Method, device, equipment and medium for predicting production probability of multiple time series well patterns

Abstract

The present application discloses a method, device, equipment and medium for predicting the probability of well network production in multiple time series, and relates to the field of deep learning technology for predicting oil and natural gas production, including determining a training set and a test set; establishing an autoregressive neural network and a production probability distribution function, and constructing a deep autoregressive network model; performing maximum likelihood estimation on the production probability distribution function to obtain production probability distribution function parameters, optimizing the maximum likelihood function, and obtaining target production probability distribution function parameters; judging whether the target production probability distribution function parameters meet the parameter index, and if so, predicting the well network production of multiple time series of the test set, obtaining the well network production prediction results, and realizing the evaluation of the well network production prediction results of multiple time series of the deep autoregressive network model after training. It realizes training and prediction for multiple wells, makes full use of oilfield data to improve the efficiency of multi-well modeling training, reduces the time cost of modeling, and improves the prediction effect of unconventional tight gas production.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology for oil and natural gas production prediction, and in particular to a method, device, equipment and medium for predicting the production probability of a multiple time series well network. Background Art

[0002] Conventional production prediction methods include DCA (Decision Curve Analysis), a production decline curve analysis method, material balance method, and machine learning method. The production decline curve method is the most widely used because it is simple and effective. Assuming that the bottom hole pressure is constant and the flow rate gradually decreases, the corresponding decline pattern is fitted by actual production data to obtain the decline rate and predicted final recovery. The material balance method overcomes the challenges brought by variable flow and variable pressure conditions. This method aims to solve the situation where flow and pressure conditions change over time. In recent years, there has been a significant increase in the research on deep learning technology for oil and gas production prediction. Among these methods, the most widely used algorithm is the time series-RNN (Recurrent Neural Network) algorithm. Many studies have explored the powerful nonlinear processing and data assimilation capabilities of deep learning in this field. However, the research objects of traditional methods are all point models for single wells. They cannot process hundreds or thousands of wells at the same time, cannot perform probabilistic predictions, and cannot perform classification predictions.

[0003] From the above, it can be seen that how to realize training prediction for multiple wells, make full use of oil field data to improve the efficiency of multi-well modeling training, reduce the time cost of modeling, and thus improve the effect of unconventional tight gas production prediction is a problem to be solved in this field. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for predicting the production probability of multiple time series well patterns, which can realize training prediction for multiple wells, make full use of oil field data to improve the efficiency of multi-well modeling training, reduce the time cost of modeling, and thus improve the effect of predicting the production of unconventional tight gas. The specific scheme is as follows:

[0005] In a first aspect, the present application discloses a multiple time series well pattern production probability prediction method, which is applied to a preset multiple time series well pattern production probability prediction model, comprising:

[0006] Acquire characteristic data of well pattern production of multiple time series, and determine a training set and a test set based on the characteristic data;

[0007] Determine the well pattern production prediction characteristics of multiple time series, establish an autoregressive neural network and a production probability distribution function, and construct a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function;

[0008] Performing maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, optimizing the preset maximum likelihood function based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the deep autoregressive network model and target production probability distribution function parameters after training;

[0009] Determine whether the target production probability distribution function parameters meet the preset parameter indicators. If the target production probability distribution function parameters meet the preset parameter indicators, use the trained deep autoregressive network model to perform multiple time series well network production prediction on the test set to obtain the well network production prediction results of the multiple time series. Based on the well network production prediction results of the multiple time series and the actual results of the well network production of the multiple time series, evaluate the well network production prediction results of the multiple time series of the deep autoregressive network model after training.

[0010] Optionally, the step of obtaining characteristic data of well pattern production in multiple time series and determining a training set and a test set based on the characteristic data includes:

[0011] Acquire characteristic data of well pattern production of multiple time series, and preprocess the characteristic data to obtain processed data; wherein the preprocessing includes missing value processing, outlier processing, data smoothing and dynamic covariate normalization processing;

[0012] The processed data is divided according to a preset data set time range to obtain a training set and a test set.

[0013] Optionally, determining the well pattern production prediction characteristics of multiple time series includes:

[0014] The well network production prediction characteristics of the multiple time series are extracted and constructed based on the characteristics of reservoir engineering and seepage mechanics data; wherein the well network production prediction characteristics of the multiple time series include the geographical location of the wells, the type of wells and the statistical characteristics of historical production data.

[0015] Optionally, the establishing of an autoregressive neural network and a yield probability distribution function includes:

[0016] The autoregressive neural network is established using a long short-term memory network or a recurrent neural network as a basic framework, and the yield probability distribution function is constructed.

