Oil field development prediction model, optimization method and prediction method of oil field development

Abstract

The embodiment of the invention provides an oil field development prediction model, an optimization method and an oil field development prediction method. The method comprises the steps of obtaining development data of historical oil reservoirs; performing data set processing on the development data to obtain feature data; determining the correlation degree according to the Pearson correlation coefficients of any two in the feature data; optimizing the feature data according to the correlation degree to obtain optimized feature data; substituting the optimized feature data into a random forest model to determine the contribution degree of the optimized feature data to the daily oil production; and adjusting and optimizing the oil field development prediction model according to the contribution degree. According to the method, the prediction precision of oil reservoir development is greatly improved, and the oil reservoir development effect is improved.

Description

Technical Field

[0001] The present invention relates to the fields of oilfield development and artificial intelligence, and particularly relates to an oilfield development prediction model and optimization method, and an oilfield development prediction method. Background Art

[0002] The prediction of reservoir development indicators is a very important and complex task, which needs to comprehensively consider the influence of development factors such as reservoir development history, structural geology, drilling and completion, and well pattern and well type, and belongs to the prediction problem of a typical non-linear complex system. The more factors and variables considered in the prediction model, the higher the prediction accuracy, but the lower the prediction efficiency.

[0003] Currently, the prediction methods of reservoir development indicators include three categories, namely model-driven, data-driven, and dual-combination-driven prediction methods. Among them, the model-driven prediction method requires a definite development law and a perfect prediction model as the basis. The characteristics of this method are clear mechanism and control variables, easy to use, and relatively simple. However, it is too dependent on human experience and local information and is difficult to obtain the global optimal solution. The data-driven prediction method includes artificial intelligence (AI). The artificial neural network is a learning process based on layer-by-layer feature transformation. This method directly mines useful information from the data, but the calculation process is relatively complex, non-explainable, and highly non-linear. The dual-combination-driven prediction method is to predict separately by multiple methods and then take the weighted average as the final prediction value, but the problem of weight selection still needs to be solved. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide an oilfield development prediction model and optimization method, and an oilfield development prediction method, which greatly improve the prediction accuracy of reservoir development and thus improve the effect of reservoir development.

[0005] In order to achieve the above purpose, the embodiments of the present invention provide an optimization method for an oilfield development prediction model, and the method includes:

[0006] Obtain the development data of the historical reservoir;

[0007] Perform dataset processing on the development data to obtain feature data;

[0008] Determine the correlation degree according to the Pearson correlation coefficient between any two of the feature data;

[0009] Optimize the feature data according to the correlation degree to obtain optimized feature data;

[0010] Input the optimized feature data into the random forest model to determine the contribution degree of the optimized feature data to the daily oil production;

[0011] Optimize the oilfield development prediction model according to the contribution degree.

[0012] Optionally, the dataset processing includes interpolation processing, missing value processing, and data normalization processing.

[0013] Optionally, the missing value processing includes: performing autocorrelation analysis on each development data to obtain the time points of data missing;

[0014] performing partial autocorrelation analysis on the development data at this time point to obtain the correlation between any two points in the development data, and the types of the correlation include positive correlation and negative correlation;

[0015] analyzing the development data with the same type of correlation, and filling the data missing according to the analysis value.

[0016] Optionally, the method for obtaining the Pearson correlation coefficient is:

[0017]

[0018] where: ρ(x, y) is the Pearson correlation coefficient between variables x and y,

[0019] x and y are two feature variables,

[0020] σ(x) is the standard variance of variable x,

[0021] cov(x, y) is the covariance of variables x and y.

[0022] Optionally, optimizing the feature data according to the degree of correlation includes:

[0023] sorting the degree of correlation, and removing the feature data with the degree of correlation less than a certain threshold.

