Oil-water well yield prediction method and device based on neural network model

Abstract

The invention relates to the technical field of oil-water well yield prediction, in particular to an oil-water well yield prediction method and device based on a neural network model. The method comprises the following steps: constructing a neural network model and establishing a model database connected with the neural network model; performing primary processing on the historical data and dividing the historical data into a training set and a test set; inputting the training set data into the neural network model for training; testing is carried out through the test set data, whether the test result meets the requirement or not is judged, and if not, optimization training is carried out on the model till the model subjected to optimization training meets the requirement; and carrying out yield prediction by using the neural network model meeting the requirements. The problems that in the prior art, a model for yield prediction cannot automatically obtain and update latest data, so that the prediction precision is gradually reduced when the model carries out subsequent yield prediction, manual operation is needed during optimization, updating and prediction, the labor cost is increased, and the working efficiency is reduced are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and water well production prediction, and in particular to an oil and water well production prediction method and device based on a neural network model. Background Art

[0002] In the development of oil and gas fields, in order to increase the oil production rate and recovery rate, a variety of production and injection measures are often taken. However, the current measures are often formulated according to production needs, relying on past experience, and lack of systematicity and scientificity. In order to improve this situation, machine learning models are widely used in different fields of oil exploration and development.

[0003] The neural network method is a machine learning method that simulates the structure and working principle of the human brain's nervous system. It uses a network hierarchy composed of multiple nodes (neurons) to transmit and process information, and learns the relationship between input and output by continuously adjusting the network weights. In oil and gas field development, the neural network method can be applied to establish a complex nonlinear relationship model between oil well production and various parameters (such as geological attributes, production parameters, etc.).

[0004] Machine learning methods can use a large amount of historical data and monitoring data for training, and make predictions and decisions based on the learned patterns and laws. Compared with traditional rules of thumb, machine learning models can more accurately predict oil well production, optimize production-increasing measures, and provide a scientific basis to guide oilfield development decisions. At the same time, machine learning models can also be updated and optimized based on real-time data to achieve intelligent oil and gas field management and optimization, and can help solve the problem of lack of systematicity and scientificity in the formulation of measures, thereby improving the efficiency and output of oilfield development.

[0005] However, current machine learning usually uses the model that was initially established and trained to make subsequent production predictions multiple times. As oil wells continue to be developed and the number of fracturing operations increases, the effectiveness and accuracy of the original model's prediction results gradually decrease and cannot meet development needs. Summary of the invention

[0006] The present invention proposes a method and device for predicting oil and water well production based on a neural network model to solve the problem that the production prediction model in the prior art cannot automatically obtain and update the latest data, resulting in a gradual decrease in prediction accuracy when the model performs subsequent production predictions, and manual operations are required during optimization, updating and prediction, which increases labor costs and reduces work efficiency.

[0007] According to one aspect of the present invention, a method for predicting oil and water well production based on a neural network model is provided, comprising:

[0008] Constructing a neural network model and establishing a model database connected to the neural network model, wherein the model database is used to obtain historical data updated in the work area in real time or at predetermined time intervals;

[0009] Performing preliminary processing on the updated historical data;

[0010] Dividing the preliminarily processed historical data into a training set and a test set;

[0011] Inputting the training set data into the neural network model to train the model;

[0012] The trained neural network model is tested by the test set data to determine whether the test result meets the requirements. If not, the model is optimized and trained until the optimized and trained model meets the requirements.

[0013] The neural network model that meets the requirements is used to predict production.

[0014] Preferably, the historical data includes the following main parameters related to oil production:

[0015] Among them, the main parameters include at least: cumulative oil increase, oil production layer before pressure, daily liquid production, effective thickness, perforated sandstone thickness layer, water saturation, original saturation pressure, pump diameter, middle depth of oil layer, fracture pressure, fracture layer, sandstone thickness, production days, permeability, porosity, dynamic liquid level, top depth, pump depth, bottom depth, small block code and layer segment.

