Training method and device for gas reservoir recovery ratio prediction model

Through deep learning technology combining dynamic and static characteristic data, a gas reservoir recovery prediction model is formed, which solves the problem of compound gas reservoir recovery prediction, improves the accuracy and stability of prediction, and provides effective gas reservoir development Adjustment basis.

CN119026720APending Publication Date: 2024-11-26HEFEI UNIV OF TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410826502.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-24
Publication Date
2024-11-26

Smart Images

  • Figure CN119026720A_ABST
    Figure CN119026720A_ABST
Patent Text Reader

Abstract

The invention discloses a training method and device for a gas reservoir recovery ratio prediction model, which can be applied to the field of deep learning, and the method comprises the steps: obtaining historical feature data of a plurality of gas reservoirs in a preset time period, and then performing iterative training on a first network in the gas reservoir recovery ratio prediction model by using dynamic continuous feature data in the historical feature data, and in each iterative process, adjusting network parameters of the first network until a loss function of the first network reaches a minimum value. And performing iterative training on the second network by using static discrete feature data in the historical feature data, and in each iteration process, adjusting network parameters of the first network until a loss function of the second network reaches a minimum value. The trained gas reservoir recovery ratio prediction model can be used to predict the recovery ratio of the gas reservoir, and the two deep learning models are coupled, so that the generalization ability of the model can be enhanced, and the prediction accuracy and stability can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of deep learning technology, and in particular to a training method and device for a gas reservoir recovery rate prediction model. Background Art

[0002] A gas reservoir is a geological structure that stores natural gas in underground rocks.

[0003] At present, natural gas can be extracted from gas reservoirs. Since most gas reservoirs are composite gas reservoirs controlled by structures, faults and formations, various technical countermeasures for improving recovery have been implemented in the process of developing natural gas. However, due to different geological conditions, the effects of the measures vary, resulting in large differences in recovery rates. Therefore, it is necessary to deepen the understanding of the recovery rates of different types of gas reservoirs.

[0004] Therefore, it is urgent to establish a gas reservoir recovery rate prediction method to predict the recovery rate of gas reservoirs, so as to carry out research on technical countermeasures to improve the recovery rate and provide a basis for gas reservoir development adjustment. Summary of the invention

[0005] The present application provides a method for training a gas reservoir recovery rate prediction model, so as to predict the recovery rate of a gas reservoir. The present application also provides a training device for the gas reservoir recovery rate prediction model.

[0006] In a first aspect, the present application provides a method for training a gas reservoir recovery factor prediction model, wherein the gas reservoir recovery factor prediction model includes a first network and a second network, and the method includes:

[0007] Obtain historical characteristic data of multiple gas reservoirs within a preset time period;

[0008] Iteratively training the first network using the dynamic continuous feature data in the historical feature data, and adjusting the network parameters of the first network in each iteration until the loss function of the first network reaches a minimum value;

[0009] The second network is iteratively trained using the static discrete feature data in the historical feature data, and in each iteration, the network parameters of the second network are adjusted until the loss function of the second network reaches a minimum value.

[0010] Optionally, the first network is a recurrent neural network-long short-term memory network RNN-LSTM, and the second network is a multi-layer perceptron MLP.

[0011] Optionally, the gas reservoir recovery rate prediction model further includes: an attention layer and a fully connected layer, and the method further includes:

[0012] Obtain historical characteristic data of the target gas reservoir within a preset time period;

[0013] The attention layer and the fully connected layer are iteratively trained using the historical characteristic data of the target gas reservoir within a preset time period. In each iteration, the network parameters of the attention layer and the network parameters of the fully connected layer are adjusted until the loss function of the attention layer and the loss function of the fully connected layer reach a minimum value.

[0014] Optionally, the method further includes:

[0015] The historical characteristic data of the target gas reservoir within a preset time period is input into a trained gas reservoir recovery rate prediction model including an attention layer and a fully connected layer to obtain the recovery rate of the target gas reservoir in the target period.

[0016] Optionally, the iterative training of the attention layer and the fully connected layer using the historical feature data of the target gas reservoir within a preset time period includes:

[0017] Preprocessing the historical characteristic data of the target gas reservoir within a preset time period;

[0018] The attention layer and the fully connected layer are iteratively trained using the preprocessed historical feature data.

