Formation pressure prediction model optimization method and device, electronic equipment and storage medium

Abstract

The invention discloses a model optimization method and device for formation pressure prediction, electronic equipment and a storage medium, and the method comprises the steps: obtaining logging-while-drilling data of a first exploratory well and abnormal pressure factors of each depth range, and carrying out the arrangement to obtain a source domain sample; taking an actually measured pressure value or an abnormal state corresponding to the source domain sample of the first exploratory well as a source domain label; performing first training on a preset prediction model by using the source domain sample and the source domain label to obtain a first model; obtaining a target domain sample of a second exploratory well, and performing parameter fine tuning on the first model by adopting transfer learning based on the target domain sample to obtain a second model; wherein the second model is used for predicting the formation pressure of the second exploratory well. By fusing the abnormal pressure factor and the transfer learning strategy, a modeling framework of real-time calculation of the formation pressure is expanded, model optimization of formation pressure prediction can be efficiently realized, and the method can be widely applied to the technical field of data prediction.

Description

Technical Field

[0001] The present invention relates to the field of data prediction technology, and in particular to a model optimization method, device, electronic equipment and storage medium for formation pressure prediction. Background Art

[0002] In the field of oil and gas exploration, accurate prediction of formation pressure is a core technology for ensuring drilling safety, optimizing wellbore structure design, and reducing engineering risks. Traditional formation pressure prediction methods mainly rely on empirical formulas or physical models of single logging curves (such as acoustic time difference, resistivity, etc.), and have the following limitations: Traditional methods have limited ability to extract deep correlation features of multi-dimensional logging-while-drilling data, and it is difficult to comprehensively reflect the formation mechanism of abnormal pressure under complex geological conditions. Different exploration wells are located in different geological structures, lithologic combinations, and pressure genesis mechanisms. Models trained based on single exploration well data are difficult to directly apply to new well predictions. A large amount of labeled data needs to be re-collected, resulting in high costs and low timeliness.

[0003] In recent years, although deep learning technologies (such as convolutional neural networks and recurrent neural networks) have shown potential in well logging data analysis, their dependence on large amounts of labeled data is in stark contrast to the reality of scarce target well data in the early stages of exploration. Summary of the Invention

[0004] The present invention aims to at least partially address the limitations of related technologies. To this end, the present invention provides a model optimization method, device, electronic device, and storage medium for formation pressure prediction, which can efficiently optimize the model for formation pressure prediction.

[0005] In one aspect, an embodiment of the present invention provides a model optimization method for formation pressure prediction, comprising the following steps:

[0006] Obtain the logging while drilling data of the first exploration well and the abnormal pressure factors in each depth range, and then organize them to obtain source domain samples;

[0007] Using the measured pressure value or abnormal state corresponding to the source domain sample of the first exploration well as the source domain label;

[0008] Performing a first training on a preset prediction model using source domain samples and source domain labels to obtain a first model;

[0009] Among them, the prediction models include convolutional neural networks, recurrent neural networks, and fully connected networks;

[0010] Obtain target domain samples of the second exploration well, and fine-tune the parameters of the first model using transfer learning based on the target domain samples to obtain a second model;

[0011] The second model is used to predict the formation pressure of the second exploration well.

[0012] Optionally, obtaining the abnormal pressure factor of each depth range of the first exploration well includes the following steps:

[0013] Obtaining acoustic wave values and density values at various depth points of the first exploration well;

[0014] Based on the acoustic wave value and density value at each depth point, the slope of the interaction graph between the acoustic wave and density at each depth range of the first exploration well is calculated using a sliding window;

[0015] Determine the abnormal pressure factor of each depth range of the first exploration well according to the slope of the interaction graph;

[0016] Among them, the expression of abnormal pressure factor is:

[0017]

[0018] Where V P(i) Represents the acoustic wave value at the i-th depth point; R HOB(i) represents the density value of the i-th depth point; W represents the width of the sliding window; G i It represents the abnormal pressure factor in the W depth range after the i-th depth point.

