usually decision trees) to achieve accurate prediction of data.

[0125] S3.1 Model selection

[0126] This paper selects the GBM model as the supervised learning model. The GBM model can effectively capture the nonlinear relationship in the data and has shown excellent performance in various data science competitions and practical applications. By optimizing the loss function, a new model is added at each step to minimize the residual of the previous model, thereby gradually improving the overall prediction ability.

[0127] S3.2 Model Training

[0128] In the model training stage, the present invention uses XGBoot (eXtreme Gradient Boosting) and LightGBM (Light Gradient Boosting Machine) libraries to implement GBM model training. These libraries provide efficient and scalable GBM algorithms, which speed up training through parallel computing and highly optimized algorithms while reducing memory usage.

[0129] 1. XGBoo step t is an optimized implementation of Gradient Boosting. Its core idea is to minimize the loss function by iteratively adding new models. XGBoo step t uses the second-order Taylor expansion to optimize the objective function, which can converge to the optimal solution faster.

[0130] In the training data set D = {x, y i )}, xi is the feature of the i-th sample, and yi is the corresponding label. The goal is to minimize the loss function in is the model's prediction for the i-th sample.

[0131]

[0132] Among them, l is the loss function, such as square loss or logistic loss; Ω is the regularization term used to control the complexity of the model; f k is the model of the kth tree.

[0133] In each round of iteration, the loss function is approximated using a second-order Taylor expansion:

[0134]

[0135] Among them, g i and h i are the first and second order derivatives of the loss function.

[0136] Then, minimizing the approximate value of the objective function yields:

[0137]

[0138] Where T is the structure of the tree, γ and λ are regularization parameters, is the weight of the leaf node.

[0139] By solving this optimization problem, we can get the weight of each tree Thus constructing the final model.

[0140] LightGBM: Uses gradient-based single-side sampling and histogram algorithms to reduce the amount of computation while maintaining high prediction accuracy.

[0141] In LightGBM, a histogram algorithm is used to speed up tree construction. Specifically, the continuous eigenvalues ​​are discretized into multiple intervals, and then the sum of the gradients and the sum of the second-order gradients in each interval are counted. In this way, when building the tree, only the information of these intervals can be considered, which greatly reduces the amount of calculation.

[0142] In addition, LightGBM also uses gradient-based unilateral sampling to reduce the size of training data. In each round of iteration, only samples with larger gradients are retained, which can reduce the amount of calculation while maintaining the prediction accuracy of the model.

[0143] Through these optimizations, LightGBM can greatly reduce computation time and memory usage when training large-scale datasets.

[0144] S3.3 Model Evaluation

[0145] In order to evaluate the performance of the model, the present invention adopts a cross-validation method. Cross-validation is a resampling technique used to estimate the generalization ability of the model. In the present invention, k-fold cross-validation is used to evaluate the performance of the model, wherein the data set is divided into k mutually exclusive subsets of equal size, the model is trained on k-1 subsets, and tested on the remaining subset. This process is repeated k times, each time a different subset is selected as the test set, and the final performance evaluation result is the average of the k evaluation results.

[0146] S3.4 Parameter Optimization

[0147] The present invention also involves the optimization of model parameters. By adjusting key parameters such as learning rate, number of trees, tree depth, maximum number of leaf nodes, etc., the optimal model configuration can be found to maximize the prediction performance of the model.

[0148] Through these steps, the present invention can train an accurate supervised learning model with strong generalization ability to predict and classify water injection effects, providing powerful decision support for oilfield development.

[0149] S4. Using genetic algorithm to optimize supervised learning model

[0150] The present invention uses a genetic algorithm (GA) to optimize the supervised learning model, especially to optimize the parameters and structure of the gradient boosting machine (GBM) model. The genetic algorithm is an efficient global optimization algorithm, which is inspired by the theory of biological evolution and optimizes problems by simulating natural selection and genetic mechanisms.

[0151] The following is the principle process of genetic algorithm:

[0152] 1. Population Initialization: Set the population size to N, the chromosome length of each individual to L, then the population is represented by P = {p1, p2, ..., p N}, where each individual p i is a string of length L.

[0153] 2. Fitness evaluation (Fitne step Evaluation): For each individual p i , calculate its fitness f(p i ). Fitness function f(p i ) is designed based on the specific needs of the problem and is used to assess an individual's performance in problem solving.

[0154] 3. Selection (election step): The selection operation can be performed based on the fitness ratio, such as roulette wheel selection. i The probability of being selected is P(p i ) is expressed as:

[0155]

[0156] 4. Crossover (Cro step over): The crossover operation can be performed with probability p c To control. For the two selected parent individuals p a and p b , the offspring individuals c1 and c2 after crossover can be expressed as:

[0157] c1=p a[:m]+p b [m:]

[0158] c2=p b [:m]+p a [m:]

[0159] Among them, m is the crossover point, indicating the gene position where the crossover occurs.