[0024] Optionally, the development data is at least one of monthly injection-production ratio, monthly underground deficit, open well rate of old wells, monthly oil increment of old wells, open well rate of new wells, monthly oil production of new wells, natural decline rate, comprehensive decline rate, number of flowing wells in production, number of pumping wells in production, average flowing liquid level, monthly oil production, monthly gas production, monthly oil production time rate, monthly water cut, monthly gas-oil ratio, wellhead oil production rate, wellhead liquid production rate, wellhead production degree, monthly injection volume, and oil price.

[0025] Optionally, the oilfield development prediction model is an LSTM neural network model.

[0026] Optionally, the method further includes

[0027] tuning the oilfield development prediction model according to hyperparameters;

[0028] The hyperparameters are at least one of time step, batch size, number of LSTM network layers, learning rate, number of iterative training times, and dropout rate.

[0029] On the other hand, the present invention also provides a prediction model for oilfield development, which is an oilfield development prediction model optimized according to the method for optimizing the oilfield development prediction model described above.

[0030] On the other hand, the present invention also provides a prediction method for oilfield development. Development data of a reservoir is input into the prediction model for oilfield development described above to obtain development prediction data of the reservoir, and the development prediction data is used to guide subsequent reservoir development.

[0031] The method for optimizing the oilfield development prediction model of the present invention includes: obtaining development data of a historical reservoir; performing data set processing on the development data to obtain feature data; determining the correlation degree according to the Pearson correlation coefficient between any two of the feature data; optimizing the feature data according to the correlation degree to obtain optimized feature data; inputting the optimized feature data into a random forest model to determine the contribution degree of the optimized feature data to the daily oil production; and optimizing the oilfield development prediction model according to the contribution degree. According to the time-varying nature of reservoir development data, the present application timely obtains the changes in development indicators at different time scales, forms a data-driven overall time series prediction process, greatly improves the prediction accuracy of reservoir development, and further improves the effect of reservoir development.

[0032] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. Description of the Drawings

[0033] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification. Together with the following specific implementation, they are used to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0034] Figure 1 is a schematic flow chart of a method for optimizing an oilfield development prediction model of the present invention;

[0035] Figure 2 is a schematic diagram of the correlation analysis between development indicators of the Tahe fracture-vuggy reservoir;

[0036] Figure 3 is a schematic diagram of the analysis of the influence degree of the Tahe oilfield fracture-vuggy reservoir on the daily oil index;

[0037] Figure 4 is a schematic flow chart of the prediction of production indicators of the fracture-vuggy reservoir based on the LSTM model;

[0038] Figure 5 is a schematic diagram of the hyperparameter optimization process of the LSTM prediction model;

[0039] Figure 6It is a schematic diagram of the optimization results of the hyperparameters of the LSTM optimization model;

[0040] Figure 7 It is a schematic diagram of three prediction results of the overall development index - monthly oil production of Tahe fractured-vuggy reservoir;

[0041] Figure 8 It is a schematic diagram of the prediction of the overall development index - monthly oil production index of Tahe fractured-vuggy reservoir;

[0042] Figure 9 It is a schematic diagram of the prediction of the overall development index - comprehensive decline rate of Tahe fractured-vuggy reservoir;

[0043] Figure 10 It is a schematic diagram of the prediction of the overall development index - comprehensive water cut of Tahe fractured-vuggy reservoir;

[0044] Figure 11 It is a schematic diagram of the production capacity change curve within 200 months when 5 connected bodies are put into production simultaneously;

[0045] Figure 12 It is a schematic diagram of the monthly oil production prediction of the seventh area of the fractured-vuggy reservoir in Tahe Oilfield;

[0046] Figure 13 It is a schematic diagram of the comprehensive water cut prediction of the seventh area of the fractured-vuggy reservoir in Tahe Oilfield;

[0047] Figure 14 It is a schematic diagram of the open well rate prediction of the seventh area of the fractured-vuggy reservoir in Tahe Oilfield;

[0048] Figure 15 It is a schematic diagram of the oil production prediction of Well S48 in the fractured-vug unit of Tahe Oilfield;