[0016] Preferably, the preliminary processing includes:

[0017] Obtaining a field explanation table of each main parameter in the historical data, including a data set structure, field meaning, and data type;

[0018] Determine whether there is invalid data or an empty string in the field interpretation table, and if so, process it by filling, deleting or interpolating;

[0019] Determine whether there are abnormal values ​​in the field interpretation table, and if yes, process them by deletion, replacement or interpolation;

[0020] Determine whether there is a data type error in the field interpretation table, and if so, convert it to a correct data type;

[0021] Determine whether there are duplicate data rows in the field interpretation table, and if yes, delete the duplicate data rows;

[0022] Perform text cleaning on the data in the field explanation table;

[0023] Determine whether the format of the same type of data in the field interpretation table is consistent, and if not, modify it to a unified format;

[0024] Normalizing and screening the data in the field explanation table;

[0025] The data in the field explanation table is verified to determine whether it meets the accuracy and completeness requirements. If not, the data that does not meet the accuracy and completeness requirements is deleted.

[0026] Preferably, the preliminary processing further comprises:

[0027] The main parameters in the historical data are linearized.

[0028] Preferably, before using the neural network model that meets the requirements to perform production forecasting, the method further includes:

[0029] Determine whether the total number of training times of the neural network model is greater than or equal to a predetermined number of times. If not, re-divide the historical data into a training set and a test set, and input the data into the neural network model to re-train and test the model;

[0030] If yes, determine whether the test results of the retrained model meet the requirements; if not, re-process the historical data, and use the training set and test set divided by the re-processed historical data to train and test the neural network model.

[0031] Preferably, 10 times.

[0032] Preferably, the neural network model is: a one-dimensional convolutional fully connected neural network model.

[0033] According to one aspect of the present invention, there is provided an oil and water well production prediction device based on a neural network model, comprising:

[0034] A database and model building unit, used to build a neural network model and to build a model database connected to the neural network model, wherein the model database is used to obtain historical data updated in the work area in real time or at predetermined time intervals;

[0035] A data preliminary processing unit, used for performing preliminary processing on the updated historical data;

[0036] A data division unit, used for dividing the historical data after the preliminary processing into a training set and a test set;

[0037] A model training unit, used for inputting the training set data into the neural network model to train the model;

[0038] A model optimization unit is used to test the trained neural network model through the test set data to determine whether the test result meets the requirements. If not, the model is optimized and trained until the optimized and trained model meets the requirements.

[0039] The yield prediction unit is used to perform yield prediction using the neural network model that meets the requirements.

[0040] The present invention has at least the following beneficial effects:

[0041] The present invention proposes an oil and water well production prediction method and device based on a neural network model. The neural network model is connected to a real-time updated model database to obtain the latest historical data, and the network model is continuously trained to keep the model in an optimal state. The production can be predicted at any time or regularly, thereby ensuring that the accuracy of the production prediction result each time is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The accompanying drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and, together with the specification, are used to explain the technical solutions of the present invention.

[0043] Figure 1 A flow chart of a method for predicting oil and water well production based on a neural network model according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0044] Various exemplary embodiments, features and aspects of the present invention will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0045] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0046] The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there may be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the term "at least one" herein represents any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C can represent including any one or more elements selected from the set consisting of A, B, and C.

[0047] In addition, in order to better illustrate the present invention, numerous specific details are provided in the following specific embodiments. It should be understood by those skilled in the art that the present invention can be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present invention.

[0048] Figure 1 FIG. 2 shows a flow chart of a method for predicting oil and water well production based on a neural network model according to an embodiment of the present invention. Figure 1 As shown, a method for predicting oil and water well production based on a neural network model includes: step S01: constructing a neural network model and establishing a model database connected to the neural network model, the model database is used to obtain historical data updated in a work area in real time or at a predetermined time interval; step S02: preliminarily processing the updated historical data; step S03: dividing the preliminarily processed historical data into a training set and a test set; step S04: inputting the training set data into the neural network model to train the model; step S05: testing the trained neural network model through the test set data to determine whether the test result meets the requirements, if not, optimizing the model until the optimized trained model meets the requirements; and using the neural network model that meets the requirements to predict production.