[0160] 5. Mutation: The mutation operation can be performed with probability p m To control. For individual p i For each gene g in , the mutated gene g′ can be expressed as:

[0161] g′=g+Δ

[0162] Here, Δ is a random small perturbation.

[0163] 6. Replacement: The replacement operation can be implemented with different strategies, such as elitism or roulette selection. Elitism retains the best individuals in the current population, and other individuals are replaced by new individuals.

[0164] 7. Termination Condition: The termination condition can be reaching a predetermined number of iterations T, or the fitness change in the population is less than a certain threshold ∈.

[0165] The specific steps of optimizing the GBM model using genetic algorithm are as follows:

[0166] S4.1. Optimization algorithm selection

[0167] The present invention selects genetic algorithm as the optimization algorithm because it has unique advantages in dealing with global optimization problems. Genetic algorithm represents candidate solutions through individuals in the population, and generates new solutions through iterative selection, crossover and mutation operations, which can effectively explore the solution space and find the global optimal or approximate optimal solution.

[0168] S4.2. Design of integration framework

[0169] The present invention designs an integrated framework, which uses the prediction results of the GBM model as part of the fitness evaluation of the genetic algorithm. In this framework, the GBM model is used to predict the water injection effect, and the genetic algorithm is used to optimize the parameters of the GBM model to improve the accuracy of the prediction.

[0170] S4.3. Design fitness function

[0171] In order to evaluate the fitness of individuals in the genetic algorithm, the present invention designs a fitness function, which comprehensively considers the prediction accuracy of the GBM model and the practicality of the water injection scheme. The purpose of the fitness function is to quantify the pros and cons of individuals (i.e., the parameter configuration of the GBM model) in solving the problem of water injection effect prediction.

[0172] S4.4. Setting parameters

[0173] In the implementation of the genetic algorithm, the present invention has carefully set the key parameters of the genetic algorithm, including population size, crossover rate, mutation rate, etc. The setting of these parameters has an important impact on the convergence speed of the algorithm and the quality of the final solution. The population size determines the number of candidate solutions considered at the same time, the crossover rate controls the degree of mixing of genetic information, and the mutation rate determines the ability of the algorithm to explore new solution space.

[0174] Through these steps, the present invention can optimize the GBM model using a genetic algorithm, thereby improving the accuracy and practicability of water injection effect prediction.

[0175] S5. Verify and adjust the supervised learning model

[0176] S5.1. Verify the performance of the integrated model

[0177] The present invention adopts the hold-out method and time series cross-validation method to evaluate the performance of the integrated model. The hold-out method involves dividing the data set into a training set and a test set, wherein the training set is used to train the model, and the test set is used to evaluate the predictive performance of the model. The time series cross-validation rule is a validation method particularly suitable for time series data, which ensures the independence of validation by splitting the data in chronological order.

[0178] S5.2. Adjust model parameters based on validation results

[0179] According to the validation results, the present invention adjusts the parameters of GBM and GA to optimize the prediction ability of the model. The adjustment strategy includes but is not limited to:

[0180] Modify GBM parameters such as learning rate, number and depth of trees, maximum number of leaf nodes, etc. to improve the model's fitting ability.

[0181] Adjust GA parameters such as population size, crossover rate, mutation rate, etc. to improve the algorithm's search ability and convergence speed.

[0182] Based on the performance of the model on the validation set, add or delete features or change the feature engineering method to improve the generalization ability of the model.

[0183] Through these adjustment strategies, the present invention can ensure that the integrated model has good predictive performance on new, unseen data, thereby providing reliable decision support for oilfield development.

[0184] S6. Deploy the verified model to the production environment to calculate the injection volume of the injection well

[0185] This step involves deploying the trained and validated model into a production environment and describes its workflow in real-world applications.

[0186] S6.1. Model deployment

[0187] The present invention uses web frameworks such as Flask and Django to create APIs so that the trained models can be deployed in the production environment. These web frameworks can provide lightweight servers so that users can interact with the models through HTTP requests. The deployment of the model includes configuring the model files and related dependent environments on the server and ensuring the stable operation and efficient response of the model.

[0188] S6.2 Model Application

[0189] In actual applications, users input current water injection parameters through the API, including wellhead pressure, water content, oil production, etc. The server receives these parameters and passes them to the deployed model for processing. The model calculates based on the input parameters and returns the recommended injection amount to guide users in water injection operations.