[0019] Optionally, the dynamic continuous characteristic data of the historical characteristic data include: the monthly water-gas ratio of the gas reservoir, the recovery rate of the gas reservoir within a preset time period, the water-gas ratio of the gas reservoir in total months within a preset period, the number of production wells in the gas reservoir, the well-controlled area of ​​the gas reservoir and the number of water production wells in the gas reservoir.

[0020] In a second aspect, the present application further provides a training device for a gas reservoir recovery rate prediction model, the gas reservoir recovery rate prediction model includes a first network and a second network, and the device includes:

[0021] A first acquisition unit is used to acquire historical characteristic data of multiple gas reservoirs within a preset time period;

[0022] a first adjustment unit, configured to iteratively train the first network using the dynamic continuous feature data in the historical feature data, and adjust the network parameters of the first network in each iteration until the loss function of the first network reaches a minimum value;

[0023] The second adjustment unit is used to iteratively train the second network using the static discrete feature data in the historical feature data, and in each iteration, adjust the network parameters of the second network until the loss function of the second network reaches a minimum value.

[0024] Optionally, the first network is a recurrent neural network-long short-term memory network RNN-LSTM, and the second network is a multi-layer perceptron MLP.

[0025] Optionally, the gas reservoir recovery rate prediction model further includes: an attention layer and a fully connected layer, and the device further includes:

[0026] A second acquisition unit is used to acquire historical characteristic data of the target gas reservoir within a preset time period;

[0027] The third adjustment unit is used to iteratively train the attention layer and the fully connected layer using the historical characteristic data of the target gas reservoir within a preset time period. In each iteration, the network parameters of the attention layer and the network parameters of the fully connected layer are adjusted until the loss function of the attention layer reaches a minimum value and the loss function of the fully connected layer reaches a minimum value.

[0028] Optionally, the device further comprises:

[0029] The acquisition unit is used to input the historical characteristic data of the target gas reservoir within a preset time period into a trained gas reservoir recovery rate prediction model including an attention layer and a fully connected layer to obtain the recovery rate of the target gas reservoir in the target time period.

[0030] Optionally, the third adjusting unit is specifically used to:

[0031] Preprocessing the historical characteristic data of the target gas reservoir within a preset time period;

[0032] The attention layer and the fully connected layer are iteratively trained using the preprocessed historical feature data.

[0033] In a third aspect, an embodiment of the present application provides a device, comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes any method described in the first aspect.

[0034] In a fourth aspect, an embodiment of the present application provides a computer storage medium, in which a code is stored. When the code is executed, a device executing the code implements the method described in any one of the first aspects above.

[0035] Compared with the prior art, this application has the following beneficial effects:

[0036] The present application can obtain historical characteristic data of multiple gas reservoirs within a preset time period, and then use the dynamic continuous characteristic data in the historical characteristic data to iteratively train the first network in the gas reservoir recovery rate prediction model. In each iteration, the network parameters of the first network are adjusted until the loss function of the first network reaches the minimum. The second network is iteratively trained using the static discrete characteristic data in the historical characteristic data. In each iteration, the network parameters of the first network are adjusted until the loss function of the second network reaches the minimum, thereby completing the training of the gas reservoir recovery rate prediction model. In the present application, the two deep learning models of the first network and the second network are coupled into a new model. The first network and the second network are trained by obtaining the historical characteristic data of multiple gas reservoirs within a preset time period, thereby realizing the training of the gas reservoir recovery rate prediction model, so that the trained gas reservoir recovery rate prediction model can be used to predict the recovery rate of the gas reservoir, and the two deep learning models are coupled, which can enhance the generalization ability of the model and improve the accuracy and stability of the prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in this embodiment or 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 some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A schematic diagram of the structure of a gas reservoir recovery rate prediction model provided in an embodiment of the present application;

[0039] Figure 2 A flow chart of a method for training a gas reservoir recovery factor prediction model provided in an embodiment of the present application;

[0040] Figure 3 A schematic structural diagram of a specific implementation of a training device for a gas reservoir recovery factor prediction model provided in this application. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0042] It should be noted that the training method and device of a gas reservoir recovery rate prediction model provided in the present application are used in the field of deep learning technology. The above is only an example and does not limit the application field of the method and device provided in the present application.

[0043] A gas reservoir is a geological structure that stores natural gas in underground rocks.

