Shale gas single well daily production prediction method and system
By employing the LSTM algorithm and cross-validation with typical curves in shale gas well areas, the accuracy and adaptability issues of single-well production prediction for shale gas have been resolved, achieving more efficient production prediction, which is particularly suitable for situations lacking geological static parameters.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PETROCHINA CO LTD
- Filing Date
- 2022-06-30
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies suffer from insufficient adaptability and low accuracy in predicting single-well production of shale gas. In particular, the accuracy of empirical algorithms from conventional oil and gas production methods in shale gas production remains to be observed under different geographical and geological conditions.
A single-well daily production prediction model is constructed by cross-validating the deep learning algorithm LSTM with typical curves of shale gas well areas. By establishing typical daily gas production curves of well areas and LSTM models, and combining the time-series attention mechanism to optimize the weight coefficients, the prediction accuracy is improved.
It improves the accuracy and adaptability of shale gas single-well daily production prediction, especially in the absence of geological static parameters, reduces labor costs, has high application value, and its short-term prediction effect is better than that of traditional power functions.
Smart Images
Figure CN117365430B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of shale gas exploration and development technology, and in particular to a method and system for predicting the daily production of a single shale gas well. Background Technology
[0002] In recent years, China's shale gas extraction industry has flourished. Major oil and gas companies have invested significant manpower and capital in shale gas exploration and development. Particularly in information technology infrastructure, for example, dynamic databases and related management systems have been established at production sites, meticulously recording the daily production status (gas production, water production, pressure, etc.) of each shale gas well. How to integrate new IT technologies and approaches to deepen the application of this data and improve management and research capabilities is a challenge facing these oil and gas companies.
[0003] Shale gas single-well production prediction is a task of great concern to management and research personnel. It serves as the basis for single-well productivity evaluation and an important basis for production planning and tracking. As an emerging unconventional energy source, shale gas research methods mainly draw upon and evolve from conventional oil and gas research and development methods.
[0004] Currently, there are numerous research methods and related software tools in China for gas reservoir dynamic analysis. Most are based on "gas reservoir engineering theory" and "dynamic analysis procedures," integrating gas reservoir description, dynamic analysis of gas reservoir production, gas reservoir engineering calculations, and production capacity evaluation. Algorithms for gas well production prediction mainly include analytical model methods, production decline methods, material balance methods, and probabilistic methods. In practical applications, these methods have achieved good results under certain regional and data conditions, but they also have significant limitations. For example, requiring too many parameters reduces the method's adaptability, and different regions and geological conditions can lead to deviations in empirical algorithms and chart parameters. In particular, most of these algorithms are accumulated from conventional oil and gas production experience, and their accuracy in shale gas production remains to be observed.
[0005] In recent years, data-driven deep learning algorithms have gradually demonstrated significant effectiveness in the field of oil and gas energy extraction. Therefore, applying deep learning algorithms to shale gas single-well production prediction, as a supplementary and verification method, has great application value.
[0006] LSTM (Long Short-Term Memory) is a type of temporal recurrent neural network specifically designed to address the long-term dependency problem inherent in general RNNs (Recurrent Neural Networks). LSTM controls the transmission state through gating, remembering information that requires long-term memory while forgetting less important information. It is suitable for processing and predicting important events with very long intervals and delays in time series (such as the continuous daily production of a single well).
[0007] Typical curves are curves with certain typical characteristics. For example, shale gas production wells in the same well area have similar geological conditions, similar engineering implementation parameters, and similar production parameters. Moreover, the wells are completely disconnected from each other. Therefore, we can assume that the production trends of each well are similar. After a sufficient number of wells have been put into operation in the well area, the daily production data of these wells can be used to calculate the typical values for each day using a probability method, and a continuous typical production curve can be plotted. This typical curve can be used as a reference when predicting the production of wells that have not yet been put into operation. Summary of the Invention
[0008] The purpose of this invention is to provide a method and system for predicting the daily production of a single shale gas well. This invention is entirely based on the characteristics of shale gas extraction, primarily employing a deep learning algorithm (LSTM algorithm) and using typical curves that fully reflect actual production characteristics for cross-validation. This invention can effectively solve the problem of predicting the daily production of a single well. To achieve the above objective, this invention provides the following technical solution:
[0009] A method for predicting the daily production of a single shale gas well, the method comprising the following steps:
[0010] Based on the total daily gas production data of all wells in the shale gas well area, a typical daily gas production curve for the well area is established.