[0190] S6.3. Model monitoring and maintenance

[0191] After the model is deployed, the present invention also includes the continuous monitoring and maintenance of the model. This includes monitoring the performance and stability of the model, and regularly updating the model to reflect new data and changes in conditions.

[0192] like Figure 8 As shown, the present invention is trial applied to Well Chen 3-38 of the Jiangsu Oilfield Production Plant No. 1. The upper left side is the real-time performance diagram data of the water injection well corresponding to the oil well, including the number of flushing, liquid production, and acquisition time; the upper right side is the downhole formation connection diagram of the water injection well, i.e. Well Chen 3-38, and the corresponding connected wells, i.e. Oil wells Chen 2-59A and Chen 3 Ping 21.

[0193] The lower left side shows the daily water injection volume of Chen 3-38 well during the test period. It can be seen that the daily water injection volume of the present invention shows a downward trend. The lower right side shows the daily liquid production of the water injection well and its corresponding connected oil well. It can be seen that during the test period, when the water injection volume decreased significantly, the liquid production remained stable. This shows that the oil well adopts the optimized water injection scheme of the present invention, which achieves a decrease in water injection volume but no decrease in liquid production, thereby reducing energy consumption.

[0194] like Fig. 9As shown, the present invention is trial applied to Well Chen 3-43 of the Jiangsu Oilfield Production Plant No. 1. The upper left side is the real-time performance diagram data of the water injection well corresponding to the oil well, including the number of flushing, liquid production, and acquisition time; the upper right side is the downhole formation connection diagram of the water injection well, i.e. Well Chen 3-43, and the corresponding connected wells, i.e. Oil wells Chen 3-101, Chen 3-102, and Chen 3-8.

[0195] The lower left side shows the daily water injection volume of Chen 3-43 well during the test period. It can be seen that the daily water injection volume of the present invention shows a downward trend. The lower right side shows the daily liquid production of the water injection well and its corresponding connected oil well. It can be seen that during the test period, when the water injection volume decreased significantly, the liquid production remained stable. This shows that the oil well adopts the water injection scheme optimized by the present invention, and also achieves a decrease in water injection volume but no decrease in liquid production, thereby reducing energy consumption.

[0196] The above is only a preferred embodiment of the present invention, which shows and describes the basic principles, main features and advantages of the present invention, but does not limit the scope of patent protection of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. In addition to the above embodiments, the present invention may have other implementation modes without departing from the spirit and scope of the present invention. The present invention may also have various changes and improvements, and all technical solutions formed by equivalent substitution or equivalent transformation fall within the scope of protection required by the present invention. The scope of protection required by the present invention is defined by the attached claims and their equivalents. The technical features not described in the present invention can be realized by or using existing technologies, which will not be repeated here.

Claims

1. An intelligent calculation method for injection volume of water injection wells based on big data and artificial intelligence, characterized in that: The steps are as follows: S1, collect and pre-process data related to water injection; S2, extract and optimize time series features; S3. Establish a supervised learning model and conduct training; S4, using genetic algorithm to optimize supervised learning model; S5. Verify and adjust the supervised learning model; S6. Deploy the verified model to the production environment to calculate the injection volume of the injection well.

2. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized in that: Step S1 specifically includes the following steps: S1.

1. Collect data related to water injection; S1.

2. Clean the collected data; S1.3, standardize and normalize the data; S1.

4. Visualize the data.

3. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 2 is characterized in that: Step S1 also includes: S1.

5. Introduce data quality assessment indicators to monitor data quality.

4. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized in that: In step S2, seasonal adjustment technology, trend fitting technology and periodic analysis technology are used to capture the seasonal, trend and periodic characteristics in the time series data.

5. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized by: In step S2, a sliding window technique is used to extract statistical features over a period of time in the past. The statistical features include but are not limited to mean, variance, standard deviation, skewness and kurtosis, as well as maximum and minimum values, reflecting the historical trend of the data.

6. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized by: In step S2, the domain knowledge of oil field development is introduced to deeply explore various factors affecting the water injection effect, including but not limited to oil field geological characteristics, bottom hole conditions and formation pressure; through the knowledge of expert systems or domain experts, new features are constructed, including derivative features of formation permeability, interactive features of water injection volume and oil production, to enhance the model's ability to describe data relationships.

7. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized by: In step S2, an advanced feature selection technique is used to screen out the feature subset that contributes most to the model performance. The advanced feature selection technique includes stability selection, tree model-based selection, and clustering-based selection.