[0044] At present, natural gas can be extracted from gas reservoirs. Since most gas reservoirs are composite gas reservoirs controlled by structures, faults and formations. Various technical countermeasures to improve recovery have been implemented in the process of developing natural gas. However, due to different geological conditions, the effects of the measures vary, resulting in large differences in recovery rates. Therefore, it is necessary to deepen the understanding of the recovery rates of different types of gas reservoirs. In addition, since the production of gas reservoirs is affected by multiple factors such as geological factors, development methods, and the number of wells, the overall law is characterized by high dimensionality, strong nonlinearity, and large differences in development time, which makes the recovery rate of gas reservoirs difficult to predict. Therefore, it is urgent to establish a recovery rate prediction method for gas reservoirs to predict the recovery rate of gas reservoirs, so as to carry out research on technical countermeasures to improve recovery rates and provide a basis for gas reservoir development adjustments.

[0045] In view of this, the inventor has proposed the technical solution of the present application after research. The present application can obtain the historical characteristic data of multiple gas reservoirs within a preset time period, and then use the dynamic continuous characteristic data in the historical characteristic data to iteratively train the first network in the gas reservoir recovery rate prediction model. In each iteration, the network parameters of the first network are adjusted until the loss function of the first network reaches the minimum value. The second network is iteratively trained using the static discrete characteristic data in the historical characteristic data. In each iteration, the network parameters of the first network are adjusted until the loss function of the second network reaches the minimum value, thereby completing the training of the gas reservoir recovery rate prediction model. In the present application, the first network and the second network are coupled into a new model. The first network and the second network are trained by obtaining the historical characteristic data of multiple gas reservoirs within a preset time period, thereby realizing the training of the gas reservoir recovery rate prediction model, so that the trained gas reservoir recovery rate prediction model can be used to predict the recovery rate of the gas reservoir, and the two deep learning models are coupled, which can enhance the generalization ability of the model and improve the accuracy and stability of the prediction.

[0046] The method provided in the embodiment of the present application can be executed by software on a terminal device, which can be, for example, a mobile phone, a tablet computer, a computer, or the like.

[0047] In order to make the technical personnel in this field better understand the present application, the present application is further described in detail below in combination with the drawings and specific implementation methods. The method provided in the present application embodiment is described below by taking the first device as an example.

[0048] The embodiment of the present application provides a method for training a gas reservoir recovery rate prediction model. Before introducing the training method of the present application, the gas reservoir recovery rate prediction model provided by the present application can be introduced first.

[0049] Specifically, the gas reservoir recovery factor prediction model may include a first network and a second network.

[0050] The first network may be a network with feature extraction capability and sequence dependency learning function, such as a recurrent neural network-long short-term memory network RNN-LSTM and a convolutional network-long short-term memory network CNN-LSTM.

[0051] The second network can be a network model that can process sequence data, such as a multi-layer perceptron MLP or a Transformer model. Among them, RNN-LSTM processes continuous data and can capture the time dependency and dynamic changes of data in the time series, and can more effectively utilize historical feature data because they can remember the information of the previous time step, and can also adjust the network structure according to specific tasks, such as increasing the number of layers, changing the number of hidden units, etc., to adapt to different prediction needs.

[0052] MLP can capture the complex nonlinear relationship between gas reservoir recovery and influencing factors through its nonlinear activation function. MLP has a simpler model structure than CNN or RNN, so MLP has a faster training speed and is relatively more flexible.

[0053] Therefore, when the first network in the gas reservoir recovery rate prediction model is RNN-LSTM and the second network is MLP, the gas reservoir recovery rate prediction model will have a better effect on predicting gas reservoir recovery rate than models with other structures, and the prediction accuracy will be higher.

[0054] The following is an introduction using the first layer network of the present application as RNN-LSTM and the second layer network as MLP as an example. The structure of the gas reservoir recovery rate prediction model can be as follows: Figure 1 As shown, Figure 1 The gas reservoir recovery factor prediction model shown may also include other layers, wherein the other layers may be fully connected network layers for coupling the output results of the first network and the second network as the final output results of the gas reservoir recovery factor prediction model.

[0055] In some possible implementations, a Relu activation function can also be set before the fully connected network layer in other layers, where the Relu activation function is a function added to artificial neural networks to help the network learn complex patterns in data. Similar to the neuron-based model in the human brain, it ultimately determines what to emit to the next neuron.

[0056] The models for predicting gas reservoir recovery in the prior art generally only analyze and predict based on one of static data or dynamic data, and cannot fully capture the interdependence between static data and dynamic data in gas reservoir development.

[0057] The gas reservoir recovery rate prediction model in this application can effectively combine dynamic and static data and simulate the gas reservoir development process.