[0019] Optionally, before the step of arranging and obtaining source domain samples, the method further includes the following steps:

[0020] Normalize abnormal stress factors.

[0021] Optionally, normalizing the abnormal pressure factors comprises the following steps:

[0022] The abnormal pressure factor is normalized using a logarithmic function;

[0023] The expression for standardization is:

[0024] C=lg(G+1)

[0025] Where G represents the original abnormal pressure factor; C represents the abnormal pressure factor after normalization; and lg is the applied logarithmic function.

[0026] Optionally, performing a first training on a preset prediction model using source domain samples and source domain labels to obtain a first model includes the following steps:

[0027] Input the source domain samples into the prediction model to obtain the prediction results;

[0028] A loss function is constructed based on the prediction results and source domain labels corresponding to the source domain samples. The model parameters of the prediction model are then adjusted through convergence optimization feedback of the loss function until the loss function meets the preset conditions. The prediction model is then trained.

[0029] Among them, the expression of the loss function is:

[0030]

[0031] Where, L s represents the loss function; N s Indicates the number of source domain samples; represents the i-th source domain sample; Indicates the source domain label corresponding to the i-th source domain sample; Represents the prediction result corresponding to the i-th source domain sample; L represents the sub-loss function applied to the prediction result corresponding to the i-th source domain sample and the source domain label.

[0032] Optionally, fine-tuning parameters of the first model using transfer learning based on target domain samples includes the following steps:

[0033] Get the target domain label corresponding to the target domain sample;

[0034] Input the target domain sample into the first model for processing to obtain the predicted output;

[0035] A target loss function is constructed based on the predicted output and target domain label corresponding to the target domain sample, and the model parameters of the first model are fine-tuned based on the target loss function and a preset training strategy to obtain a prediction model;

[0036] Among them, the expression of the objective loss function is:

[0037]

[0038] Where, L t represents the loss function; N t Indicates the number of target domain samples; represents the i-th target domain sample; represents the target domain label corresponding to the i-th target domain sample; Represents the predicted output corresponding to the i-th target domain sample; L represents the sub-loss function applied to the predicted output corresponding to the i-th target domain sample and the target domain label.

[0039] Optionally, fine-tuning the model parameters of the first model based on the target loss function in combination with a preset training strategy includes the following steps:

[0040] Freeze the convolutional neural network in the first model;

[0041] Fine-tune the model parameters of the first model based on the loss function and the preset learning rate;

[0042] Among them, the learning rate is preset by the following relationship:

[0043]

[0044] Where θ t represents the model parameters of the first model after fine-tuning with target domain samples; θ s represents the model parameters of the first model; α represents the learning rate; Represents the target loss function for the model parameter θ t gradient.

[0045] On the other hand, an embodiment of the present invention provides a model optimization device for formation pressure prediction, comprising:

[0046] The first module is used to obtain the logging while drilling data of the first exploration well and the abnormal pressure factors in each depth range, and then organize them to obtain source domain samples;

[0047] The second module is configured to use the measured pressure value or abnormal state corresponding to the source domain sample of the first exploration well as the source domain label;

[0048] The third module is configured to perform a first training on a preset prediction model using source domain samples and source domain labels to obtain a first model;

[0049] Among them, the prediction models include convolutional neural networks, recurrent neural networks, and fully connected networks;

[0050] The fourth module is used to obtain target domain samples of the second exploration well, and fine-tune the parameters of the first model using transfer learning based on the target domain samples to obtain a second model;

[0051] The second model is used to predict the formation pressure of the second exploration well.

[0052] Optionally, the apparatus further includes a fifth module configured to perform the following operations:

[0053] Normalize abnormal stress factors.

[0054] On the other hand, an embodiment of the present invention provides an electronic device, comprising: a processor and a memory; the memory is used to store programs; the processor executes the program to implement the above-mentioned model optimization method for formation pressure prediction.