[0011] An LSTM model is constructed based on the daily production data of a single well in the shale gas well area.
[0012] By combining the typical daily gas production curve of the well area and the LSTM model, a prediction curve for the daily gas production of a single well is established.
[0013] Preferably, the shale gas well area includes shale gas well areas with natural production and shale gas well areas with controlled production.
[0014] Preferably, establishing a typical daily gas production curve for the well area includes the following steps:
[0015] Obtain all daily gas production data for all wells in shale gas well areas using natural production methods or controlled production methods;
[0016] Remove records with a daily gas production of 0;
[0017] Sort the daily gas production data of each well by date and label them with sequence numbers 1, 2...N;
[0018] The daily gas production value of the same sequence number in all wells is statistically analyzed to obtain the daily gas production statistics corresponding to each sequence number;
[0019] Based on the daily gas production statistics corresponding to each sequence number, a typical daily gas production curve for the well area is constructed.
[0020] Preferably, the step of statistically analyzing the daily gas production values for the same sequence number in all wells includes,
[0021] The daily gas production values for the same sequence number in all wells were statistically analyzed using the average value, P50, or M50 statistical algorithms; among them,
[0022] P50 represents the middle daily gas production value after all wells with the same sequence number are sorted from low to high.
[0023] M50 represents the average daily gas production value from the 25th percentile to the 75th percentile after sorting all wells with the same sequence number from low to high.
[0024] Preferably, the daily production data of a single well in the shale gas well area includes,
[0025] Wells with fewer than 30 days of production data records, and wells with 30 or more days of production data records; among them,
[0026] For single wells with fewer than 30 days of production data records, the value of M50 is used as the daily production data to predict the daily production data of the single well.
[0027] For single wells with more than or equal to 30 days of production data records, the actual production data or daily production data of a certain year is taken, and records with a daily production data of 0 are removed to construct an LSTM model.
[0028] The production day data includes one or more of the following: production date, daily gas production, daily water production, and average casing pressure.
[0029] Preferably, the construction of the LSTM model includes,
[0030] Normalize the daily production data of individual wells in the shale gas well area;
[0031] For the normalized production day data of a single well, divide the data into a training set and a test set;
[0032] Based on the training and test sets of data, an LSTM model is constructed.
[0033] Preferably, the step of constructing an LSTM model based on the training and test data sets includes:
[0034] The LSTM neural network processes the parameters of the training data set to construct the LSTM model;
[0035] Based on the test data set, the LSTM model is validated and analyzed.
[0036] Based on the validation analysis results, the LSTM model is evaluated.
[0037] Preferably, the LSTM neural network processes the parameters of the training data set to construct the LSTM model, including:
[0038] The LSTM neural network combines optimization algorithms and loss functions to dynamically adjust the output parameters of each input data training set parameter. After multiple iterations, the LSTM model is constructed.
[0039] Preferably, the validation analysis of the LSTM model based on the test dataset further includes the following steps:
[0040] Based on the typical curves of the shale gas well area, the daily gas production in the LSTM model is reassigned using a time-series attention mechanism.
[0041] Preferably, the step of redistributing the weight coefficients of the daily gas production in the LSTM model by combining typical curves of the shale gas well area with a time-series attention mechanism includes the following steps:
[0042] Calculate the sequence values N1 to NT for the "Nth day of production" corresponding to dates r1 to rT, where r1 represents the first date and rT represents the Tth date.