[0058] Specifically, in the process of training the gas reservoir recovery prediction model, the dynamic data (also called dynamic continuous data) can be divided into multiple groups (for example, more than one thousand groups) as input, and the static data (also called static discrete data) remains unchanged, which can effectively simulate the gas reservoir development process. The change of dynamic data can enable the gas reservoir recovery prediction to learn and adapt to the changing conditions of different development stages, while the static data can provide important information on stable factors such as gas reservoir characteristics and geological conditions.

[0059] Furthermore, the gas reservoir recovery rate prediction model in this application is trained using a combination of dynamic data and static data, which can more comprehensively analyze various factors and their changing trends in the gas reservoir development process and reveal the correlation between different factors, thereby improving the accuracy and reliability of the gas reservoir recovery rate prediction model.

[0060] Combined with Figure 1 The gas reservoir recovery rate prediction model shown in the figure introduces the training method of the gas reservoir recovery rate prediction model provided by the present application, such as Figure 2 As shown, the method includes:

[0061] S201: Acquire historical characteristic data of multiple gas reservoirs within a preset time period.

[0062] The first device can obtain historical characteristic data of multiple gas reservoirs within a preset time.

[0063] In some possible implementations, the multiple gas reservoirs may be target gas reservoirs (gas reservoirs for which recovery factor prediction is required) and gas reservoirs with characteristic data similar to the target gas reservoirs. For example, if the recovery factor difference between the gas reservoirs within a preset time period is less than a threshold, the characteristic data of the two gas reservoirs may be considered similar. Of course, this is just an example, and there may be other implementations, which are not limited here.

[0064] In some possible implementations, the multiple gas reservoirs may not include the target gas reservoir, and the multiple gas reservoirs are multiple gas reservoirs having characteristic data similar to the target gas reservoir.

[0065] The following is an example in which the target gas reservoir is not included in the multiple gas reservoirs and the multiple gas reservoirs are multiple gas reservoirs with characteristic data similar to the target gas reservoir.

[0066] The first device obtains historical characteristic data of multiple gas reservoirs similar to the characteristic data of the target gas reservoir within a preset time, wherein the preset time can be set according to demand.

[0067] For example, if the target gas reservoir is gas reservoir A, and multiple gas reservoirs with similar characteristic data to the target gas reservoir are gas reservoir B, gas reservoir C, and gas reservoir D, and the recovery rate of the target gas reservoir in July needs to be predicted, then the historical characteristic data of gas reservoir B from January to June, the historical characteristic data of gas reservoir C from January to June, and the historical characteristic data of gas reservoir D from January to June can be obtained. For another example, if the recovery rate of the target gas reservoir in August needs to be predicted, the historical characteristic data of gas reservoir B from January to July, the historical characteristic data of gas reservoir C from January to July, and the historical characteristic data of gas reservoir D from January to July can be obtained. Of course, this is just an example and is not limited here.

[0068] Specifically, the acquired historical feature data may include dynamic continuous feature data and static discrete feature data. Among them, dynamic continuous feature data is data that changes over time and can take any real value. This type of data not only has the characteristics of continuity, but also includes dynamic changes in time series.

[0069] Exemplarily, dynamic continuous characteristic data may include monthly water-gas ratio, gas reservoir recovery rate within a preset time period, total monthly water-gas ratio of the gas reservoir within a preset period, number of production wells, well-controlled area, and number of water-producing wells.

[0070] Among them, the monthly water-gas ratio refers to the volume ratio of water contained in unit volume of natural gas in the gas reservoir every month during the natural gas production process. The water-gas ratio can be determined by the ratio of the total amount of water produced in the gas reservoir every month and the volume of natural gas produced every month.

[0071] The recovery rate refers to the ratio of the cumulative amount of natural gas produced in a gas reservoir to the total volume of natural gas in the gas reservoir in its original state.

[0072] The water-gas ratio stage refers to the volume ratio of water contained in a unit volume of natural gas in a gas reservoir within a preset period (for example, every three months) during the natural gas production process. The water-gas ratio can be determined by the total amount of water produced in the gas reservoir within the preset period and the volume ratio of the natural gas produced within the preset period.

[0073] The number of production wells refers to the number of gas wells producing natural gas in a gas reservoir. For example, the time period for obtaining the number of production wells may be in months, and the number of production wells per month in the gas reservoir may be obtained. It should be noted that the number of production wells per month in the gas reservoir may be different.