[0055] On the other hand, an embodiment of the present invention provides a computer storage medium storing a program executable by a processor. When the program is executed by the processor, it is used to implement the above-mentioned model optimization method for formation pressure prediction.

[0056] The present invention obtains logging-while-drilling (LWD) data and abnormal pressure factors at various depths from a first exploration well, and then organizes them to obtain source domain samples. The measured pressure values or abnormal conditions corresponding to the source domain samples of the first exploration well are used as source domain labels. A preset prediction model is first trained using the source domain samples and source domain labels to obtain a first model, wherein the prediction model includes a convolutional neural network, a recurrent neural network, and a fully connected network. Target domain samples are obtained from a second exploration well, and transfer learning is used to fine-tune the parameters of the first model based on the target domain samples to obtain a second model, wherein the second model is used to predict formation pressure in the second exploration well. The present invention trains an initial model based on the source domain samples and source domain labels of the first exploration well. Through transfer learning, prior knowledge, such as formation pressure response patterns and feature association patterns, is transferred to the target domain of the second exploration well, significantly reducing the target well's reliance on large amounts of labeled data and addressing the data scarcity problem in the early stages of exploration. Furthermore, the present invention utilizes a transfer learning parameter fine-tuning strategy to address data distribution differences in the target domain. While retaining the general feature extraction capabilities of the source domain, the model's deep parameters are adaptively adjusted to accommodate the specificity of the target well, ensuring cross-well generalization. By integrating abnormal pressure factors with transfer learning strategies, the present invention expands the modeling framework for real-time calculation of formation pressure and can efficiently achieve model optimization for formation pressure prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the technical solution of the present invention and do not constitute a limitation to the technical solution of the present invention.

[0058] Figure 1 This is a schematic diagram of an implementation environment of a model optimization method for formation pressure prediction provided by an embodiment of the present invention;

[0059] Figure 2 1 is a flow chart of a model optimization method for formation pressure prediction provided by an embodiment of the present invention;

[0060] Figure 3 Schematic diagram of acoustic wave and density logging curves of three well examples provided in an embodiment of the present invention;

[0061] Figure 4 A schematic diagram of an example of a model structure of a prediction model provided by an embodiment of the present invention;

[0062] Figure 5 A schematic diagram of a stratum sedimentary characteristics extraction unit structure based on an improved attention mechanism provided in an embodiment of the present invention;

[0063] Figure 6 A schematic diagram showing the correspondence between abnormal pressure factors and abnormal formation pressures provided in an embodiment of the present invention;

[0064] Figure 7 A schematic diagram of a transfer learning strategy framework based on equivalent depth provided by an embodiment of the present invention;

[0065] Figure 8 A schematic diagram of the distribution of formation pressure calculation accuracy results before and after optimization of the equivalent depth-based transfer learning strategy provided by an embodiment of the present invention;

[0066] Figure 9 A schematic diagram of the curves showing the calculated and true pressure values of three wells after optimization of the transfer learning strategy based on equivalent depth provided in an embodiment of the present invention;

[0067] Figure 10 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0069] It should be noted that although the system diagrams illustrate functional module divisions and the flowcharts illustrate a logical sequence, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the system or the sequence in the flowcharts. The terms "first / S100," "second / S200," and the like in the specification, claims, and drawings are used to distinguish similar objects and are not necessarily intended to describe a specific sequence or precedence.

[0070] References to "embodiments" in this disclosure mean that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the disclosure. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0071] It is understandable that the model optimization method for formation pressure prediction provided in the embodiment of the present invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal is a smart phone, tablet computer, laptop computer, desktop computer, etc., but is not limited to this.

[0072] To facilitate understanding of the technical solutions of the present invention, the following are first explained regarding the technical features that may appear in the embodiments of the present invention:

[0073] like Figure 1 FIG. 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention. Figure 1 , the implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected to the network in a wireless or wired manner to complete data transmission and exchange.