[0017] Optionally, performing maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and optimizing the preset maximum likelihood function based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters, including:

[0018] Using the maximum likelihood method to perform maximum likelihood estimation on the parameters in the yield probability distribution function to obtain the yield probability distribution function parameters;

[0019] A gradient descent algorithm is used to optimize the preset maximum likelihood function based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters.

[0020] Optionally, the determining whether the target yield probability distribution function parameters meet preset parameter indicators includes:

[0021] Determine whether the target yield probability distribution function parameters meet the preset root mean square error index, and determine whether the target yield probability distribution function parameters meet the preset mean absolute error index.

[0022] Optionally, after determining whether the target yield probability distribution function parameters meet preset parameter indicators, the method further includes:

[0023] If the parameters of the target production probability distribution function do not meet the preset parameter indicators, the hyperparameters of the deep autoregressive network model after training are adjusted to obtain the adjusted deep autoregressive network model, and then the adjusted deep autoregressive network model is used to perform well network production prediction for multiple time series on the test set to obtain well network production prediction results for multiple time series.

[0024] In a second aspect, the present application discloses a multiple time series well pattern production probability prediction device, which is applied to a preset multiple time series well pattern production probability prediction model, comprising:

[0025] A training and test set determination module, used to obtain characteristic data of well pattern production of multiple time series, and determine a training set and a test set based on the characteristic data;

[0026] A model building module, used to determine the well pattern production prediction characteristics of multiple time series, establish an autoregressive neural network and a production probability distribution function, and build a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function;

[0027] An optimization calculation module, used for performing maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, optimizing the preset maximum likelihood function based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the deep autoregressive network model and target production probability distribution function parameters after training;

[0028] The module for predicting the well network production of multiple time series is used to determine whether the parameters of the target production probability distribution function meet the preset parameter indicators. If the parameters of the target production probability distribution function meet the preset parameter indicators, the deep autoregressive network model after training is used to predict the well network production of multiple time series on the test set to obtain the prediction results of the well network production of multiple time series. Based on the prediction results of the well network production of multiple time series and the actual results of the well network production of multiple time series, the evaluation of the prediction results of the well network production of multiple time series of the deep autoregressive network model after training is realized.

[0029] In a third aspect, the present application discloses an electronic device, comprising:

[0030] Memory, used to store computer programs;

[0031] The processor is used to execute the computer program to implement the aforementioned multiple time series well pattern production probability prediction method.

[0032] In a fourth aspect, the present application discloses a computer storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the steps of the aforementioned disclosed multiple time series well network production probability prediction method are implemented.

[0033] It can be seen that the present application provides a method for predicting the probability of production of a well network with multiple time series, including obtaining characteristic data of the production of the well network with multiple time series, determining a training set and a test set based on the characteristic data; determining the characteristics of the production prediction of the well network with multiple time series, establishing an autoregressive neural network and a production probability distribution function, and constructing a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function; performing maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and based on the training set, the production prediction characteristics of the well network with multiple time series, the deep autoregressive network model and the production probability distribution function parameters, constructing a deep autoregressive network model; The preset maximum likelihood function is optimized to obtain the deep autoregressive network model and target production probability distribution function parameters after training; whether the target production probability distribution function parameters meet the preset parameter indicators, if the target production probability distribution function parameters meet the preset parameter indicators, the deep autoregressive network model after training is used to perform well network production prediction of multiple time series on the test set to obtain the well network production prediction results of multiple time series, and the well network production prediction results of multiple time series of the deep autoregressive network model after training are evaluated based on the well network production prediction results of multiple time series and the actual results of the well network production of multiple time series. The present application determines a training set and a test set, establishes an autoregressive neural network and a production probability distribution function, and constructs a deep autoregressive network model. The production probability distribution function is estimated by maximum likelihood, and the production probability distribution function parameters are obtained. The maximum likelihood function is optimized to obtain the target production probability distribution function parameters, thereby realizing training and prediction for multiple wells, making full use of oil field data to improve the efficiency of multi-well modeling training, reducing the time cost of modeling, and judging through preset parameter indicators whether the target production probability distribution function parameters meet the preset parameter indicators, thereby improving the accuracy and stability of the evaluation of the well network production prediction results of multiple time series of the deep autoregressive network model after training, and improving the effect of predicting unconventional tight gas production. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0035] Figure 1 A flow chart of a method for predicting the probability of production of a multiple time series well pattern disclosed in this application;

[0036] Figure 2A flow chart of a method for predicting the probability of production of a multiple time series well pattern disclosed in this application;