[0049] The oil and water well production prediction method based on the neural network model provided in the embodiment of the invention specifically includes the following steps:

[0050] Step S01: construct a neural network model and establish a model database connected to the neural network model, wherein the model database is used to obtain historical data updated in the work area in real time or at predetermined time intervals.

[0051] In the embodiment of the present invention, the model database is connected to the oilfield information center database to obtain all data related to oilfield production and construction in the work area, that is, historical data, which may include: basic data, geological data, production data, test data and layer data, etc. As the production process continues, the production data is also constantly updated and increased. The database can be set to obtain updated historical data in real time or at a predetermined time interval. If the predetermined time interval is 30 days, the database will obtain the newly generated historical data within the previous 30 days every 30 days.

[0052] In the present invention, the historical data contains main parameters related to oil production: wherein, the main parameters include at least: cumulative oil increase, oil production layer before pressure, daily liquid production, effective thickness, perforated sandstone thickness layer, water saturation, original saturation pressure, pump diameter, middle depth of oil layer, fracture pressure, fracture layer, sandstone thickness, production days, permeability, porosity, dynamic liquid level, top depth, pump depth, bottom depth, small block code, layer segment.

[0053] In the embodiment of the present invention, since the historical data contains many types of parameters, too much data during training will cause the training time to be too long or affect the accuracy of the training model. Therefore, it is necessary to screen the data and select the parameters that are more critical to the output.

[0054] The parameters that meet the conditions are initially screened out through the previous geological and process parameters. The database provided by the Oilfield Information Center contains oil storage and reservoir data, process measures data records, etc. The main data include: single well basic data, geological basic data, single well stratification data, single well connectivity data, single well small layer interlayer data, formation stratification data, oil layer stratification data, small layer grouping data, sandstone grouping data, perforation data, single well perforation stratification data, reservoir basic data, reservoir property data, small layer evaluation, oilfield development foundation, water well daily data, water well monthly data, oil well daily data, oil well monthly data, flow static pressure data, stratified flow static pressure data, water well stratification data, production profile test, injection profile test, operation log, operation record, acidizing record, fracturing record, skin coefficient, etc.

[0055] The database of the oilfield information center is mirrored, and the model database of the present invention is built on the server, so that the static data and dynamic data of the oilfield can be synchronized or regularly updated in the model database.

[0056] The parameters that already exist in the database of the oilfield information center, such as effective thickness, monthly production, and the number of connected wells, are directly extracted from the database. The parameters related to the remaining oil reserves lack accurate measurement parameters, and there is a lack of effective standard values ​​for comparison during calculation. They are not suitable for prediction by artificial intelligence algorithms, and can be obtained through calculations using relevant data provided by the geological department. Relevant parameters such as formation pressure, permeability, and porosity will change with production dynamics, and these parameters are highly correlated. Changes in their parameters will be reflected in changes in formation pressure. Therefore, in order to ensure the accuracy of the prediction, the formation pressure must be calculated and predicted in the early stage. Generally, historical formation pressure or formation pressure of other related wells is used for simulation calculations.

[0057] Through relevant literature, the formulas and theoretical knowledge related to formation pressure are summarized, and appropriate adjustments are made according to the parameters and data provided. The relationship between important parameters is analyzed, the parameters are classified according to different standards, and the parameters with high prediction correlation are screened out.

[0058] Three rounds of screening were conducted among the parameters with high correlation with the prediction. According to the parameters provided, relevant information was consulted in a targeted manner, and parameters with high frequency of occurrence in the prediction and parameters with high theoretical importance were screened, and finally the parameters selected in the first round were obtained.