[0074] Server 101 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0075] In addition, server 101 can also be a node server in a blockchain network. Blockchain is a new application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, and encryption algorithm.

[0076] The terminal 102 may be a smart phone, tablet computer, laptop computer, desktop computer, smart speaker, smart watch, etc., but is not limited thereto. The terminal 102 and the server 101 may be connected directly or indirectly via wired or wireless communication, which is not limited in this embodiment of the present invention.

[0077] Based on the example Figure 1In the implementation environment shown, an embodiment of the present invention provides a model optimization method for formation pressure prediction. The following is explained using the example of the model optimization method for formation pressure prediction being applied in the server 101. It can be understood that the model optimization method for formation pressure prediction can also be applied in the terminal 102.

[0078] Reference Figure 2 , Figure 2 The flowchart of the model optimization method for formation pressure prediction applied to the server provided in the embodiment of the present invention is provided. The execution subject of the model optimization method for formation pressure prediction can be any of the aforementioned computer devices (including servers or terminals). Figure 2 , the method comprises the following steps:

[0079] S100, obtaining logging while drilling data of a first exploration well and abnormal pressure factors in various depth ranges, and then arranging them to obtain source domain samples;

[0080] It should be noted that, in some embodiments, obtaining the abnormal pressure factor for each depth range of the first exploration well may include the following steps: obtaining the acoustic wave value and density value at each depth point of the first exploration well; based on the acoustic wave value and density value at each depth point, using a sliding window to calculate the slope of an interaction graph of acoustic wave and density for each depth range of the first exploration well; and determining the abnormal pressure factor for each depth range of the first exploration well according to the slope of the interaction graph; wherein the expression of the abnormal pressure factor is:

[0081]

[0082] Where V P(i) Represents the acoustic wave value at the i-th depth point; R HOB(i) represents the density value of the i-th depth point; W represents the width of the sliding window; G i It represents the abnormal pressure factor in the W depth range after the i-th depth point.

[0083] For example, in some specific embodiments, the present invention combines the formation abnormal pressure formation cause mechanism and its characteristics, proposes a model optimization method that considers the formation abnormality cause (i.e., abnormal pressure factor), and innovatively proposes the concept of formation abnormal pressure cause factor as the input of the model. Specifically, the present invention proposes the concept of abnormal pressure cause factor (G) for the first time, aiming to provide quantifiable abnormal pressure characteristics for the pressure calculation model. The quantification principle is: use a sliding window to calculate the slope of the acoustic wave-density intersection diagram in this section at this depth, such as Figure 3 As shown in the red box, the calculation formula is as follows:

[0084]

[0085] Among them, V P(i)Indicates the sound wave value at depth point i, R HOB(i) represents the density value at depth point i, W represents the width of the sliding window, that is, the abnormal pressure factor G calculated within the depth range of W i .

[0086] In some embodiments, before the step of arranging and obtaining source domain samples, the method may further include the following step: performing standardization processing on abnormal pressure factors.

[0087] For example, in some specific implementations, when monitoring the causes of abnormal pressure, extremely large values often represent specific abnormal conditions. Normalization compresses these extreme values, preventing the model from correctly understanding their importance in monitoring. Conversely, excessively large values may interfere with model training and affect the learning of other features. Therefore, standardization is a more appropriate processing strategy, ensuring that the model effectively captures the key role of this feature while avoiding distortion of outliers during data processing.

[0088] In some embodiments, the normalization process of the abnormal pressure factor may include the following steps: normalizing the abnormal pressure factor using a logarithmic function; wherein the normalization process expression is:

[0089] C=lg(G+1)

[0090] Where G represents the original abnormal pressure factor; C represents the abnormal pressure factor after normalization; and lg is the applied logarithmic function.