[0074] Well controlled area refers to the ratio of the total area of ​​gas wells producing natural gas in the gas reservoir to the number of production wells in the time period. For example, if the time period is a month, then the well controlled area refers to the ratio of the total area of ​​gas wells producing natural gas in the gas reservoir in the current month to the number of production wells in the current month.

[0075] The number of water-producing wells refers to the number of wells that produce water in a gas reservoir within a preset stage. For example, if the preset stage is a month, the number of water-producing wells refers to the number of wells that produce water in the gas reservoir every month.

[0076] Static continuous characteristic data may include: water-bearing area, gas-water distribution characteristics, effective thickness, porosity, original formation pressure, geological reserves, total number of development wells, total number of water-producing wells, number of invalid wells, stable production capacity scale, unimpeded flow (single well parameter statistics), external transmission pressure, wellhead pressure after pressurization, water invasion intensity, gas production rate, effective thickness (single well median), effective thickness (single average), effective permeability (single well median) and effective permeability (single well average), etc., data that do not change with time.

[0077] Water-Bearing Area: The area or region occupied by water in an underground gas reservoir.

[0078] Gas-water distribution characteristics: describes the distribution of natural gas and water in underground gas reservoirs. It can include the relative proportion of gas and water, the distribution pattern in the reservoir (such as gas above water, water above gas, etc.), and the vertical and horizontal distribution range.

[0079] Effective thickness: refers to the thickness of the effective rock layer in the gas reservoir that can be used for gas flow. It can be the effective flow area of ​​gas.

[0080] Porosity: The percentage of pores (or holes) in a reservoir rock, used to describe the amount of space in the rock that can hold fluids (such as natural gas and water).

[0081] Original formation pressure: refers to the initial pressure state when the natural gas reservoir is formed, which can be used to calculate the production capacity and production strategy of the gas reservoir.

[0082] Geological reserves: refers to the amount of natural gas that can be reasonably expected to be extracted under known technical and economic conditions.

[0083] Total number of development wells: refers to the total number of wells drilled and put into use when developing a gas reservoir.

[0084] Total water-producing wells: The total number of wells used to discharge or treat water during the development of a gas reservoir.

[0085] Number of invalid wells: refers to the number of wells that do not produce effective production or are abandoned during the development of gas reservoirs.

[0086] Stable production capacity scale: refers to the amount of natural gas that a gas reservoir can continuously produce under stable production conditions.

[0087] Unobstructed flow rate (single well parameter statistics): refers to the maximum gas production rate that a single well can achieve under unrestricted conditions (without production capacity constraints).

[0088] Output pressure: refers to the pressure when natural gas is exported from a gas reservoir or gas well to a pipeline system or processing facility.

[0089] Wellhead pressure after boosting: The pressure of natural gas when it flows out of the wellhead after being boosted by the boosting equipment.

[0090] Water invasion intensity: refers to the degree of water invasion in the gas reservoir or the intensity of mixing of water and natural gas.

[0091] Gas production rate: refers to the amount of natural gas produced from a single well or the entire gas reservoir per unit time.

[0092] Effective thickness (median of a single well): refers to the effective gas reservoir thickness at the median level of a single well.

[0093] Effective thickness (single well average): refers to the effective gas reservoir thickness at the average level of a single well.

[0094] Effective permeability (median of a single well): refers to the effective permeability at the median level of a single well, which can be used to describe the ability of gas to flow in the reservoir.

[0095] Effective permeability (single well average): refers to the effective permeability of the average level of a single well, which can be used to describe the ability of gas to flow in the reservoir.

[0096] S202: Iteratively train the first network using the dynamic continuous feature data in the historical feature data, and adjust the network parameters of the first network in each iteration until the loss function of the first network reaches a minimum value.

[0097] The first device can use the dynamic continuous feature data in the historical feature data to iteratively train the first network. During each iteration, the network parameters of the first network (including the weights and biases of neurons) are adjusted until the loss function of the first network reaches a minimum value.

[0098] Specifically, the first layer network can be composed of a large number of neurons connected to each other. The formula of a single neuron can be as follows:

[0099]

[0100] Among them, x i As the input of the neuron; ω ij , b i As the weight and bias of a single neuron; yj is the signal x i The corresponding output sum. It should be noted that those skilled in the art can set the number of neurons in the first network according to the needs. The appropriate number of neurons can improve the fitting ability of the network, reduce the error in the experiment, allow the network to better explore the hidden relationship between the data, and improve the network learning ability. For example, the hidden layer is set to be twice the input layer, and the quantitative relationship between the hidden layers is at least twice, so that the learning ability of the first network can be relatively good.