[0059] On this basis, since there are too many parameters involved and it is not convenient for actual operation, only one of the several parameters with correlation is retained to reduce the number of parameters and improve the actual operability. At the same time, it is noted that some parameters have a large amount of missing data, which means that these parameters are difficult to obtain in actual production and are not convenient to use as input for analysis and prediction. Therefore, the parameters with a large amount of missing data are eliminated. In order to reasonably select the input features, the parameters are screened for the second time by calculating the correlation coefficients of the screened parameters. The commonly used correlation coefficients include the Pearson coefficient, the Spearman coefficient, and the Kendall coefficient. The Spearman coefficient can well express the linear correlation between the variables, and finally obtain the parameters screened in the second round.

[0060] The accuracy of the model training obtained by using the parameters selected in the second round as feature input is only about 72%. Through analysis, it is found that the existence of some redundant parameters affects the accuracy of the prediction. Therefore, the parameters are screened for the third round. The importance of the influencing factors of each parameter to the model accuracy is calculated through the random forest algorithm, and the importance is sorted according to the importance. After sorting the results, it can be seen that some parameters are less important in the accuracy of the model. After multiple tests, it is finally determined that the cumulative oil increase, pre-pressure oil production layer, daily liquid production, effective thickness, perforated sandstone thickness layer, water saturation, original saturation pressure, pump diameter, middle depth of the oil layer, fracture pressure, fracture layer, sandstone thickness, production days, permeability, porosity, dynamic liquid level, top depth, pump depth, bottom depth, small block code, layer segment and other 21 influencing factors can achieve the best prediction effect when used as model input. Other parameters can be added as needed.

[0061] Step S02: Preliminary processing is performed on the updated historical data.

[0062] In the present invention, the preliminary processing includes: obtaining the field interpretation table of each main parameter in the historical data, including the data set structure, field meaning and data type; judging whether there is invalid data or empty string in the field interpretation table, and if so, processing by filling, deleting or interpolation method; judging whether there is abnormal value in the field interpretation table, and if so, processing by deletion, replacement or interpolation method; judging whether there is data type error in the field interpretation table, and if so, converting it to the correct data type; judging whether there are duplicate data rows in the field interpretation table, and if so, deleting the duplicate data rows; performing text cleaning on the data in the field interpretation table; judging whether the format of the same type of data in the field interpretation table is consistent, and if not, modifying it to a unified format; normalizing and screening the data in the field interpretation table; verifying the data in the field interpretation table to judge whether it meets the accuracy and completeness requirements, and if not, deleting the data whose accuracy and completeness cannot meet the requirements.

[0063] In an embodiment of the present invention, preliminary processing includes data cleaning, wherein the data cleaning process includes obtaining field explanation tables, identifying and processing missing values, identifying and processing outliers, data type conversion, removing duplicate values, text cleaning, data format unification, data integration and normalization, data screening and subset selection, and data verification.

[0064] Obtain the field explanation table of the main parameters to understand the structure of the dataset, field meanings and data types, as well as possible data problems or anomalies.

[0065] Identify and handle missing values: If there is invalid data (NULL) or empty string in the field interpretation table, handle it by filling, deleting or interpolating.

[0066] Identify and process outliers: Identify outliers in the field explanation table after missing value processing, such as values ​​that are too large or too small. These values ​​may be caused by data collection or input errors and are processed by deletion, replacement, or interpolation.

[0067] Data type conversion to: Identify data type errors in the field interpretation table after outlier processing, and convert the erroneous data to the correct data type, such as converting a string to the corresponding numeric or date data.

[0068] Remove duplicate values: Identify duplicate data rows in the field interpretation table after type conversion, and delete the duplicate data rows in the field interpretation table to avoid repeated calculations or analysis.