[0049] Figure 16 It is a schematic diagram of the water cut prediction of Well S48 in the fractured-vug unit of Tahe Oilfield;

[0050] Figure 17 It is a schematic diagram of the oil production prediction of Well S45X in the condensate gas reservoir of the No. 4 fault zone in Shunbei Oilfield. Specific embodiments

[0051] The following further elaborates on the specific embodiments of the embodiments of the present invention with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only used to illustrate and explain the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0052] Embodiment 1

[0053] Figure 1 It is a schematic diagram of the optimization method process of an oilfield development prediction model of the present invention. As Figure 1 shown, an optimization method of an oilfield development prediction model of the present invention includes:

[0054] Step S101 is to obtain the development data of the historical reservoir. Specifically, the development data is at least one of the monthly injection-production ratio, monthly underground deficit, open well rate of old wells, monthly oil increase of old wells, open well rate of new wells, monthly oil production of new wells, natural decline rate, comprehensive decline rate, number of flowing wells, number of pumping wells, average flowing liquid level, monthly oil production, monthly gas production, monthly oil production time rate, monthly water cut, monthly gas-oil ratio, wellhead oil production rate, wellhead liquid production rate, wellhead recovery factor, monthly water injection volume, and oil price.

[0055] Step S102 is to perform dataset processing on the development data to obtain characteristic data. The dataset processing is data cleaning and normalization processing. Specifically, based on the collection of oilfield production dynamic data, the correlation degree analysis of the factors affecting the development indicators is carried out to determine the main control factors, and the data bodies that do not meet the conditions are deleted through data cleaning to complete the dataset processing. The dataset processing includes interpolation processing, missing value processing, and data standardization processing.

[0056] During the development process of deep-ultra-deep fractured-vuggy reservoirs, oil wells often stop flowing and producing due to reservoir factors such as rapid water flooding, water channeling, gas channeling, and insufficient formation energy in the reservoir. In addition, during the production process of oil and water wells, the production of oil and water wells is often interrupted due to engineering technology factors such as wellbore treatment of oil and water wells, bottom-hole operations and measures of oil and water wells, and ground pipeline maintenance. As a result, data gaps are shown in the collected production dynamic data table. Compared with sandstone reservoirs, the fractured-vuggy system has more development influencing factors and poor system operation stability. The degree of data gaps in the production dynamic data of oil and water wells in fractured-vuggy reservoirs is much higher than that of conventional sandstone reservoirs. Therefore, it is necessary to specifically process the data gap problem of fractured-vuggy reservoirs. For missing values, the general processing method is to perform data smoothing, find the time variation law of the data itself, and perform fitting processing and supplementation on the missing data. For deep-ultra-deep fractured-vuggy reservoirs, due to the high frequency and large amplitude of data changes, the data smoothing method often cannot obtain regular results, so data filling cannot be performed.

[0057] The present invention proposes to perform missing value processing on the data, specifically including: performing autocorrelation analysis on each development data to obtain the time points of data gaps; performing partial autocorrelation analysis on the development data at these time points to obtain the correlation between any two points in the development data, and the types of the correlation include positive correlation and negative correlation; analyzing the development data with the same type of correlation, and filling the data gaps according to the analysis values. Among them, partial autocorrelation measures the correlation between a time series and its lagged version under the influence of controlling other lag terms.

[0058] According to a specific embodiment, the process of filling missing value data in the present invention includes four steps: The first step: perform autocorrelation analysis on each variable. By establishing the autocorrelation function (ACF) of a certain characteristic quantity changing with time, determine the correlation of a specific time series with itself at that lag moment (that is, whether there is a correlation between any two points with a fixed distance, and finally obtain at what time points there are missing values). This method can be implemented for both the case of changing according to a fixed function and the case of random change.

[0059] The second step: determination of relevant variables. After determining the time points of data missing, it is also necessary to determine which are the relevant variables. This method is partial autocorrelation function (PACF) analysis, and the purpose is to determine the correlation between two characteristic variables under specific values in the same system. This method does not find the correlation between the lag and the current like ACF, but finds the correlation between the residual and the next lag value.