[0037] Figure 3 A structural diagram of a multiple time series well pattern production probability prediction model disclosed in this application;

[0038] Figure 4 A prediction result diagram of Well 1 disclosed in this application;

[0039] Figure 5 A prediction result diagram of Well 2 disclosed in this application;

[0040] Figure 6 A prediction result diagram of Well 3 disclosed in this application;

[0041] Figure 7 This is a schematic diagram of the structure of a multiple time series well pattern production probability prediction device disclosed in this application;

[0042] Figure 8 A structural diagram of an electronic device provided for this application. DETAILED DESCRIPTION

[0043] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0044] Conventional production prediction methods include DCA, a production decline curve analysis method, material balance method, and machine learning method. The production decline curve method is the most widely used because it is simple and effective. Assuming that the bottom hole pressure is constant and the flow rate gradually decreases, the corresponding decline pattern is fitted by actual production data to obtain the decline rate and predicted ultimate recovery. The material balance method overcomes the challenges brought by variable flow and variable pressure conditions. This method aims to solve the situation where flow and pressure conditions change over time. In recent years, there has been a significant increase in the research on deep learning technology for oil and gas production prediction. Among these methods, the most widely used algorithm is the time series-RNN algorithm. Many studies have explored the powerful nonlinear processing and data assimilation capabilities of deep learning in this field. However, the research objects of traditional methods are all point models for single wells. They cannot process hundreds or thousands of wells at the same time, cannot perform probabilistic predictions, and cannot perform classification predictions. As can be seen from the above, how to realize training prediction for multiple wells, make full use of oil field data to improve the efficiency of multi-well modeling training, reduce the time cost of modeling, and thus improve the effect of unconventional tight gas production prediction is a problem to be solved in this field.

[0045] See also Figure 1 As shown, the embodiment of the present invention discloses a method for predicting the probability of production of a well pattern with multiple time series, which is applied to a preset well pattern production probability prediction model with multiple time series, and specifically may include:

[0046] Step S11: Acquire characteristic data of well pattern production in multiple time series, and determine a training set and a test set based on the characteristic data.

[0047] In this embodiment, characteristic data of well network production of multiple time series are obtained, and the characteristic data are preprocessed to obtain processed data; wherein the preprocessing includes missing value processing, outlier processing, data smoothing and dynamic covariate normalization processing; the processed data are divided according to a preset data set time range to obtain a training set and a test set.

[0048] Specifically, the characteristic data of well network production in multiple time series are obtained, including dynamic data and static data of multi-well production. The dynamic data (daily production, monthly production, oil pressure, casing pressure) and static data (well type, layer, block, initial casing pressure, water-steam ratio) of multi-well production are sorted and preprocessed, including missing value processing, outlier processing, dynamic covariate normalization processing, data smoothing, etc., and then the processed data are divided into training set and test set according to the time range of the data set.

[0049] Step S12: Determine the well network production prediction characteristics of multiple time series, establish an autoregressive neural network and a production probability distribution function, and construct a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function.

[0050] In this embodiment, the well network production prediction characteristics of the multiple time series are extracted and constructed based on the reservoir engineering and seepage mechanics data characteristics; wherein the well network production prediction characteristics of the multiple time series include the geographical location of the well, the type of well, and the statistical characteristics of the historical production data, and then, an autoregressive neural network and a production probability distribution function are established, and a deep autoregressive network model is constructed based on the autoregressive neural network and the production probability distribution function. The specific process of determining the well network production prediction characteristics of the multiple time series is: according to the knowledge and data characteristics in the field of reservoir engineering and seepage mechanics, the well network production prediction characteristics of the multiple time series suitable for prediction are extracted and constructed, and the well network production prediction characteristics of the multiple time series may include the geographical location of the well, the type of well, the statistical characteristics of the historical production data, etc.

[0051] Step S13: Perform maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and optimize the preset maximum likelihood function based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the trained deep autoregressive network model and target production probability distribution function parameters.

[0052] Step S14: Determine whether the target production probability distribution function parameters meet the preset parameter indicators. If the target production probability distribution function parameters meet the preset parameter indicators, use the trained deep autoregressive network model to perform multiple time series well network production prediction on the test set to obtain the well network production prediction results of the multiple time series. Based on the well network production prediction results of the multiple time series and the actual results of the well network production of the multiple time series, the well network production prediction results of the multiple time series of the deep autoregressive network model after training are evaluated.