[0069] Text cleaning is to clean and standardize the text data in the field interpretation table after removing duplicate values, remove special characters, fix spelling errors, unify uppercase and lowercase, etc.

[0070] Data format unification is to ensure that the same type of data in the field interpretation table after text cleaning has a consistent format in the entire data set, and adjust the same type of data to a unified format, such as unifying the date format, number format or currency format.

[0071] Data integration and normalization include: integrating the data in the field interpretation tables of multiple data sources after the format is unified, and normalizing the data to ensure the consistency and comparability of the data.

[0072] Data screening and subset selection are as follows: in the normalized field explanation table, specific data rows or fields are screened out as needed to meet specific analysis requirements.

[0073] Data validation is to validate the data in the filtered field explanation table to ensure the accuracy and completeness of the data.

[0074] In the present invention, the preliminary processing also includes: linearizing the main parameters in the historical data.

[0075] In the embodiment of the present invention, linearization is to determine the data with linear relationship among all the main parameters. Due to the monotonic relationship between yield and factors (parameters), the yield influencing factors usually show strong linear correlation, i.e., multicollinearity, which affects the yield prediction accuracy.

[0076] In order to clarify the linear relationship between the main parameters (random variables), Spearman correlation analysis was performed. 1 , x 2 , …, x n ) and Y=(y 1 ,y 2 , …, y n ) is:

[0077]

[0078] In the formula, ρ is the rank correlation coefficient, m is the number of samples, and a i and b i x i and i Ranking, a and are the means of random variables X and Y respectively, i=1,2,3...n.

[0079] When 0<ρ≤1, there is a positive correlation, when -1≤ρ<1, there is a negative correlation, and when ρ=0, there is no correlation.

[0080] The data after preliminary processing is divided into a training set and a test set. The training set is used to train the model, and the test set is used to test the model performance.

[0081] Step S03: Divide the historical data after preliminary processing into a training set and a test set.

[0082] Step S04: input the training set data into the neural network model to train the model.

[0083] In the present invention, the neural network model is: a one-dimensional convolutional fully connected neural network model.

[0084] In the embodiment of the present invention, after comparing dozens of algorithms through parameter training, it is determined that a one-dimensional convolutional fully connected neural network (1D Convolutional Neural Network) algorithm can achieve good results.

[0085] Step S05: Test the trained neural network model using the test set data to determine whether the test result meets the requirements. If not, optimize the model until the optimized trained model meets the requirements; use the neural network model that meets the requirements to predict production.

[0086] In an embodiment of the present invention, the training set data divided after preliminary processing is input into the established one-dimensional convolutional fully connected neural network model to train the model; the sample data in the test set is input into the model, and the prediction result is obtained by running it. The prediction result is compared with the label data in the test set to determine the error of the model prediction result. If the error is too large, that is, the error is greater than a predetermined value, it means that the test result does not meet the requirements, and the algorithm parameters of the model need to be adjusted. Step S04 is executed to input the training set data into the neural network model after adjusting the algorithm parameters for re-training until the error of the trained model is less than or equal to the predetermined value, and the model optimization training is completed.

[0087] In the present invention, before using the neural network model that meets the requirements to predict production, it also includes: judging whether the total number of training times of the neural network model is greater than or equal to the predetermined number of times, if not, re-dividing the historical data into training sets and test sets, and inputting them into the neural network model to re-train and test the model; if yes, judging whether the test results of the retrained model meet the requirements, if not, re-preliminarily processing the historical data, and using the training sets and test sets divided by the historical data after the preliminary processing to train and test the neural network model.

[0088] In the present invention, the predetermined number of times is: 10 times.

[0089] In an embodiment of the present invention, after each training, the number of model training times is accumulated, and it is determined whether the total number of training times after accumulation reaches 10 times. If the number of training times is less than 10 times, step S03 is executed to readjust the data, divide the historical data, and input the re-divided training set and test set into the model for training and testing. After the total number of training times reaches 10 times, it is determined whether the test results meet the requirements. If yes, the final neural network model training is completed. If not, step S02 is executed to re-process the historical data, and input the re-preliminary processed historical data into the model to re-train the model.