[0091] For example, in some specific implementations, logarithmic normalization is an effective data preprocessing method suitable for processing features with extreme values, skewed distributions, or exponential growth. By applying a logarithmic transformation to each data point, the range of large values can be compressed, reducing the interference of extreme values on model training, while smoothing the data distribution to make the data more consistent with the model training requirements. The principle is as follows:

[0092] C=lg(G+1)

[0093] Where G is the original abnormal formation pressure factor, and C is the normalized abnormal formation pressure factor. The addition of 1 (G+1) is to avoid infinity or mathematical errors when the logarithmic function processes zero or negative values. To ensure that the model can effectively capture the key characteristics of the parameters while avoiding distortion of outliers during data processing, the present invention uses the lg function as the function for normalizing the logarithmic function.

[0094] S200: Using the measured pressure value or abnormal state corresponding to the source domain sample of the first exploration well as a source domain label;

[0095] For example, in some specific implementations, the data in the source domain is recorded as s ={(X s ,Y s )}, where X s is the feature of the source domain sample, Y s is the label of the source domain sample, the number of samples is N s , the label Y of the source domain s Indicates the calculated value or abnormal state of formation pressure.

[0096] S300: Performing a first training on a preset prediction model using source domain samples and source domain labels to obtain a first model;

[0097] Among them, the prediction models include convolutional neural networks, recurrent neural networks, and fully connected networks;

[0098] For example, in some specific embodiments, the present invention utilizes the mechanism characteristics of formation deposition and logging while drilling data to construct a high-precision, real-time calculation model of formation pressure during drilling from the perspective of local feature extraction of formation space, time series model feature processing, and machine learning related principles. The schematic diagram of the model structure is shown in FIG. Figure 4 As shown, specifically:

[0099] Convolutional neural networks (CNNs) are used to perceive multiple local spatial features. LWD data is multidimensional and highly autocorrelated, reflecting changes in the current (local) formation state and, in turn, characterizing the changing trends in effective stress. This high-dimensional LWD input contains rich spatiotemporal features. Therefore, in the formation pressure calculation model (i.e., the prediction model), extracting these high-dimensional, autocorrelated features through a CNN helps the model better perceive the spatial information of the local formation.

[0100] The stratigraphic sedimentary characteristics extraction model adopts a variable sequence feature quantization unit based on a recurrent neural network and a stratigraphic sedimentary characteristics extraction unit based on an improved attention mechanism, where:

[0101] This paper utilizes LSTM to capture the long-range dependencies of various well logging data. LSTM is a type of recurrent neural network specifically designed to address the difficulty traditional RNNs have in capturing long-range dependencies in long sequences of data. By introducing a "memory unit" and a "gating mechanism" (including input gates, forget gates, and output gates), LSTM can selectively remember or forget information during training, effectively preserving the memory of important historical data while filtering out irrelevant information.

[0102] Moreover, in the present invention, since the input parameters are multi-source logging data, these parameters are still multi-dimensional physical and chemical features that characterize the sedimentary characteristics of the formation after spatial feature extraction and sequential feature extraction. Conventional methods only focus on the dynamic assignment (autocorrelation) of a single logging parameter at different depth points, which often cannot capture the mutual correlation between different logging data. Therefore, the present invention proposes an improved attention mechanism to extract the sedimentary characteristics of the formation, and its structure is as follows: Figure 5 As shown:

[0103] Where Z = Concat(Z1, Z1, Z1, ..., Z n ), Z1, Z1, Z1, ..., Z n is the attention weight of different logging parameters on the depth sequence, and its calculation process is as follows:

[0104]

[0105] in, is a scaling factor used to prevent the dot product value from being too large, thereby affecting the operation of the softmax() function. Q and K are the query vector and key vector in the attention mechanism, respectively. The calculation process is as follows:

[0106] Q = x n W q n

[0107] K=x n W k n

[0108] Among them, W q n and W k n is the corresponding learnable weight matrix. After obtaining the weight value of each logging parameter at depth, the weight value of each logging parameter at depth is calculated using the following formula:

[0109] O=Z n ×V n

[0110]

[0111] Where O is x n The weighted value output of the parameter at depth; V n For x n a matrix of values on the parameters; is the corresponding learnable weight matrix.