[0053] In this embodiment, it is determined whether the prediction accuracy of the target production probability distribution function parameters on the training set meets the preset root mean square error index, and it is determined whether the prediction accuracy of the target production probability distribution function parameters on the training set meets the preset mean absolute error index; the preset parameter index includes but is not limited to the preset root mean square error index and the preset mean absolute error index. If the target production probability distribution function parameters meet the preset parameter index, the trained deep autoregressive network model is used to predict the well network production of multiple time series on the test set to obtain the well network production prediction results of multiple time series, and the well network production prediction results of multiple time series of the trained deep autoregressive network model are evaluated based on the well network production prediction results of multiple time series and the actual well network production results of multiple time series; if the target production probability distribution function parameters do not meet the preset parameter index, the trained deep autoregressive network model is hyperparameter adjusted to obtain the adjusted deep autoregressive network model, and then the adjusted deep autoregressive network model is used to predict the well network production of multiple time series on the test set to obtain the well network production prediction results of multiple time series.

[0054] Specifically, the deep autoregressive network model after training is evaluated using preset parameter indicators, and various evaluation indicators, such as RMSE (Root-mean-square error), MAE (Mean Absolute Error), etc., can be used to evaluate the prediction performance of the model. If the parameters of the target production probability distribution function do not meet the preset parameter indicators, the hyperparameters of the trained deep autoregressive network model are adjusted, such as network structure, learning rate, batch size, etc., to optimize the performance and prediction accuracy of the trained deep autoregressive network model. Finally, the adjusted deep autoregressive network model is used to predict the well network production of multiple time series on the test set to obtain the well network production prediction results of multiple time series, and the production of different time periods can be predicted as needed, such as the production of the next day, week or month. In addition, the well network production prediction results of multiple time series can also be analyzed to evaluate the accuracy and stability of the adjusted deep autoregressive network model.

[0055] In this embodiment, characteristic data of well network production of multiple time series are obtained, and a training set and a test set are determined based on the characteristic data; the prediction characteristics of well network production of multiple time series are determined, an autoregressive neural network and a production probability distribution function are established, and a deep autoregressive network model is constructed based on the autoregressive neural network and the production probability distribution function; the maximum likelihood estimation is performed on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and the preset maximum likelihood function is estimated based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters. The method comprises the following steps: optimizing the deep autoregressive network model after training and obtaining the target production probability distribution function parameters; judging whether the target production probability distribution function parameters meet the preset parameter indicators; if the target production probability distribution function parameters meet the preset parameter indicators, using the deep autoregressive network model after training to perform well network production prediction of multiple time series on the test set to obtain the well network production prediction results of multiple time series; and evaluating the well network production prediction results of multiple time series of the deep autoregressive network model after training based on the well network production prediction results of multiple time series and the actual results of the well network production of multiple time series. The present application determines a training set and a test set, establishes an autoregressive neural network and a production probability distribution function, and constructs a deep autoregressive network model. The production probability distribution function is estimated by maximum likelihood, and the production probability distribution function parameters are obtained. The maximum likelihood function is optimized to obtain the target production probability distribution function parameters, thereby realizing training and prediction for multiple wells, making full use of oil field data to improve the efficiency of multi-well modeling training, reducing the time cost of modeling, and judging through preset parameter indicators whether the target production probability distribution function parameters meet the preset parameter indicators, thereby improving the accuracy and stability of the evaluation of the well network production prediction results of multiple time series of the deep autoregressive network model after training, and improving the effect of predicting unconventional tight gas production.

[0056] See also Figure 2 As shown, the embodiment of the present invention discloses a method for predicting the probability of production of a well pattern with multiple time series, which is applied to a preset well pattern production probability prediction model with multiple time series, and specifically may include:

[0057] Step S21: Acquire characteristic data of well pattern production in multiple time series, and determine a training set and a test set based on the characteristic data.

[0058] Step S22: Determine the well network production prediction characteristics of multiple time series, use the long short-term memory network or the recurrent neural network as the basic framework to establish the autoregressive neural network, and construct the production probability distribution function, and construct a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function.

[0059] In this embodiment, an autoregressive neural network is established using LSTM (Long Short-Term Memory) or RNN as the basic framework, and a production probability distribution function (normal distribution, T distribution, etc.) is constructed. A deep autoregressive network model is constructed based on the autoregressive neural network and the production probability distribution function. The input of the deep autoregressive network model is the production of multiple wells in the previous n-1 days, and the output is the predicted target production probability distribution function parameters.

[0060] Step S23: Use the maximum likelihood method to perform maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and use the gradient descent algorithm to optimize the preset maximum likelihood function based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the deep autoregressive network model and target production probability distribution function parameters after training.