[0101] The first device can determine the true label of the first network training, and then input the dynamic continuous characteristic data in the historical characteristic data of each gas reservoir into the initial first network to obtain the predicted value of the gas reservoir recovery rate. Then, the loss function of the first network can be used to determine the difference between the true label and the predicted value, and then the network parameters of the first network can be adjusted according to the difference between the true label and the predicted value.

[0102] Repeat the above steps until the loss function reaches the minimum value or the value of the loss function does not decrease significantly, stop training the first network, and complete the training of the first network.

[0103] S203: Iteratively train the second network using the static discrete feature data in the historical feature data, and adjust the network parameters of the second network in each iteration until the loss function of the second network reaches a minimum value.

[0104] The first device can use the static discrete feature data in the historical feature data to iteratively train the second network. During each iteration, the network parameters of the second network are adjusted until the loss function of the second network reaches a minimum value or the value of the loss function does not decrease significantly, and the training of the second network is stopped, thereby completing the training of the second network.

[0105] The network structure and specific training principle of the second network are similar to those of the first network in step S102, and are not described in detail here.

[0106] When the training of the first network and the second network is completed, it can be determined that the training of the gas reservoir recovery factor prediction model is completed.

[0107] In some possible implementation methods, after the gas reservoir recovery rate prediction model is trained, the present application can also obtain the historical characteristic data of the target gas reservoir within a preset time period, and then input the historical characteristic data of the target gas reservoir within the preset time period into the trained gas reservoir recovery rate prediction model, and then obtain the recovery rate of the target gas reservoir in the target time period, thereby realizing the recovery rate of the target gas reservoir in the target time period using the trained gas reservoir recovery rate prediction model, wherein the target time period refers to the time period that needs to be predicted.

[0108] The present application can obtain historical characteristic data of multiple gas reservoirs within a preset time period, and then use the dynamic continuous characteristic data in the historical characteristic data to iteratively train the first network in the gas reservoir recovery rate prediction model. In each iteration, the network parameters of the first network are adjusted until the loss function of the first network reaches the minimum. The second network is iteratively trained using the static discrete characteristic data in the historical characteristic data. In each iteration, the network parameters of the first network are adjusted until the loss function of the second network reaches the minimum, thereby completing the training of the gas reservoir recovery rate prediction model. In the present application, the two deep learning models of the first network and the second network are coupled into a new model. The first network and the second network are trained by obtaining the historical characteristic data of multiple gas reservoirs within a preset time period, thereby realizing the training of the gas reservoir recovery rate prediction model, so that the trained gas reservoir recovery rate prediction model can be used to predict the recovery rate of the gas reservoir, and the two deep learning models are coupled, which can enhance the generalization ability of the model and improve the accuracy and stability of the prediction.

[0109] The above embodiment can be referred to as the first embodiment. On the basis of the above embodiment, in order to improve the accuracy and efficiency of the gas reservoir recovery rate prediction model in predicting the gas reservoir recovery rate, an attention mechanism and a fully connected network can be introduced into the gas reservoir recovery rate prediction model trained in the first embodiment. This embodiment can be referred to as the second embodiment, and the second embodiment is introduced below.

[0110] Specifically, an attention layer and a fully connected layer can be added after the trained network model structure of the gas reservoir recovery prediction model. Among them, the fully connected layer is a fully connected neural network model, which is a multi-layer perceptron (MLP). The principle of the perceptron is to find the most reasonable and robust hyperplane between categories. The most representative perceptron is the SVM support vector machine algorithm. Neural networks draw on both perceptrons and bionics. Generally speaking, after receiving a signal, animal nerves will send to each neuron. After receiving the input, each neuron will generate an output signal based on its own judgment and then aggregate it to achieve recognition and classification of the information source.

[0111] The attention layer can make the model more accurate and effective when processing sequence data. In traditional neural networks, the output of each neuron depends only on the output of all neurons in the previous layer, while in the attention mechanism, the output of each neuron not only depends on the output of all neurons in the previous layer, but can also be weighted according to different parts of the input data, that is, different weights are given to different parts. This allows the model to pay more attention to the key information in the input sequence, thereby improving the accuracy and efficiency of the model.

[0112] The new gas reservoir recovery rate prediction model generated by adding the attention layer and the fully connected layer can be trained.