[0112] After calculating the weighted values of each logging parameter at depth, considering that the factors affecting the formation pressure at different levels during the formation sedimentation process are different, the present invention introduces a weight parameter matrix W to assign corresponding weights to different logging parameters to obtain richer formation sedimentation characteristics. The process is as follows:

[0113] Output=O×W

[0114] This improved attention mechanism dynamically adjusts the weight of each logging parameter based on its impact on formation pressure. This allows the model to effectively learn the local and global characteristics of different logging parameters and ensures that the importance of different features is weighted differently when calculating pressure, ensuring that each logging parameter has a distinct impact on the final formation pressure prediction.

[0115] It should be noted that, in some embodiments, step S300 may include the following steps: inputting the source domain sample into the prediction model for processing to obtain a prediction result; constructing a loss function based on the prediction result and the source domain label corresponding to the source domain sample, and then adjusting the model parameters of the prediction model through convergence optimization feedback of the loss function until the loss function meets the preset conditions, thereby training the prediction model; wherein the expression of the loss function is:

[0116]

[0117] Where, L s represents the loss function; N s Indicates the number of source domain samples; represents the i-th source domain sample; Indicates the source domain label corresponding to the i-th source domain sample; Represents the prediction result corresponding to the i-th source domain sample; L represents the sub-loss function applied to the prediction result corresponding to the i-th source domain sample and the source domain label.

[0118] For example, in some specific implementations, in the specific migration process of the present invention, the source domain data O is first used. s ={(X s ,Y s )} for training, the goal of the model is to minimize the loss function on the source domain training data

[0119]

[0120] in, It represents the prediction (regression) results of the real-time calculation model of formation pressure on the source domain data. is the loss function (used to construct the loss value between the prediction result of the i-th source domain sample and the source domain label), θ sis the parameter of the source domain model, N s is the amount of source domain data. The training process is performed by minimizing To learn the parameters θ of the model s , which enables the model to perform effective formation pressure prediction on source domain data.

[0121] S400 , obtaining a target domain sample of a second exploration well, and fine-tuning parameters of the first model using transfer learning based on the target domain sample to obtain a second model.

[0122] The second model is used to predict the formation pressure of the second exploration well.

[0123] It should be noted that, in some embodiments, fine-tuning the parameters of the first model using transfer learning based on the target domain samples may include the following steps: obtaining a target domain label corresponding to the target domain sample; inputting the target domain sample into the first model for processing to obtain a predicted output; constructing a target loss function based on the predicted output corresponding to the target domain sample and the target domain label; fine-tuning the model parameters of the first model based on the target loss function in combination with a preset training strategy to obtain a prediction model; wherein the expression of the target loss function is:

[0124]

[0125] Where, L t represents the loss function; N t Indicates the number of target domain samples; represents the i-th target domain sample; represents the target domain label corresponding to the i-th target domain sample; Represents the predicted output corresponding to the i-th target domain sample; L represents the sub-loss function applied to the predicted output corresponding to the i-th target domain sample and the target domain label.

[0126] Among them, in some optional implementations, when the target domain data does not have a label, the target domain label corresponding to the target domain sample can be obtained by self-training to construct a predicted pseudo-label; specifically, the self-training step can be implemented as follows: predict pseudo-labels for the target domain data based on the first model trained on the source domain data (select high-confidence samples); add the pseudo-label data to the training set and retrain the model; iterate and optimize to gradually improve the quality of the pseudo-labels.

[0127] For example, in some specific implementations, in the target domain fine-tuning stage, the present invention only uses a portion of the target domain data D t ={(X t ,Y t )}, the goal of fine-tuning is to adjust the model trained in the source domain so that the model can better adapt to the target domain data. Assume that the model parameters after source domain training are θs , in the process of fine-tuning in the target domain, we hope to optimize the model parameters θ t , so that the loss of the model on the target domain data is minimized. The objective function in the fine-tuning process can be expressed as:

[0128]

[0129] in, represents the predicted output of the fine-tuned model on the target domain, is the loss function (used to construct the loss value between the prediction result of the i-th target domain sample and the target domain label), θ t is the model parameter after fine-tuning in the target domain, N t is the number of target domain samples.