[0061] Step S24: Determine whether the target production probability distribution function parameters meet the preset parameter indicators. If the target production probability distribution function parameters meet the preset parameter indicators, use the trained deep autoregressive network model to perform multiple time series well network production prediction on the test set to obtain the well network production prediction results of the multiple time series. Based on the well network production prediction results of the multiple time series and the actual results of the well network production of the multiple time series, the well network production prediction results of the multiple time series of the trained deep autoregressive network model are evaluated.

[0062] The present application is applied to the well network production probability prediction model of the preset multiple time series. In order to verify the effect of using the deep autoregressive network model to predict the well network production of multiple time series of tight gas, the actual 2000 tight gas production data and the corresponding dynamic and static characteristic data are used. These characteristic data include the production information of each well and various characteristics related to the production, such as wellhead pressure, formation permeability, bottom hole flow pressure, etc. First, the characteristic data containing the production data of multiple wells and related characteristic data in the oil field are collected and sorted. The characteristic data are preprocessed, including missing value processing, outlier processing, data smoothing, etc., and the preprocessed data are divided into training set and test set according to the preset data set time range. According to the knowledge and data characteristics of the fields of reservoir engineering and seepage mechanics, the well network production prediction characteristics of multiple time series of dynamic time-varying covariates, static classification covariates and static time-invariant covariates suitable for prediction are extracted and constructed, and the well network production prediction characteristics of the multiple time series may include the statistical characteristics of the production layer of the well, the type of well, the block, etc. The autoregressive neural network is constructed using LSTM or RNN as the basic framework, and a production probability distribution function (normal distribution, T distribution, etc.) is constructed. A deep autoregressive network model is constructed based on the autoregressive neural network and the production probability distribution function. The deep autoregressive neural network algorithm is used to establish a multiple time series well network production probability prediction model, which is a sequence prediction model based on a recurrent neural network. The input of the model includes historical production data and corresponding feature data, as well as the time period to be predicted. The model predicts production in the future by learning patterns and trends in historical data. Some parameter settings of the well network production probability prediction model for multiple time series are as follows: Figure 3 As shown, the well network production sequence of multiple time series, the daily or monthly production data of all wells in the overall sample; covariate features, where covariate features refer to daily production time or monthly production time; dynamic and static classification features, here refers to classification features, which divide each well into different types, such as (well type, layer, block, initial casing pressure, water vapor ratio); LSTMn (long short-term memory network), the forward recurrent neural network unit type in the system of the probability classification prediction method of the well network production of multiple time series, which can use LSTM, GRU (Gate Recurrent Unit, a type of recurrent neural network), RNN. Here, the basic forward recurrent neural network framework is defined, including the number of layers of the network, and the number of basic units in each layer (both of which are hyperparameters that need to be optimized). W μ is the weight of the linear transformation of the mean μ of the likelihood parameter: μ(h i,t )=w μ h i,t +b μ ;

[0063] Among them, hi,t is the output of the last hidden layer of the last recurrent neural network, b μ is the mean;

[0064] W σ is the weight of the linear transformation of the likelihood parameter standard deviation σ. Since the standard deviation needs to be always greater than 0, the Softplus function is used for nonlinear transformation to ensure that its value is greater than 0;

[0065]

[0066] softplus(x)=log(1+e x );

[0067] Choosing Gaussian distribution as the likelihood function can be expressed as:

[0068]

[0069] Among them, b σ is the bias of the standard deviation transformation, and z is the time series value at time t. This application uses the maximum likelihood to estimate the parameters in the probability distribution function, uses the gradient descent algorithm to optimize the maximum likelihood function, obtains the minimum value of the maximum likelihood function, obtains the yield probability distribution, and uses the preset parameter indicators to evaluate the deep autoregressive network model after training. Various evaluation indicators, such as RMSE (Root-mean-square error), MAE (Mean Absolute Error), etc., can be used to evaluate the prediction performance of the model. If the parameters of the target yield probability distribution function do not meet the preset parameter indicators, the hyperparameters of the deep autoregressive network model after training are adjusted, such as network structure, learning rate, batch size, etc., to optimize the performance and prediction accuracy of the deep autoregressive network model after training. The specific adjusted hyperparameters are shown in Table 1:

[0070] Table 1

[0071]