[0060] The third step: analysis of the directionality of correlation. There are two situations of positive correlation and negative correlation between the correlations of two variables. Therefore, it is necessary to classify the directionality of the correlation between variables to achieve the analysis of the directionality of correlation. According to the calculation result of the correlation coefficient, if it is greater than 0, it is positive correlation, otherwise it is negative correlation, and 0 means no correlation. Therefore, before determining the size of the missing value, it is necessary to analyze according to positive correlation, negative correlation, and no correlation, and consider the directionality of the correlation between variables in the later big data prediction model, and perform quantitative calculation of the Pearson coefficient for variables with the same type of correlation direction.

[0061] The fourth step: determination of the size of the missing value. To determine the size of the data value, the knn nearest neighbor algorithm is used for filling. For each data point with a missing value, calculate its distance from other data points, and find the K nearest data points to it. These data points will be the estimates of the missing value.

[0062] Such as Figure 4As shown, the standardization process first calculates the mean and standard deviation of each feature, and transforms the value of each feature so that it is divided by the standard deviation after subtracting the mean, which ensures that the data features are on the same scale. The processed data is divided into a training set and a test set according to a certain time ratio. The training set is used to train the model, and the test set is used to test the model performance; a suitable time-series data set is selected as the training set, the initial values of the hyperparameters are set, and the training set data is introduced into the model for training and learning. The present invention preferably uses the Adam gradient descent algorithm to perform backpropagation and update on the model parameters to train the final model; the performance of the trained model is verified with the test set, and the prediction results of the test set are given. When the prediction results of the test set do not meet the requirements, the hyperparameter values need to be reset and the model is trained again until the prediction accuracy requirements are met.

[0063] Step S103 is to determine the degree of correlation according to the Pearson correlation coefficient between any two of the feature data. The Pearson correlation coefficient is used to characterize the change trend and degree of correlation between two variables, and its value range is [-1, 1]. When it is close to 1, it indicates that the two have a positive correlation; when it is close to -1, it indicates a negative correlation; and when the value is close to 0, it indicates a low correlation.

[0064] Specifically, the method for obtaining the Pearson correlation coefficient is as follows:

[0065]

[0066] Among them: ρ(x, y) is the Pearson correlation coefficient between variables x and y, x and y are two feature variables, σ(x) is the standard variance of variable x, and cov(x, y) is the covariance of variables x and y.

[0067] Specifically, the classification boundaries of the Pearson correlation degree between variables x and y are shown in Table 1 below:

[0068] Table 1:

[0069]

[0070] Step S104 is to optimize the feature data according to the degree of correlation to obtain optimized feature data. According to a specific implementation manner, the optimizing the feature data according to the degree of correlation includes: sorting the degree of correlation and removing the feature data with a degree of correlation less than a certain threshold.

[0071] According to a specific implementation manner, such as Figure 2As shown, based on data cleaning and normalization, big data analysis and prediction are carried out on twenty overall development indicators of the oilfield collected as feature variables, including monthly injection-production ratio, monthly underground deficit, production rate of old wells, monthly oil increment of old wells, production rate of new wells, monthly oil production of new wells, natural decline rate, comprehensive decline rate, number of flowing wells in production, number of pumping wells in production, average flowing fluid level, monthly oil production, monthly gas production, monthly oil production time rate, monthly water cut, monthly gas-oil ratio, wellhead oil production rate, wellhead fluid production rate, wellhead recovery factor, monthly injection volume, oil price. The Pearson correlation coefficient analysis method is used to analyze the correlation between them, and features with a correlation less than 0.2 are removed from the analysis results, such as the three features with relatively weak correlations: production rate of new wells, monthly gas-oil ratio, and wellhead fluid production rate.

[0072] Step S105 is to input the optimized feature data into the random forest model to determine the contribution degree of the optimized feature data to the daily oil production.