[0043] Take the predicted production day data from points N1 to NT on the typical curve;
[0044] The predicted production day data from points N1 to NT are fitted using a power function to obtain the fitted curve;
[0045] The absolute value of the vertical distance between the actual or predicted production day data value and the fitted curve for each point from r1 to rT is used to determine the weighting coefficients.
[0046] The larger the vertical distance, the smaller the weighting coefficient; the smaller the vertical distance, the larger the weighting coefficient.
[0047] Preferably, validation analysis of the LSTM model based on the test dataset includes:
[0048] The prediction error is calculated based on the predicted daily gas production data and the data test set.
[0049] Based on the prediction error results, the LSTM model is judged: the smaller the prediction error, the more accurate the LSTM model.
[0050] The formula for calculating the prediction error, based on the predicted daily gas production data and the data test set, is as follows:
[0051] In the formula, RMSE represents the root mean square error of the prediction, n represents the number of prediction date points, i represents the prediction date point number; ai represents the actual daily gas production value of the i-th date point, and pi represents the predicted daily gas production value of the i-th date point.
[0052] Preferably, the step of establishing a single-well daily gas production prediction curve by combining typical daily gas production curves of the well area and an LSTM model includes the following steps:
[0053] The predicted daily gas production is calculated based on the LSTM model, and the corresponding "Nth day of production" is marked for each value.
[0054] Obtain the typical value for day N from the typical curve based on "day N after production".
[0055] The predicted value is corrected based on ±10% of the typical value, and controlled within the normal value range.
[0056] Based on the corrected predicted values, a single-well production prediction curve is constructed.
[0057] A shale gas single-well daily production prediction system, the system comprising,
[0058] The typical curve establishment unit is used to establish the typical daily gas production curve of the well area based on the daily gas production data of all wells in the shale gas well area;
[0059] The model building unit is used to build an LSTM model based on the daily production data of a single well in the shale gas well area.
[0060] The prediction curve establishment unit is used to combine the typical daily gas production curve of the well area with the LSTM model to establish the daily gas production prediction curve of a single well.
[0061] Preferably, the typical curve establishment unit is used to establish the typical daily gas production curve of the well area, including the following steps:
[0062] Typical curve establishment unit obtains all daily gas production data of all wells in shale gas well areas with natural production mode or controlled production mode;
[0063] Remove records with a daily gas production of 0;
[0064] Sort the daily gas production data of each well by date and label them with sequence numbers 1, 2...N;
[0065] The daily gas production value of the same sequence number in all wells is statistically analyzed to obtain the daily gas production statistics corresponding to each sequence number;
[0066] Based on the daily gas production statistics corresponding to each sequence number, a typical daily gas production curve for the well area is established.
[0067] Preferably, the construction of the LSTM model includes,
[0068] The model building unit normalizes the daily production data of a single well in the shale gas well area;
[0069] The model building unit divides the normalized daily production data of a single well into a training set and a test set.
[0070] The model building unit constructs an LSTM model based on the training and test data sets.
[0071] Preferably, the step of constructing an LSTM model based on the training and test data sets includes:
[0072] The model building unit, the LSTM neural network, processes the parameters of the training data set to construct the LSTM model;
[0073] The model building unit uses the test dataset to validate and analyze the LSTM model;
[0074] The model building unit evaluates the LSTM model based on the validation analysis results.
[0075] The technical effects and advantages of this invention are as follows:
[0076] This innovative technical approach proposes a method for predicting the daily production of shale gas wells based on dual verification using Long Short-Term Memory (LSTM) networks and typical curves. This method can improve the research on shale gas well production prediction based on data-driven approaches. It is particularly suitable for situations where static geological parameters are lacking, has lower manpower costs and higher application value, and is also one of the concrete implementations of artificial intelligence (deep learning) in unconventional oil and gas resources.