[0072] This application uses various evaluation indicators, such as root mean square error (RMSE), mean absolute error (MAE), etc. to evaluate the prediction performance of the model. The production forecast results of the deep autoregressive network model for the next 12 months are as follows: Figure 4-Figure 6 As shown, Figure 4 Predicted results for well 1, Figure 5 Predict the results for Well 2, Figure 6This is the prediction result of Well 3. It can be seen from the figure that compared with the traditional RNN series of time series neural network models, the deep autoregressive network model has a very good ability to capture the characteristics of switch wells, and is particularly suitable for the production capacity prediction of intermittent wells. The prediction curve of the deep autoregressive network model can be abrupt and non-smooth, which is a reflection of the comprehensive results of the trend factors and random factors of the production time series. The average determination coefficient R2 on the prediction set of more than 2,000 wells is 0.8, indicating that the prediction performance of the global model established by the deep autoregressive network model is very good, and it can realize simultaneous training and prediction of multiple wells, so that oilfield data can be more fully utilized and the time cost of single well modeling can be saved. In addition, the classification prediction is used to greatly improve the effect of unconventional tight gas production prediction.

[0073] In this embodiment, characteristic data of well network production of multiple time series are obtained, and a training set and a test set are determined based on the characteristic data; the prediction characteristics of well network production of multiple time series are determined, an autoregressive neural network and a production probability distribution function are established, and a deep autoregressive network model is constructed based on the autoregressive neural network and the production probability distribution function; the maximum likelihood estimation is performed on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and the preset maximum likelihood function is estimated based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters. The method comprises the following steps: optimizing the deep autoregressive network model after training and obtaining the target production probability distribution function parameters; judging whether the target production probability distribution function parameters meet the preset parameter indicators; if the target production probability distribution function parameters meet the preset parameter indicators, using the deep autoregressive network model after training to perform well network production prediction of multiple time series on the test set to obtain the well network production prediction results of multiple time series; and evaluating the well network production prediction results of multiple time series of the deep autoregressive network model after training based on the well network production prediction results of multiple time series and the actual results of the well network production of multiple time series. The present application determines a training set and a test set, establishes an autoregressive neural network and a production probability distribution function, and constructs a deep autoregressive network model. The production probability distribution function is estimated by maximum likelihood, and the production probability distribution function parameters are obtained. The maximum likelihood function is optimized to obtain the target production probability distribution function parameters, thereby realizing training and prediction for multiple wells, making full use of oil field data to improve the efficiency of multi-well modeling training, reducing the time cost of modeling, and judging through preset parameter indicators whether the target production probability distribution function parameters meet the preset parameter indicators, thereby improving the accuracy and stability of the evaluation of the well network production prediction results of multiple time series of the deep autoregressive network model after training, and improving the effect of predicting unconventional tight gas production.

[0074] See also Figure 7 As shown, the embodiment of the present invention discloses a multiple time series well pattern production probability prediction device, which is applied to a preset multiple time series well pattern production probability prediction model, and specifically may include:

[0075] The training and test set determination module 11 is used to obtain characteristic data of well pattern production of multiple time series, and determine the training set and the test set based on the characteristic data;

[0076] A model building module 12 is used to determine the well pattern production prediction characteristics of multiple time series, establish an autoregressive neural network and a production probability distribution function, and build a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function;

[0077] An optimization calculation module 13 is used to perform maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and optimize the preset maximum likelihood function based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the deep autoregressive network model and target production probability distribution function parameters after training;

[0078] The multiple time series well network production prediction module 14 is used to determine whether the target production probability distribution function parameters meet the preset parameter indicators. If the target production probability distribution function parameters meet the preset parameter indicators, the trained deep autoregressive network model is used to perform multiple time series well network production prediction on the test set to obtain the multiple time series well network production prediction results. Based on the multiple time series well network production prediction results and the actual results of the multiple time series well network production, the multiple time series well network production prediction results of the trained deep autoregressive network model are evaluated.

[0079] In this embodiment, characteristic data of well network production of multiple time series are obtained, and a training set and a test set are determined based on the characteristic data; the prediction characteristics of well network production of multiple time series are determined, an autoregressive neural network and a production probability distribution function are established, and a deep autoregressive network model is constructed based on the autoregressive neural network and the production probability distribution function; the maximum likelihood estimation is performed on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and the preset maximum likelihood function is estimated based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters. The method comprises the following steps: optimizing the deep autoregressive network model after training and obtaining the target production probability distribution function parameters; judging whether the target production probability distribution function parameters meet the preset parameter indicators; if the target production probability distribution function parameters meet the preset parameter indicators, using the deep autoregressive network model after training to perform well network production prediction of multiple time series on the test set to obtain the well network production prediction results of multiple time series; and evaluating the well network production prediction results of multiple time series of the deep autoregressive network model after training based on the well network production prediction results of multiple time series and the actual results of the well network production of multiple time series. The present application determines a training set and a test set, establishes an autoregressive neural network and a production probability distribution function, and constructs a deep autoregressive network model. The production probability distribution function is estimated by maximum likelihood, and the production probability distribution function parameters are obtained. The maximum likelihood function is optimized to obtain the target production probability distribution function parameters, thereby realizing training and prediction for multiple wells, making full use of oil field data to improve the efficiency of multi-well modeling training, reducing the time cost of modeling, and judging through preset parameter indicators whether the target production probability distribution function parameters meet the preset parameter indicators, thereby improving the accuracy and stability of the evaluation of the well network production prediction results of multiple time series of the deep autoregressive network model after training, and improving the effect of predicting unconventional tight gas production.