[0113] Specifically, the historical characteristic data of the target gas reservoir within a preset time period is obtained, and the historical characteristic data of the target gas reservoir within a preset time period is used to iteratively train the attention layer and the fully connected layer. In each iteration, the network parameters of the attention layer and the network parameters of the fully connected layer are adjusted until the loss function of the attention layer reaches the minimum value or has no obvious decrease and the loss function of the fully connected layer reaches the minimum value or has no obvious decrease. The training of the new gas reservoir recovery rate prediction model is completed. It should be noted that in the process of training the new gas reservoir recovery rate prediction model, the network parameters of the first network and the second network are fixed.

[0114] In some possible implementation methods, the historical characteristic data of the target gas reservoir within a preset time period may be preprocessed, wherein the preprocessing may include data normalization and data cleaning, etc., the purpose of which is to convert the original data into a format suitable for model training and improve the data quality.

[0115] Then, the preprocessed historical feature data can be used to iteratively train the attention layer and the fully connected layer, thereby improving the efficiency of model training.

[0116] After the training of the new gas reservoir recovery rate prediction model is completed, the recovery rate of the gas reservoir can be predicted using the new gas reservoir recovery rate prediction model. That is, the historical characteristic data of the target gas reservoir in the preset time period is obtained, and the historical characteristic data of the target gas reservoir in the preset time period is input into the trained new gas reservoir recovery rate prediction model to obtain the recovery rate of the target gas reservoir in the target period.

[0117] In this embodiment, an attention layer and a fully connected layer are added after the trained network model structure of the gas reservoir recovery rate prediction model. For data with time series characteristics in the gas reservoir, the attention layer can help the model capture long-term dependencies, which is very important for predicting long-term recovery rate trends. In addition, the attention mechanism can assign different weights to different parts of the model input data, so that the model can focus on features that have a greater impact on the prediction results, such as water-gas ratio per month, number of production wells, etc. The fully connected layer can learn the complex mapping relationship from input features to prediction targets, thereby increasing the expressive power of the model.

[0118] In this way, by adding an attention layer and a fully connected layer after the trained network model structure of the gas reservoir recovery rate prediction model, and training the attention layer and the fully connected layer, the accuracy and efficiency of the gas reservoir recovery rate prediction model in predicting the gas reservoir recovery rate can be improved.

[0119] The above are some specific implementation methods of the training method of the gas reservoir recovery rate prediction model provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization. The device and the training method of the gas reservoir recovery rate prediction model described above can be referenced to each other.

[0120] Figure 3 The structural block diagram of the training device of the gas reservoir recovery rate prediction model provided by the embodiment of the present invention is referred to as the specific implementation method three. The gas reservoir recovery rate prediction model includes a first layer network and a second layer network. Figure 3 The device may include:

[0121] The first acquisition unit 300 is used to acquire historical characteristic data of multiple gas reservoirs within a preset time period;

[0122] A first adjustment unit 310 is used to iteratively train the first network using the dynamic continuous feature data in the historical feature data, and adjust the network parameters of the first network in each iteration until the loss function of the first network reaches a minimum value;

[0123] The second adjustment unit 320 is used to iteratively train the second network using the static discrete feature data in the historical feature data, and adjust the network parameters of the second network in each iteration until the loss function of the second network reaches a minimum value.

[0124] Optionally, the first network is a recurrent neural network-long short-term memory network RNN-LSTM, and the second network is a multi-layer perceptron MLP.

[0125] Optionally, the gas reservoir recovery rate prediction model further includes: an attention layer and a fully connected layer, and the device further includes:

[0126] A second acquisition unit is used to acquire historical characteristic data of the target gas reservoir within a preset time period;

[0127] The third adjustment unit is used to iteratively train the attention layer and the fully connected layer using the historical characteristic data of the target gas reservoir within a preset time period. In each iteration, the network parameters of the attention layer and the network parameters of the fully connected layer are adjusted until the loss function of the attention layer reaches a minimum value and the loss function of the fully connected layer reaches a minimum value.

[0128] Optionally, the device further comprises:

[0129] The acquisition unit is used to input the historical characteristic data of the target gas reservoir within a preset time period into a trained gas reservoir recovery rate prediction model including an attention layer and a fully connected layer to obtain the recovery rate of the target gas reservoir in the target time period.