[0090] When the model training and testing results meet the requirements, the model should be trained no less than 10 times to ensure the accuracy of the model.

[0091] After obtaining the final trained model, the oil well production in the work area is predicted through the model, and the operating wells are selected based on the prediction results.

[0092] The model database regularly obtains updated historical data on oil well production in the work area, and automatically inputs it into the neural network model after preliminary processing to optimize the model training, and then uses the trained model to predict production, that is, re-execute steps S01 to S05, so as to ensure that the model is always in the latest state and the accuracy of each prediction result is the highest.

[0093] The neural network model is integrated into the oil field's well selection platform. The platform automatically performs well selection operations based on the prediction results output by the model each time, thereby improving work efficiency and saving manpower.

[0094] It can be understood that the above-mentioned various method embodiments mentioned in the present invention can be combined with each other to form a combined embodiment without violating the principle logic. Due to space limitations, the present invention will not go into details.

[0095] The execution subject of the oil and water well production prediction method based on the neural network model can be an oil and water well production prediction device based on the neural network model. For example, the oil and water well production prediction method based on the neural network model can be executed by a terminal device or a server or other processing device, wherein the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the oil and water well production prediction method based on the neural network model can be implemented by a processor calling a computer-readable instruction stored in a memory.

[0096] Those skilled in the art will appreciate that, in the above method of specific implementation, the order in which the steps are written does not imply a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of the steps should be determined by their functions and possible internal logic.

[0097] The present invention also provides an oil and water well production prediction device based on a neural network model, comprising: a database and model building unit, used to construct a neural network model and establish a model database connected to the neural network model, the model database is used to obtain historical data updated in a work area in real time or at a predetermined time interval; a data preliminary processing unit, used to perform preliminary processing on the updated historical data; a data division unit, used to divide the preliminary processed historical data into a training set and a test set; a model training unit, used to input the training set data into the neural network model to train the model; a model optimization unit, used to test the trained neural network model through the test set data to determine whether the test result meets the requirements, and if not, optimize the model until the optimized trained model meets the requirements; a production prediction unit, used to use the neural network model that meets the requirements to predict production.

[0098] In some embodiments, the functions or modules and units included in the device provided by the embodiment of the present invention can be used to execute the method described in the above method embodiment. Its specific implementation can refer to the description of the above method embodiment. For the sake of brevity, it will not be repeated here.

[0099] The present invention can bring multiple benefits and functions by integrating a one-dimensional convolutional fully connected neural network model into the well selection platform. First, the model can accurately predict the production of oil and water wells by learning the relationship between production parameters and production information in the historical data of oil wells, which will provide engineers with a more reliable decision-making basis and help them better plan and manage oil and water well resources.

[0100] Secondly, by integrating this model, the well selection platform can realize the automatic production prediction function. Traditional prediction methods usually rely on expert experience and complex model building processes, while neural network models can automatically learn and predict from large amounts of data, greatly reducing the complexity and time cost of manual modeling.

[0101] In addition, the one-dimensional convolutional fully connected neural network model is also highly flexible and adaptable. It can process different types of oil and water well data, including various key parameters such as temperature, pressure, fluid content, etc., to provide comprehensive and accurate prediction results, which enables the well selection platform to meet the needs of different projects and working conditions.

[0102] Finally, by predicting the production of oil and water wells, the well selection platform can help engineers optimize oilfield production. By analyzing the differences between the predicted results and the actual production data, the platform can help find problems and potential bottlenecks in the production process and provide corresponding suggestions and improvement measures to improve the well's productivity and economic benefits.