[0080] In some specific embodiments, the training and test set determination module 11 may specifically include:

[0081] A preprocessing module, used to obtain characteristic data of well pattern production of multiple time series, and preprocess the characteristic data to obtain processed data; wherein the preprocessing includes missing value processing, outlier processing, data smoothing and dynamic covariate normalization processing;

[0082] The data partitioning module is used to partition the processed data according to a preset data set time range to obtain a training set and a test set.

[0083] In some specific embodiments, the model building module 12 may specifically include:

[0084] A module for constructing the production prediction characteristics of a well network of multiple time series is used to extract and construct the production prediction characteristics of the well network of multiple time series based on the characteristics of reservoir engineering and seepage mechanics data; wherein the production prediction characteristics of the well network of multiple time series include the geographical location of the wells, the type of wells and the statistical characteristics of historical production data.

[0085] In some specific embodiments, the model building module 12 may specifically include:

[0086] The autoregressive neural network establishment module is used to establish the autoregressive neural network using a long short-term memory network or a recurrent neural network as a basic framework, and construct the yield probability distribution function.

[0087] In some specific embodiments, the optimization calculation module 13 may specifically include:

[0088] A maximum likelihood estimation module, used to perform maximum likelihood estimation on the parameters in the yield probability distribution function using a maximum likelihood method to obtain the parameters of the yield probability distribution function;

[0089] An optimization module is used to optimize a preset maximum likelihood function using a gradient descent algorithm based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model, and the production probability distribution function parameters.

[0090] In some specific embodiments, the multiple time series well pattern production prediction module 14 may specifically include:

[0091] The judgment module is used to judge whether the target yield probability distribution function parameters meet the preset root mean square error index, and judge whether the target yield probability distribution function parameters meet the preset mean absolute error index.

[0092] In some specific embodiments, the multiple time series well pattern production prediction module 14 may specifically include:

[0093] The module for determining the well network production prediction results of multiple time series is used to adjust the hyperparameters of the deep autoregressive network model after training if the parameters of the target production probability distribution function do not meet the preset parameter indicators to obtain the adjusted deep autoregressive network model, and then use the adjusted deep autoregressive network model to perform well network production prediction of multiple time series on the test set to obtain the well network production prediction results of multiple time series.

[0094] Figure 8A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the multiple time series well network production probability prediction method performed by the electronic device disclosed in any of the aforementioned embodiments.

[0095] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0096] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon include an operating system 221, a computer program 222 and data 223, etc. The storage method can be temporary storage or permanent storage.

[0097] Among them, the operating system 221 is used to manage and control the hardware devices and computer programs 222 on the electronic device 20 to realize the operation and processing of the data 223 in the memory 22 by the processor 21, which can be Windows, Unix, Linux, etc. In addition to including a computer program that can be used to complete the multiple time series well network production probability prediction method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks. In addition to including data transmitted from an external device received by the multiple time series well network production probability prediction device, the data 223 can also include data collected by its own input and output interface 25, etc.

[0098] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0099] Furthermore, an embodiment of the present application also discloses a computer-readable storage medium, in which a computer program is stored. When the computer program is loaded and executed by a processor, the steps of the multiple time series well network production probability prediction method disclosed in any of the aforementioned embodiments are implemented.