[0130] Optionally, the third adjusting unit is specifically used to:

[0131] Preprocessing the historical characteristic data of the target gas reservoir within a preset time period;

[0132] The attention layer and the fully connected layer are iteratively trained using the preprocessed historical feature data.

[0133] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0134] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method described in any embodiment of the present application.

[0135] The computer storage medium stores codes, and when the codes are executed, a device executing the codes implements the method described in any embodiment of the present application.

[0136] The "first" and "second" in the names such as "first" and "second" (if any) mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0137] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0138] Each embodiment in this specification is described in a progressive manner, and the same and similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0139] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A training method for a gas reservoir recovery factor prediction model, It is characterized in that The gas reservoir recovery rate prediction model includes a first network and a second network, and the method includes: Obtain historical characteristic data of multiple gas reservoirs within a preset time period; Iteratively training the first network using the dynamic continuous feature data in the historical feature data, and adjusting the network parameters of the first network in each iteration until the loss function of the first network reaches a minimum value; The second network is iteratively trained using the static discrete feature data in the historical feature data, and in each iteration, the network parameters of the second network are adjusted until the loss function of the second network reaches a minimum value.

2. The method according to claim 1, It is characterized in that The first network is a recurrent neural network-long short-term memory network RNN-LSTM, and the second network is a multi-layer perceptron MLP.

3. The method according to claim 2, It is characterized in that The gas reservoir recovery rate prediction model further includes: an attention layer and a fully connected layer, and the method further includes: Obtain historical characteristic data of the target gas reservoir within a preset time period; The attention layer and the fully connected layer are iteratively trained using the historical characteristic data of the target gas reservoir within a preset time period. In each iteration, the network parameters of the attention layer and the network parameters of the fully connected layer are adjusted until the loss function of the attention layer and the loss function of the fully connected layer reach a minimum value.

4. The method according to claim 3, It is characterized in that The method further comprises: The historical characteristic data of the target gas reservoir within a preset time period is input into a trained gas reservoir recovery rate prediction model including an attention layer and a fully connected layer to obtain the recovery rate of the target gas reservoir in the target period.

5. The method according to claim 3, It is characterized in that The iterative training of the attention layer and the fully connected layer using the historical feature data of the target gas reservoir within a preset time period includes: Preprocessing the historical characteristic data of the target gas reservoir within a preset time period; The attention layer and the fully connected layer are iteratively trained using the preprocessed historical feature data.

6. The method according to claim 1, It is characterized in that The dynamic continuous characteristic data of the historical characteristic data include: the monthly water-gas ratio of the gas reservoir, the recovery rate of the gas reservoir in a preset time period, the water-gas ratio of the gas reservoir in the total months within a preset period, the number of production wells of the gas reservoir in each preset stage, the well-controlled area of ​​the gas reservoir and the number of wells producing water in the gas reservoir in each preset stage.

7. A training device for a gas reservoir recovery factor prediction model, It is characterized in that The gas reservoir recovery rate prediction model includes a first network and a second network, and the device includes: A first acquisition unit is used to acquire historical characteristic data of multiple gas reservoirs within a preset time period; a first adjustment unit, configured to iteratively train the first network using the dynamic continuous feature data in the historical feature data, and adjust the network parameters of the first network in each iteration until the loss function of the first network reaches a minimum value; The second adjustment unit is used to iteratively train the second network using the static discrete feature data in the historical feature data, and in each iteration, adjust the network parameters of the second network until the loss function of the second network reaches a minimum value.

8. The device according to claim 7, It is characterized in that The first network is a recurrent neural network-long short-term memory network RNN-LSTM, and the second network is a multi-layer perceptron MLP.

9. The device according to claim 8, It is characterized in that The gas reservoir recovery rate prediction model further includes: an attention layer and a fully connected layer, and the device further includes: A second acquisition unit is used to acquire historical characteristic data of the target gas reservoir within a preset time period; The third adjustment unit is used to iteratively train the attention layer and the fully connected layer using the historical characteristic data of the target gas reservoir within a preset time period. In each iteration, the network parameters of the attention layer and the network parameters of the fully connected layer are adjusted until the loss function of the attention layer reaches a minimum value and the loss function of the fully connected layer reaches a minimum value.

10. The device according to claim 9, It is characterized in that The device also includes: The acquisition unit is used to input the historical characteristic data of the target gas reservoir within a preset time period into a trained gas reservoir recovery rate prediction model including an attention layer and a fully connected layer to obtain the recovery rate of the target gas reservoir in the target time period.