[0077] LSTM leverages the advantages of deep learning and data-driven approaches, while the typical gas production curve demonstrates the genetic advantage of the production characteristics of the research target. Cross-validation of the two methods allows for more accurate prediction of single-well production, exhibiting the strongest adaptability and minimal error propagation. This method was tested based on existing dynamic data from shale gas fields. Test results show that compared to commonly used power function production prediction methods, this method achieves similar long-term predictions over 300 days and outperforms the power function in short-term predictions within 30 days.
[0078] This invention proposes a method for predicting the daily production of shale gas single wells using a dual-validation approach combining a Long Short-Term Memory (LSTM) network and typical curves. The method is entirely tailored to the characteristics of shale gas extraction, primarily employing a deep learning algorithm (LSTM) and cross-validating it with typical curves that fully reflect actual production characteristics. This invention effectively solves the problem of predicting the daily production of single wells.
[0079] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description
[0080] Figure 1 A flowchart of the method for predicting the daily production of a single shale gas well is shown.
[0081] Figure 2 The figure shows a typical daily gas production curve of a certain well area under natural production.
[0082] Figure 3 The figure shows a typical daily gas production curve for a certain well area under a controlled production method.
[0083] Figure 4 The LSTM model architecture diagram is shown;
[0084] Figure 5 A schematic diagram of the temporal attention mechanism in the prior art is shown;
[0085] Figure 6 The graph shows the absolute values of the vertical distance between the predicted yield value and the fitted curve, with the weighting coefficients as the plot.
[0086] Figure 7 The single-well production prediction curve is shown. Detailed Implementation
[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0088] To address the shortcomings of existing technologies, this invention discloses a method for predicting the daily production of a single shale gas well. The method includes the following steps: establishing a typical daily gas production curve for the well area based on the total daily gas production data of all wells in the shale gas well area; constructing an LSTM model to predict the daily gas production of a single well based on the daily production data of a single well in the shale gas well area; and establishing a prediction curve for the daily gas production of a single well by combining the typical daily gas production curve of the well area and the LSTM model to predict the daily gas production of a single well.
[0089] Furthermore, such as Figure 1As shown, this invention first selects a well area. On one hand, it calculates and plots typical production curves for all daily gas production data of all wells in the well area. On the other hand, it selects a single well to be predicted in the shale gas well area and determines whether the well has been in production for more than 30 days. For single wells with fewer than 30 days of production data records, the typical curve is used to predict production directly, and the value of M50 is taken as the daily gas production value. For single wells with more than or equal to 30 days of production data records, the actual daily production data or daily production data of a certain year is used. The data is preprocessed to obtain a data training set and a data test set. Then, an LSTM model is constructed based on the data training set. After the LSTM model is validated and evaluated using the data test set, a single-well production prediction model is obtained. The single-well production prediction model, combined with the typical production curve, realizes the daily production prediction of a single well.
[0090] Furthermore, the shale gas well area includes shale gas well areas with natural production and shale gas well areas with controlled production.
[0091] In one embodiment of the present invention, a well area (including several wells already in production) is first selected for production analysis. Due to different production methods, it is necessary to manually determine in advance whether the well area is a "natural production method" or a "controlled production method", or both. Then, according to different production methods, the corresponding shale gas production wells are selected, all production data are obtained, and typical curves are calculated.
[0092] In one embodiment of the present invention, the step of plotting a typical daily gas production curve for a well area includes the following steps: obtaining all daily gas production data of all wells in a shale gas well area with natural production or controlled production; removing records with a daily gas production of 0; sorting the daily gas production data of each well by date and marking them with sequence numbers 1, 2...N; statistically analyzing the daily gas production values of the same sequence number in all wells to obtain the daily gas production statistics corresponding to each sequence number; and constructing a typical daily gas production curve for the well area based on the daily gas production statistics corresponding to each sequence number.
[0093] In one embodiment of the present invention, the step of statistically analyzing the daily gas production values of the same sequence number in all wells includes using an average value, P50, or M50 statistical algorithm to statistically analyze the daily gas production values of the same sequence number in all wells; wherein, P50 represents the middle daily gas production value after sorting the daily gas production values of the same sequence number in all wells from low to high, and M50 represents the average value of the daily gas production values of the same sequence number in all wells from low to high, from the 25th to the 75th percentile.