[0130] In some embodiments, fine-tuning the model parameters of the first model based on the target loss function in combination with a preset training strategy may include the following steps: freezing the convolutional neural network in the first model; fine-tuning the model parameters of the first model based on the loss function in combination with a preset learning rate; wherein the learning rate is preset by the following relationship:

[0131]

[0132] Where θ t represents the model parameters of the first model after fine-tuning with target domain samples; θ s represents the model parameters of the first model; α represents the learning rate; Represents the target loss function for the model parameter θ t gradient.

[0133] For example, in some specific implementations, to ensure that the prior knowledge of the source domain model is not excessively lost, the following strategies are usually adopted during the fine-tuning process:

[0134] (1) Freezing some layers: During fine-tuning, some layers of the source domain model (multiple local spatial feature perception layers) are frozen because the features learned by this module are generally more universal. Only the stratigraphic sedimentary feature extraction model is fine-tuned to adapt to the sequential characteristics of the target domain data.

[0135] (2) Smaller learning rate: To avoid drastically modifying the parameters learned in the source domain, a smaller learning rate α is usually set for fine-tuning:

[0136]

[0137] in, is the target loss function for the parameter θ t gradient.

[0138] In order to explain the principle of the technical solution of the present invention in detail, the overall process of the present invention is described below in combination with some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and cannot be regarded as a limitation of the present invention.

[0139] First of all, it should be noted that in the actual oil and gas exploration process, in order to save costs and improve exploration efficiency, an exploratory well is often drilled. The sedimentary characteristics of the regional strata are inferred based on the physical and chemical properties of the formation measured by the exploratory well. However, in the field of pressure monitoring, the pressure value is affected by the combined effects of the effective stress of the sedimentary process and the pore fluid of the formation, resulting in different degrees of abnormal pressure in the same area. Therefore, the calculation accuracy of the calculation model on the lead well is poor. To address this problem, the traditional optimization method optimizes the model, but the results have little effect in terms of accuracy and the depth corresponding to the pressure value. Based on the above reasons, the present invention proposes a new model optimization method for adjacent well pressure calculation to improve the applicability of its model. First, to address the problem of poor pressure calculation accuracy of the model in adjacent wells in the region, the present invention combines the mechanism and characteristics of the formation abnormal pressure, proposes a model optimization method that takes into account the formation abnormality, and innovatively proposes the concept of the formation abnormal pressure factor as the input of the model. At the same time, in order to prevent the characteristic input from causing gradient explosion or gradient disappearance in the model, the present invention analyzes the regional strata and standardizes its value using the lg function; secondly, to further improve the calculation accuracy of the model on adjacent wells, the present invention proposes a migration method for the real-time formation pressure calculation model based on the sedimentary characteristics of the strata and the idea of the equivalent depth method; finally, to demonstrate the effectiveness of the method proposed in this paper, the present invention conducts example verification of the proposed method based on actual drilling data.

[0140] Specifically, in terms of pressure prediction (calculation), the causes of abnormal pressure are divided into two categories based on geophysical properties: undercompaction and fluid expansion. Undercompaction mainly occurs in shallower formations, which is usually manifested as insufficient compaction of the formation and higher pore water pressure. This type of pressure anomaly usually has strong local variability and a faster rate of change, especially when sediments accumulate rapidly or the formation is covered in a short period of time. This means that undercompaction often manifests as more localized abnormal pressure changes; abnormal pressure caused by fluid expansion usually occurs in deeper oil and gas reservoirs or water layers, and this pressure change is relatively smooth and has greater continuity. With the expansion of deep fluids or the accumulation of gas, the abnormal pressure caused by fluid expansion changes relatively slowly in the vertical direction, manifesting as pressure changes over a long period of time and a large range. Therefore, the present invention only studies the impact of these two types of abnormal pressure cause mechanisms on the accuracy of the calculation model.