[0100] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0101] The above is a detailed introduction to a method, device, equipment and storage medium for predicting the production probability of a multiple time series well network provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for predicting the probability of production of a multiple time series well pattern, characterized in that: The well pattern production probability prediction model applied to the preset multiple time series includes: Acquire characteristic data of well pattern production of multiple time series, and determine a training set and a test set based on the characteristic data; Extract and construct the well network production prediction features of the multiple time series based on reservoir engineering and seepage mechanics data features, establish an autoregressive neural network and a production probability distribution function, and construct a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function; the well network production prediction features of the multiple time series include the geographical location of the well, the type of well and the statistical characteristics of the historical production data; Performing maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, optimizing the preset maximum likelihood function based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the deep autoregressive network model and target production probability distribution function parameters after training; Determine whether the target production probability distribution function parameters meet the preset parameter indicators. If the target production probability distribution function parameters meet the preset parameter indicators, use the trained deep autoregressive network model to perform multiple time series well network production prediction on the test set to obtain the well network production prediction results of the multiple time series. Based on the well network production prediction results of the multiple time series and the actual results of the well network production of the multiple time series, evaluate the well network production prediction results of the multiple time series of the deep autoregressive network model after training.

2. The method for predicting the production probability of a multiple time series well pattern according to claim 1, characterized in that: The step of obtaining characteristic data of well pattern production in multiple time series and determining a training set and a test set based on the characteristic data includes: Acquire characteristic data of well pattern production of multiple time series, and preprocess the characteristic data to obtain processed data; wherein the preprocessing includes missing value processing, outlier processing, data smoothing and dynamic covariate normalization processing; The processed data is divided according to a preset data set time range to obtain a training set and a test set.

3. The method for predicting the production probability of a multiple time series well pattern according to claim 1, characterized in that: The step of establishing an autoregressive neural network and a yield probability distribution function comprises: The autoregressive neural network is established using a long short-term memory network or a recurrent neural network as a basic framework, and the yield probability distribution function is constructed.

4. The method for predicting the production probability of a multiple time series well pattern according to claim 1, characterized in that: The maximum likelihood estimation is performed on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, and the preset maximum likelihood function is optimized based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters, including: Using the maximum likelihood method to perform maximum likelihood estimation on the parameters in the yield probability distribution function to obtain the yield probability distribution function parameters; A gradient descent algorithm is used to optimize the preset maximum likelihood function based on the training set, the well network production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters.

5. The method for predicting the probability of production of a multiple time series well pattern according to claim 1, characterized in that: The determining whether the target yield probability distribution function parameters meet preset parameter indicators includes: Determine whether the target yield probability distribution function parameters meet the preset root mean square error index, and determine whether the target yield probability distribution function parameters meet the preset mean absolute error index.

6. The method for predicting the production probability of a multiple time series well pattern according to any one of claims 1 to 5, characterized in that: After determining whether the target yield probability distribution function parameters meet the preset parameter indicators, the method further includes: If the parameters of the target production probability distribution function do not meet the preset parameter indicators, the hyperparameters of the deep autoregressive network model after training are adjusted to obtain the adjusted deep autoregressive network model, and then the adjusted deep autoregressive network model is used to perform well network production prediction for multiple time series on the test set to obtain well network production prediction results for multiple time series.

7. A multiple time series well pattern production probability prediction device, characterized in that: The well pattern production probability prediction model applied to the preset multiple time series includes: A training and test set determination module, used to obtain characteristic data of well pattern production of multiple time series, and determine a training set and a test set based on the characteristic data; A model building module is used to extract and construct the well network production prediction characteristics of the multiple time series based on the characteristics of reservoir engineering and seepage mechanics data, establish an autoregressive neural network and a production probability distribution function, and build a deep autoregressive network model based on the autoregressive neural network and the production probability distribution function; the well network production prediction characteristics of the multiple time series include the geographical location of the well, the type of well and the statistical characteristics of the historical production data; An optimization calculation module, used for performing maximum likelihood estimation on the parameters in the production probability distribution function to obtain the production probability distribution function parameters, optimizing the preset maximum likelihood function based on the training set, the well pattern production prediction characteristics of the multiple time series, the deep autoregressive network model and the production probability distribution function parameters to obtain the deep autoregressive network model and target production probability distribution function parameters after training; The module for predicting the well network production of multiple time series is used to determine whether the parameters of the target production probability distribution function meet the preset parameter indicators. If the parameters of the target production probability distribution function meet the preset parameter indicators, the deep autoregressive network model after training is used to predict the well network production of multiple time series on the test set to obtain the prediction results of the well network production of multiple time series. Based on the prediction results of the well network production of multiple time series and the actual results of the well network production of multiple time series, the evaluation of the prediction results of the well network production of multiple time series of the deep autoregressive network model after training is realized.

8. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor is used to execute the computer program to implement the multiple time series well network production probability prediction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: Used to store computer programs; wherein, when the computer program is executed by a processor, the multiple time series well network production probability prediction method as described in any one of claims 1 to 6 is implemented.

Images