[0094] Figure 2 The figure shows a typical daily gas production curve of a certain well area under natural production. Figure 3This diagram shows a typical daily gas production curve for a certain well area under a controlled production method. The horizontal axis represents "Day N of production," which is the same sequence number for all wells mentioned above. Figure 2 The horizontal axis is marked "33," indicating that the daily gas production of the 33rd well is calculated using M50, and the calculated value is the Y-axis value. Figure 2 It can be observed that under natural production conditions, most wells reach their peak daily gas production on the 33rd day. Through... Figure 3 It can be observed that under the production control method, most wells reach their peak production on the 60th day, and then the overall production tends to stabilize due to the production control measures.
[0095] In one embodiment of the present invention, when selecting a well for production prediction and its production data, the system first determines whether the well has been in operation for more than 30 days (number of production data records > 30). If it has been in operation for less than 30 days, subsequent prediction using LSTM will not be accurate. For such wells, the data from the "typical curve" (taking the M50 value as the daily gas production value) is used directly. For single wells with more than or equal to 30 production data records, the actual production data or the daily production data for a certain year is used, and records with a daily gas production of 0 are removed. The daily production data includes one or more of the following: production date, daily gas production, daily water production, and average casing pressure. As shown in Table 1, Table 1 shows the daily production data of well W1:
[0096] Table 1 Daily production data of Well W1
[0097]
[0098] As shown in Table 1, the daily production data for Well W1 was recorded starting from September 30, 2021. During the recording process, if the values of "daily water production" or "average casing pressure" were empty, they could be filled with the corresponding values from the previous date.
[0099] The construction of an LSTM model for predicting daily gas production from a single well includes: normalizing the daily production data of a single well in a shale gas well area; dividing the normalized daily production data into a training set and a test set; processing the parameters of the training set using an LSTM neural network to construct an LSTM model; validating the LSTM model based on the test set; and evaluating the model based on the validation results. The LSTM neural network's processing of the training set parameters to construct the training LSTM model involves dynamically adjusting the output parameters of each input training set parameter using an optimization algorithm and a loss function, iterating multiple times to construct the LSTM model. The prediction error is calculated based on the predicted daily gas production data and the test set; the LSTM model is then evaluated based on the prediction error: the smaller the prediction error, the more accurate the LSTM model.
[0100] In one embodiment of the present invention, the LSTM model uses a multi-parameter input (daily gas production, daily water production, average casing pressure) and single-parameter output (daily gas production), therefore parameter normalization is required. This method employs a min-max normalization approach, with the transformation function as follows:
[0101]
[0102] In the formula, X* represents the transformation function, min represents the minimum actual daily gas production, max represents the maximum actual daily gas production, and x represents the actual daily gas production for each day.
[0103] As shown in Table 2, taking "daily gas production" as an example, 4210 on October 19, 2021, corresponds to the minimum actual daily gas production represented by 'min' in the min-max conversion function; 12670 on October 6, 2021, corresponds to the maximum actual daily gas production represented by 'max' in the min-max conversion function; and 10730 on September 30, 2021, corresponds to the actual daily gas production represented by 'x' for each day in the min-max conversion function. For example, the normalized daily gas production conversion function for September 30, 2021...
[0104] X * = (10730-4210) / (12670-4210) = 0.7706856
[0105] Table 2. Normalized daily production data for Well W1
[0106]
[0107]
[0108] In one embodiment of the present invention, generally, the training set of data is used to estimate the parameters in the model so that the model can reflect reality and thus predict the future or other unknown information, while the test set of data is used to evaluate the predictive performance of the model.
[0109] For example, when the method of the present invention is put into use, daily production data of 100 wells are collected, and the data content of each well is shown in Table 3:
[0110] Table 3 shows partial daily production data from 100 wells.