[0103] In summary, integrating the one-dimensional convolutional fully connected neural network model into the well selection platform can provide users with accurate production prediction, automated modeling process, flexible data processing capabilities, and help optimize oilfield production and improve resource utilization efficiency and economic benefits.

[0104] The embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for predicting oil and water well production based on a neural network model, characterized in that: include: Constructing a neural network model and establishing a model database connected to the neural network model, wherein the model database is used to obtain historical data updated in the work area in real time or at predetermined time intervals; Performing preliminary processing on the updated historical data; Dividing the preliminarily processed historical data into a training set and a test set; Inputting the training set data into the neural network model to train the model; The trained neural network model is tested by the test set data to determine whether the test result meets the requirements. If not, the model is optimized and trained until the optimized and trained model meets the requirements. The neural network model that meets the requirements is used to predict production.

2. The method for predicting oil and water well production based on a neural network model according to claim 1, characterized in that: The historical data include the main parameters related to oil production: Among them, the main parameters include at least: cumulative oil increase, oil production layer before pressure, daily liquid production, effective thickness, perforated sandstone thickness layer, water saturation, original saturation pressure, pump diameter, middle depth of oil layer, fracture pressure, fracture layer, sandstone thickness, production days, permeability, porosity, dynamic liquid level, top depth, pump depth, bottom depth, small block code and layer segment.

3. The oil and water well production prediction method based on a neural network model according to claim 1 is characterized in that: The preliminary processing includes: Obtaining a field explanation table of each main parameter in the historical data, including a data set structure, field meaning, and data type; Determine whether there is invalid data or an empty string in the field interpretation table, and if so, process it by filling, deleting or interpolating; Determine whether there are abnormal values ​​in the field interpretation table, and if yes, process them by deletion, replacement or interpolation; Determine whether there is a data type error in the field interpretation table, and if so, convert it to a correct data type; Determine whether there are duplicate data rows in the field interpretation table, and if yes, delete the duplicate data rows; Perform text cleaning on the data in the field explanation table; Determine whether the format of the same type of data in the field interpretation table is consistent, and if not, modify it to a unified format; Normalizing and screening the data in the field explanation table; The data in the field explanation table is verified to determine whether it meets the accuracy and completeness requirements. If not, the data that does not meet the accuracy and completeness requirements is deleted.

4. The oil and water well production prediction method based on a neural network model according to claim 1 is characterized in that: The preliminary processing also includes: The main parameters in the historical data are linearized.

5. The oil and water well production prediction method based on a neural network model according to any one of claims 1 to 4, characterized in that: Before using the neural network model that meets the requirements to predict production, the method further includes: Determine whether the total number of training times of the neural network model is greater than or equal to a predetermined number of times. If not, re-divide the historical data into a training set and a test set, and input the data into the neural network model to re-train and test the model; If yes, determine whether the test results of the retrained model meet the requirements; if not, re-process the historical data, and use the training set and test set divided by the re-processed historical data to train and test the neural network model.

6. The method for predicting oil and water well production based on a neural network model according to claim 5 is characterized in that: The predetermined number of times is: 10 times.

7. The method for predicting oil and water well production based on a neural network model according to any one of claims 1 to 4 and 6, characterized in that: The neural network model is: a one-dimensional convolutional fully connected neural network model.

8. An oil and water well production prediction device based on a neural network model, characterized in that: include: A database and model building unit, used to build a neural network model and to build a model database connected to the neural network model, wherein the model database is used to obtain historical data updated in the work area in real time or at predetermined time intervals; A data preliminary processing unit, used for performing preliminary processing on the updated historical data; A data division unit, used for dividing the historical data after the preliminary processing into a training set and a test set; A model training unit, used for inputting the training set data into the neural network model to train the model; A model optimization unit is used to test the trained neural network model through the test set data to determine whether the test result meets the requirements. If not, the model is optimized and trained until the optimized and trained model meets the requirements. The yield prediction unit is used to perform yield prediction using the neural network model that meets the requirements.

Images