[0073] According to a specific implementation manner, the remaining 17 feature variables related to the above-mentioned oilfield development effect are input into the random forest model for ranking the influence degree on the daily oil production (as Figure 3 shown), so as to determine which features have a greater contribution to the target variable and obtain the main controlling factors affecting the daily oil production capacity of the deep-ultra-deep fractured-vuggy reservoir. The results show that the feature variables such as wellhead oil production rate, monthly oil production time rate, average flowing fluid level, and monthly water cut have the greatest influence degree and relatively high importance, and can be used as the input features of the future oilfield overall development index prediction model. This method greatly improves the accuracy and reliability of the prediction model.

[0074] Step S106 is to optimize the oilfield development prediction model according to the contribution degree.

[0075] According to a specific implementation manner, the parameters in the model training process are combinations that control the structure of the training model, and each model parameter is different. Usually, the parameters of the model are called hyperparameters. In the process of hyperparameter tuning, the optimization problem is formalized as a minimization problem of the objective function, where the objective function is usually the performance index of the machine learning model (such as the accuracy of cross-validation or the mean squared error). According to the results of previous experiments, the next test point is adaptively selected to search for the most promising hyperparameter configuration in the search space. This method helps to improve the performance of the model and speeds up the model development process.

[0076] As Figure 5As shown, in order to ensure the optimized performance of the training model, five hyperparameters, namely time step (timestemp), batch size (batch_size), number of LSTM network layers (layers), learning rate (lr), number of iterative training epochs (epoches), and dropout rate, are selected for tuning. The present invention constructs an LSTM model and introduces the Optuna framework, with the goal of minimizing the loss error of the test set as the optimization objective. Search for the optimal combination of hyperparameters within the range of hyperparameters, and input the processed dataset into the model for training. Through the Optuna framework, repeatedly train the model to continuously search for the optimal combination of hyperparameters, and obtain a prediction model with the best performance (the smallest loss error of the test set) on the test set.

[0077] Analyze the results of repeated training of the analysis framework and the influence of each hyperparameter on the training results, and rank the importance of the influence degree of hyperparameters on the training results. As Figure 6 shown, the number of training epochs (epoches) has the highest impact on the accuracy of the model, followed by the time step (timestemp) and the hidden layer size. This initially obtains the hyperparameter number range for model training for subsequent model calculations. From the search process of each parameter value of this model, it can be seen that the value of the number of training epochs (epoches) is around 500 with a small error. Usually, the value of the hidden layer size (hiddenlayer) is between 200 and 300, and the value of the time step (timestemp) is between 20 and 30. The specific numerical size is related to the prediction time scale (year, month, day).

[0078] Example 2

[0079] Example 2 is for predicting the development indicators of the fractured-vuggy carbonate reservoir in the S Oilfield. The reservoir depth of the S Oilfield is 5000m - 7000m, belonging to a deep to ultra-deep reservoir. The reservoir type is a fractured-vuggy reservoir, and the development started in 1993. So far, the cumulative oil production has reached more than 100 million tons. Using the prediction method provided by the present invention, the overall development indicators of the fractured-vuggy reservoir in the S Oilfield are predicted, including monthly production, water cut, cumulative oil production, well opening rate, decline rate, etc. The annual oil production and decline rate for the next year are predicted, providing a basis for the oilfield development deployment in the coming year.

[0080] As Figure 7As shown, during the prediction process, the production data before December 2021 in the data was used as the training set, and the data from then until March 2023 was used as the test set for prediction. Finally, this LSTM neural network prediction method was compared with the traditional production decline method and the BP neural network. According to convention, the three methods were evaluated using the evaluation indicators of MSE, MAE, and MAPE (Table 2). The results show that although the traditional production decline method has advantages in simplicity and ease of use, its accuracy is significantly insufficient. The LSTM method has better time series data modeling capabilities and can better capture long-term dependencies in the time series.