[0111]
[0112] We can adopt the principle of "80% training data + 20% test data", select the first 80% of the date data for each well for the next step of model validation analysis, and then use the remaining 20% of the date data for production prediction to evaluate the model's performance.
[0113] In one embodiment of the present invention, the method employs the LSTM algorithm to construct an LSTM model. Recurrent Neural Networks (RNNs) are commonly used to process sequential data, but RNNs suffer from "memory loss" when processing long sequences. LSTM is a modified RNN structure that overcomes this shortcoming by introducing a gating mechanism into its internal structure, making the memory information in the time series controllable.
[0114] The LSTM algorithm supports multiple parameter inputs and outputs, with multiple hidden layers in between. The input parameters are normalized and then fed into the LSTM neural network for processing. After multiple iterations, a prediction model for the output parameters is obtained. In this invention, the input parameters are selected as daily gas production, daily water production, and average casing pressure; the output parameter is daily gas production; and two hidden layers are set in between, such as... Figure 4 As shown:
[0115] 1) The input layer uses three parameters: daily gas production, daily water production, and average casing pressure;
[0116] 2) The output layer has one parameter: daily gas production;
[0117] 3) Create two hidden layers: the first layer has 12 nodes, and the second layer has 4 nodes;
[0118] When constructing an LSTM model, the optimizer uses the Adam optimization algorithm because the Adam algorithm (an extension of stochastic gradient descent) can dynamically adjust the learning rate of each input parameter.
[0119] 4) The loss function used is MAE, Mean Absolute Error:
[0120]
[0121] In the formula, n represents the number of date points. The predicted value of each date point is subtracted from the actual value, the absolute values are summed, and finally divided by n to obtain the mean absolute error (MAE).
[0122] In one embodiment of the present invention, in order to fully leverage the influence of historical data on the current predicted value, this scheme adds a temporal attention mechanism in the construction of the LSTM model. Figure 5 The diagram illustrates the principle of temporal attention mechanisms in existing technologies, combined with... Figure 5r1 to rT represent the model training input values from date 1 to T, h represents the initial predicted value, and the predicted values from h1 to hT are validated by the typical curves of the corresponding date points to obtain the β values used to calculate the weight coefficients. Then, the softmax function is used to adjust the weight coefficients β1 to β2. t After normalization, the weight values α1 to α2 are obtained. t The predicted value is then multiplied by the initial predicted value h to obtain the final predicted value. The method of obtaining the weight coefficients through "typical curve verification" is the content of this invention. The principle of determining the weight coefficients is as follows: When using LSTM to predict the "N+1"th point, the predicted values of the "first N" points are needed, but the weight values of each of the "first N" points can be different. The actual value (or predicted value) of each of the "first N" points is compared with the value corresponding to the typical curve. The larger the difference, the lower the weight; the smaller the difference, the more accurate the prediction of this point, and the higher the weight.
[0123] The daily gas production in the constructed LSTM model is weighted and redistributed using a time-series attention mechanism, including the following steps.
[0124] Calculate the sequence values N1 to NT for the "Nth day of production" corresponding to the dates r1 to rT, where r1 represents the first date point and rT represents the Tth date point. Calculate the sequence values N1 to NT for the "Nth day of production" on the typical curve corresponding to the dates r1 to rT.
[0125] Take the daily output values from point N1 to NT on the typical curve;
[0126] The daily output forecast values from points N1 to NT are fitted using a power function. The absolute value β' of the vertical distance between the output value (actual or predicted) of each point from r1 to rT and the fitted curve is used as the basis for determining the weight coefficients (the weight coefficients are inversely proportional, β = 1 / β', that is, the larger the vertical distance, the smaller the weight coefficients). Finally, the softmax function is used to normalize the weight coefficients.
[0127] like Figure 6 As shown in the figure, the dashed line represents the curve obtained by fitting the daily production data corresponding to the typical curve date series N1 to NT using a power function. Point A corresponds to the preliminary predicted value hi by LSTM, and β' represents the absolute value of the vertical distance between the preliminary predicted value and the fitted curve.