8. The intelligent calculation method of injection volume of water injection wells based on big data and artificial intelligence according to claim 1 is characterized by: In step S2, when the feature space is high, a dimensionality reduction technique is used to reduce the dimension of the data, thereby reducing the complexity and computational cost of the model; the dimensionality reduction technique includes principal component analysis technique and linear discriminant analysis technique.

9. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 7 is characterized by: In step S2, after feature selection is completed, feature importance analysis is performed to evaluate the impact of each feature on model performance; the feature importance analysis includes but is not limited to calculating the importance score of the feature, and displaying the importance of the feature through visualization technology to provide a basis for the interpretation of the model.

10. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized by: In step S3, a gradient boosting machine is used as a supervised learning model to process and predict data related to the water injection effect, and a strong learner is formed by gradually constructing and combining multiple weak learners to achieve accurate prediction of the data.

11. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1 is characterized by: Step S3 specifically includes the following sub-steps: S3.1, select the gradient boosting machine as the supervised learning model; S3.

2. Use XGBoo step t and LightGBM library to train the gradient boosting machine model; S3.3, k-fold cross validation was used to evaluate the performance of the model; S3.

4. Optimize the parameters of the model.

12. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 11, characterized in that: In step S3.2, when using XGBoo step t for model training: In the training data set D = {x, y i )}, xi is the feature of the i-th sample, and yi is the corresponding label; The goal is to minimize the loss function in is the model's prediction for the i-th sample; Among them, l is the loss function; Ω is the regularization term used to control the complexity of the model; f k is the model of the kth tree; In each round of iteration, the loss function is approximated using a second-order Taylor expansion: Among them, g i and h i are the first and second order derivatives of the loss function; Then, minimizing the approximate value of the objective function yields: Where T is the structure of the tree, γ and λ are regularization parameters, is the weight of the leaf node; by solving this optimization problem, we can get the weight of each tree Thus constructing the final model.

13. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 11, characterized in that: In step S3.2, LightGBM uses the histogram algorithm to accelerate the construction of the tree and uses gradient-based one-side sampling to reduce the size of the training data; in each round of iteration, only samples with larger gradients are retained to reduce the amount of calculation, computing time, and memory usage while maintaining the prediction accuracy of the model.

14. The intelligent calculation method for injection volume of water injection wells based on big data and artificial intelligence according to claim 1, characterized in that: The specific steps of step S4 are as follows: S4.

1. Select genetic algorithm as the optimization algorithm, which represents candidate solutions through individuals in the population and generates new solutions through iterative selection, crossover and mutation operations, which can effectively explore the solution space and find the global optimal or approximate optimal solution; S4.

2. Design an integrated framework in which the GBM model is used to predict the water injection effect, and the genetic algorithm is used to optimize the parameters of the GBM model to improve the accuracy of the prediction; S4.3, design a fitness function that comprehensively considers the prediction accuracy of the GBM model and the practicality of the water injection scheme, and quantifies the pros and cons of individuals in solving the water injection effect prediction problem; S4.

4. The key parameters of the genetic algorithm include population size, crossover rate and mutation rate. The population size determines the number of candidate solutions considered simultaneously, the crossover rate controls the degree of mixing of genetic information, and the mutation rate determines the algorithm's ability to explore new solution space.

15. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1, characterized in that: The specific steps of step S5 are as follows: S5.

1. Validate the performance of the ensemble model: The holdout method and time series cross-validation method are used to evaluate the performance of the ensemble model. The holdout method involves dividing the dataset into a training set and a test set. The training set is used to train the model, while the test set is used to evaluate the prediction performance of the model. The time series cross-validation method ensures the independence of the validation by splitting the data in chronological order. S5.

2. Adjust the parameters of GBM and GA based on the validation results to optimize the predictive ability of the model; the adjustment strategies include but are not limited to: Modify GBM's learning rate, number and depth of trees, and maximum number of leaf nodes to improve the model's fitting ability; Adjust the population size, crossover rate, and mutation rate of GA to improve the algorithm's search ability and convergence speed.

16. The intelligent calculation method of water injection well injection amount based on big data and artificial intelligence according to claim 1, characterized in that: The specific steps of step S6 are as follows: S6.

1. Use web frameworks such as Flask and Django to create APIs and deploy trained models to production environments. Model deployment includes configuring model files and related dependent environments on the server and ensuring stable operation and efficient response of the model. S6.

2. In the actual application of the model, the user inputs the current water injection parameters, including wellhead pressure, water content, and oil production, through the API; the server receives these parameters and passes them to the deployed model for processing. The model calculates based on the input parameters, returns the recommended injection amount, and guides the user to perform water injection operations; S6.

3. After the model is deployed, the model should be continuously monitored and maintained.

Images