[0081] Table 2:

[0082]

[0083] In summary, it can be seen that the method of this application can achieve rapid prediction of the monthly production indicators of the entire oilfield or block in deep - ultra - deep fracture - cave oil reservoirs, so it can be used to guide the overall development deployment and comprehensive management of the oilfield or block.

[0084] Oil production prediction is one of the important tasks in oilfield development prediction and is also the main basis for reservoir development deployment. The method of this invention was used to predict the monthly oil production and the comprehensive decline rate of monthly oil production in the Tahe fracture - cave reservoir. The prediction results are as Figure 8 and Figure 9 shown, where the blue curve is the actual production curve and the orange curve is the optimized prediction curve. It can be seen that the prediction results of this application are very good.

[0085] Water production prediction is another important task in the reservoir. As Figure 10 shown, the main index of this prediction is the prediction of the comprehensive water cut at different development time scales, and its prediction is of great significance for specifying the workload of reasonable treatment measures.

[0086] Figure 11 The recovery factor of the deep - ultra - deep fracture - cave reservoir was predicted using the model of this invention, which belongs to the main index of oilfield development and operation benefits.

[0087] Figure 12 、 13 and Figure 14 This invention was used for the prediction of the development indicators in the seventh district of the Tahe Oilfield, including monthly oil production capacity, comprehensive water cut, and well - opening rate. It can be seen from the figure that the change range of the well - opening rate is very large.

[0088] Well S48 in the oilfield is located in the fourth district with a weathered crust karst geological background, and the single - well controlled reserves reach 3.3 million tons. This invention can use the daily production data of this well to carry out production prediction, including oil production, comprehensive water cut, etc. (as Figure 15 andFigure 16 as shown

[0089] The results show that the method of the present invention can not only predict the overall development indexes of the oilfield and the development indexes of blocks, but also predict the production data of individual wells in deep and ultra-deep fractured-vuggy reservoirs; it can not only carry out index prediction on the annual and monthly time scales, but also carry out index prediction on the single-day time scale.

[0090] such as Figure 17 As shown, the geological reserves in the second area of the Shunbei Oilfield reach 32 million tons, and the proven natural gas reserves reach 110 billion cubic meters. The method provided by the present invention is used to predict the daily oil production in this area. The results show that the method of the present invention can realize the prediction of the development indexes of production wells in ultra-deep condensate gas reservoirs.

[0091] In summary, the present invention uses the optimized neural network model LSTM and the unoptimized model to predict the monthly oil production respectively. The prediction results show that both of them show low errors on the training set, but on the test set, the optimized model (i.e., the model obtained in this application) shows better prediction performance than the unoptimized model. The intelligent optimization model of neural network adopted in this application shows higher prediction accuracy and better generalization ability on the test set, fully considering the historicity and time-variability of the development indexes of fractured-vuggy reservoirs, reflecting the changes of development indexes under different time scales, improving the accuracy and speed of index prediction, and making up for the problems of low prediction accuracy and poor effect caused by the great difficulty of mathematical modeling of fractured-vuggy reservoirs, and has strong application value.

[0092] On the other hand, the present invention also provides a prediction model for oilfield development, which is an oilfield development prediction model optimized according to the method for optimizing the oilfield development prediction model described above. This model can be used to improve the prediction accuracy and coverage of the overall development indexes of the reservoir.

[0093] On the other hand, the present invention also provides a prediction method for oilfield development. The development data of the reservoir is brought into the above-mentioned oilfield development prediction model to obtain the development prediction data of the reservoir, and the development prediction data is used to guide the subsequent reservoir development. This method gives full play to the advantages of rapid prediction of time series problems, timely reflects the changes of development indexes under different time scales, improves the prediction accuracy and coverage of the overall development indexes of the reservoir, greatly improves the real-time prediction effect of the oilfield, and has certain application prospects.