[0128] In one embodiment of the present invention, a model validation analysis is performed using the constructed trained LSTM model and a test data set. When the RMSE (Root Mean Square Error) loss is within a specified range, the LSTM model is obtained. Since the data was normalized in the previous processing steps, it is necessary to restore the predicted "daily gas production (normalized)" to the actual units of the predicted daily gas production. Then, the RMSE is used to calculate the loss. The formula for calculating the prediction error is as follows:
[0129]
[0130] In the formula, RMSE represents the root mean square error of the prediction, n represents the number of prediction time points, and i represents the prediction time point number; a i p represents the actual daily gas production value at the i-th time point. i This represents the predicted daily gas production value at the i-th time point.
[0131] In one embodiment of the invention, the final model is used to predict the daily gas production of the well for the next year.
[0132] The method of establishing a predicted daily gas production curve for a single well by combining typical daily gas production curves of the well area and an LSTM model for predicting daily gas production of a single well includes the following steps:
[0133] The predicted daily gas production value for each well is calculated using an LSTM model to predict daily gas production, and the corresponding "day N" after production commencement is marked for each value.
[0134] Obtain the typical value for day N from the typical curve based on "day N after production".
[0135] The predicted value is corrected based on a 10% margin above and below the typical value, and controlled within the normal value range.
[0136] Based on the corrected predicted values, a single-well production prediction curve is constructed.
[0137] like Figure 7 As shown, Figure 7 The diagram shows the predicted and actual gas production of a well each day: the horizontal axis represents the Nth day after the well is put into production, the actual line represents the daily gas production, and the dashed line represents the daily gas production predicted by the prediction model of this invention. The actual gas production of this well only extends to the 36th day, so the dashed lines after the 36th day can be used as a reference for the predicted production.
[0138] This invention also discloses a shale gas single-well daily production prediction system, the system comprising,
[0139] The typical curve establishment unit is used to establish the typical daily gas production curve of the well area based on the daily gas production data of all wells in the shale gas well area;
[0140] The model building unit is used to build an LSTM model to predict the daily gas production of a single well based on the daily production data of a single well in the shale gas well area.
[0141] The prediction curve establishment unit is used to combine the typical daily gas production curve of the well area with the LSTM model for predicting the daily gas production of a single well to establish the prediction curve of the daily gas production of a single well.
[0142] Regarding the aforementioned shale gas single-well daily production prediction system, the specific methods by which each unit performs its operations have been described in detail in the embodiments of the relevant method, and will not be elaborated upon here.
[0143] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting the daily production of a single shale gas well, characterized in that, The method includes the following steps: Based on the total daily gas production data of all wells in the shale gas well area, a typical daily gas production curve for the well area is established. An LSTM model is constructed based on the daily production data of a single well in the shale gas well area. By combining typical daily gas production curves of the well area and an LSTM model, a prediction curve for daily gas production of a single well is established; among which, The establishment of a typical daily gas production curve for the well area includes the following steps. Obtain all daily gas production data for all wells in shale gas well areas using natural production methods or controlled production methods; Remove records with a daily gas production of 0; Sort the daily gas production data of each well by date and label them with sequence numbers 1, 2...N; The daily gas production values for the same sequence number in all wells are statistically analyzed to obtain the daily gas production statistics corresponding to each sequence number. Based on the daily gas production statistics corresponding to each sequence number, a typical daily gas production curve for the well area is constructed. The construction of the LSTM model includes, Normalize the daily production data of individual wells in the shale gas well area; For the normalized production day data of a single well, divide the data into a training set and a test set; Based on the training and test sets of data, construct an LSTM model; The construction of the LSTM model based on the training and test data sets includes, The LSTM neural network processes the parameters of the training data set to construct the LSTM model; Based on the test data set, the LSTM model is validated and analyzed. The LSTM model is evaluated based on the validation analysis results. The LSTM neural network processes the parameters of the training data set and constructs the LSTM model by combining optimization algorithms and loss functions to dynamically adjust the output parameters of each input training data set parameter. After multiple iterations, the LSTM model is constructed. The validation analysis of the LSTM model based on the test dataset includes: The prediction error is calculated based on the predicted daily gas production data and the data test set. Based on the prediction error results, the LSTM model is judged: the smaller the prediction error, the more accurate the LSTM model. The formula for calculating the prediction error, based on the predicted daily gas production data and the data test set, is as follows: In the formula, RMSE represents the root mean square error of the prediction, n represents the number of prediction date points, and i represents the prediction date point number; a i p represents the actual daily gas production value at the i-th date. i This represents the predicted daily gas production value for the i-th date. The method of establishing a daily gas production prediction curve for a single well by combining typical daily gas production curves of the well area and an LSTM model includes the following steps: The predicted daily gas production is calculated based on the LSTM model, and the corresponding "day N of production" is marked for each value. Obtain the typical value for day N from the typical curve based on "day N after production". The predicted value is corrected within ±10% of the typical value to keep it within the normal range. Based on the corrected predicted values, a single-well production prediction curve is constructed.