[0094] The optimization method of the oilfield development prediction model of the present invention includes: obtaining the development data of historical reservoirs; performing dataset processing on the development data to obtain feature data; determining the correlation degree according to the Pearson correlation coefficient between any two of the feature data; optimizing the feature data according to the correlation degree to obtain optimized feature data; bringing the optimized feature data into a random forest model to determine the contribution degree of the optimized feature data to the daily oil production; and optimizing the oilfield development prediction model according to the contribution degree. According to the time-varying nature of reservoir development data, this application timely obtains the changes in development indicators at different time scales, forms a data-driven overall time series prediction process, greatly improves the prediction accuracy of reservoir development, and thus improves the effect of reservoir development.

[0095] The optional implementation manners of the embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details in the above implementation manners. Within the technical concept scope of the embodiments of the present invention, various simple variants can be made to the technical solutions of the embodiments of the present invention, and these simple variants all belong to the protection scope of the embodiments of the present invention.

[0096] In addition, it should be noted that, among the various specific technical features described in the above specific implementation manners, they can be combined in any suitable manner without conflict. To avoid unnecessary repetition, the embodiments of the present invention will not separately describe various possible combination manners.

[0097] Those skilled in the art can understand that all or part of the steps of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions for enabling a single-chip microcomputer, a chip or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disc that can store program codes.

[0098] In addition, any combination can be made between various different implementation manners of the embodiments of the present invention, as long as it does not violate the idea of the embodiments of the present invention, and it should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. An optimization method for an oilfield development prediction model, characterized in that, the method comprises: obtaining the development data of historical reservoirs; performing dataset processing on the development data to obtain feature data; determining the correlation degree according to the Pearson correlation coefficient between any two of the feature data; optimizing the feature data according to the correlation degree to obtain optimized feature data; bringing the optimized feature data into a random forest model to determine the contribution degree of the optimized feature data to the daily oil production; tuning the oilfield development prediction model according to the contribution degree.

2. The method according to claim 1, characterized in that, the dataset processing includes interpolation processing, missing value processing and data standardization processing.

3. The method according to claim 2, characterized in that, the missing value processing includes: performing autocorrelation analysis on each development data to obtain the time points of data missing; performing partial autocorrelation analysis on the development data at this time point to obtain the correlation between any two points in the development data, and the types of the correlation include positive correlation and negative correlation; analyzing the development data with the same type of correlation, and filling the data missing according to the analysis value.

4. The method according to claim 1, characterized in that, the method for obtaining the Pearson correlation coefficient is: where: ρ(x, y) is the Pearson correlation coefficient between variables x and y, x, y are two feature variables, σ(x) is the standard variance of variable x, cov(x, y) is the covariance of variables x and y.

5. The method according to claim 1, characterized in that, optimizing the feature data according to the correlation degree includes: sorting the correlation degree, and removing the feature data with the correlation degree less than a certain threshold.

6. The method according to claim 1, characterized in that, the development data is at least one of the monthly ratio of injection-production ratio, monthly deficit of underground deficit, open well rate of old wells, monthly oil increase of old wells, open well rate of new wells, monthly oil of new wells, natural decline rate, comprehensive decline rate, number of flowing wells in production, number of pumping wells in production, average flowing fluid level, monthly oil volume, monthly gas volume, monthly oil production time rate, monthly water cut, monthly gas-oil ratio, wellhead oil production rate, wellhead fluid production rate, wellhead recovery factor, monthly injection volume and oil price.

7. The method according to claim 1, characterized in that, the oilfield development prediction model is an LSTM neural network model.

8. The method according to claim 1, characterized in that, the method further includes tuning the oilfield development prediction model according to hyperparameters; the hyperparameters are at least one of time step, batch size, number of LSTM network layers, learning rate, number of iterative training times and dropout rate.

9. An oilfield development prediction model, characterized in that, it is an oilfield development prediction model optimized by the optimization method of the oilfield development prediction model according to any one of claims 1-8 above.

10. A prediction method for oilfield development, characterized in that, bringing the development data of the reservoir into the oilfield development prediction model in claim 9 above to obtain the development prediction data of the reservoir, and the development prediction data is used to guide the subsequent reservoir development.

Images