2. The method for predicting the daily production of a single shale gas well according to claim 1, characterized in that, The shale gas well areas include shale gas well areas with natural production and shale gas well areas with controlled production.
3. The method for predicting the daily production of a single shale gas well according to claim 1, characterized in that, The statistical analysis of daily gas production values for the same sequence number in all wells includes, The daily gas production values for the same sequence number in all wells were statistically analyzed using the average value, P50, or M50 statistical algorithms; among them, P50 represents the middle daily gas production value after all wells with the same sequence number are sorted from low to high. M50 represents the average daily gas production value from the 25th percentile to the 75th percentile after sorting all wells with the same sequence number from low to high.
4. The method for predicting the daily production of a single shale gas well according to claim 1, characterized in that, The daily production data of a single well in the shale gas well area includes... Wells with fewer than 30 days of production data records, and wells with 30 or more days of production data records; among them, For single wells with fewer than 30 days of production data records, the value of M50 is used as the daily production data to predict the daily production data of the single well. For single wells with more than or equal to 30 days of production data records, the actual production data or daily production data of a certain year is taken, and records with a daily production data of 0 are removed to construct an LSTM model. The production day data includes one or more of the following: production date, daily gas production, daily water production, and average casing pressure.
5. The method for predicting the daily production of a single shale gas well according to claim 1, characterized in that, The validation analysis of the LSTM model based on the data test set also... Includes the following steps, Based on the typical curves of the shale gas well area, the daily gas production in the LSTM model is reassigned using a time-series attention mechanism.
6. The method for predicting the daily production of a single shale gas well according to claim 5, characterized in that, By incorporating typical curves from the shale gas well region, the daily gas production in the LSTM model is weighted and redistributed using a time-series attention mechanism. Includes the following steps, Calculate the sequence values N1 to NT for the "Nth day of production" corresponding to dates r1 to rT, where r1 represents the first date and rT represents the Tth date. Take the predicted production day data from points N1 to NT on the typical curve; The predicted production day data from points N1 to NT are fitted using a power function to obtain the fitted curve; The absolute value of the vertical distance between the actual or predicted production day data value and the fitted curve for each point from r1 to rT is used to determine the weighting coefficients. The larger the vertical distance, the smaller the weighting coefficient; the smaller the vertical distance, the larger the weighting coefficient.
7. A shale gas single-well daily production prediction system, characterized in that, The system, applicable to the method of any one of claims 1-6, comprises: The typical curve establishment unit is used to establish the typical daily gas production curve of the well area based on the daily gas production data of all wells in the shale gas well area; The model building unit is used to build an LSTM model based on the daily production data of a single well in the shale gas well area. The prediction curve establishment unit is used to combine the typical daily gas production curve of the well area with the LSTM model to establish the daily gas production prediction curve of a single well.