[0141] At present, the main method for determining the cause of abnormal pressure is to draw an acoustic wave-density intersection diagram and conduct qualitative analysis in combination with the compaction trend line. This method of determination is relatively accurate, but it cannot be input as a characteristic parameter into the pressure calculation model to provide the calculation model with relevant characteristics of the floating direction and size of the abnormal pressure in the formation. Therefore, the present invention proposes the concept of abnormal pressure cause factor (G) for the first time, aiming to provide quantifiable abnormal pressure characteristics for the pressure calculation model. The quantification principle is: use a sliding window to calculate the slope of the acoustic wave-density intersection diagram in this section at this depth, such as Figure 3 As shown in the red box, the calculation formula is as follows:

[0142]

[0143] Among them, V P(i) Indicates the sound wave value at depth point i, R HOB(i) represents the density value at depth point i, W represents the width of the sliding window, that is, the abnormal pressure factor G calculated within the depth range of W i .

[0144] In some specific application scenarios, for undercompaction causes, a very small window (5–10m) can more sensitively capture local pressure changes. However, this sensitivity may lead to the amplification of high-frequency noise, making the monitoring results unstable and difficult to accurately reflect the true trend of pressure changes. Especially in undercompacted areas, where local pressure fluctuations are large, a very small window may lead to misjudgment and fail to effectively identify the overall trend of abnormal pressure changes. A window depth that is too large (20m) may prevent the model from responding to the details of local pressure changes in the formation in a timely manner. Especially in areas where the causes of abnormal pressure are more complex or pressure changes are rapid, a large window will lead to "sluggish response" to short-term pressure fluctuations. Although for fluid expansion causes, an excessively large window may help smooth data, during dynamic monitoring, the delayed response may lead to insufficient timely detection of potential anomalies.

[0145] Based on the above analysis, the present invention makes a compromise when designing the sliding window size and selects a moderate sliding window (for example, 17m, which can be adjusted according to actual application needs). The window size within this range can balance the capture of local changes and overall trends. In the undercompacted genesis area, the 17-meter window is large enough to eliminate local noise and smooth the fluctuations in the data, while not being too smooth to lose the response to subtle changes; for fluid expansion causes, the 17-meter window depth can also effectively capture relatively stable pressure changes, avoiding the inability of a window that is too small to identify pressure anomalies in a larger range. In addition, to facilitate program writing, the present invention converts the 17m window size into 100 sampling points in combination with the sampling frequency of the LWD data (6 times / meter), that is, the sliding window size is 100 sampling points, and the recognition results are as follows. Figure 6 As shown. Figure 6 It can be seen from the figure that the abnormal formation pressure factor proposed in the present invention has a better identification ability in the abnormal formation pressure section.

[0146] In some specific implementations, when calculating the slope of the acoustic wave-density crossplot within the window, it is inevitable to encounter an infinite slope value, especially in abnormal pressure formations caused by fluid expansion. Therefore, in order to complete the calculation process, the present invention sets the slope value in this case to 10 10 -1. When these features containing extremely large values are used as model input, direct normalization may lose the original meaning and value of the feature. Especially when monitoring the causes of abnormal pressure, extremely large values usually represent special abnormal states. After normalization, these extreme values will be compressed, resulting in the model being unable to correctly understand their importance in monitoring. Conversely, excessively large values may interfere with model training and affect the learning of other features. Therefore, using a customized normalization method (such as limiting the upper limit of the maximum value) is a more appropriate processing strategy, which can ensure that the model can effectively capture the key role of the feature while avoiding distortion of outliers during data processing.

[0147] Logarithmic normalization is an effective data preprocessing method suitable for handling features with extreme values, skewed distributions, or exponential growth. By applying a logarithmic transformation to each data point, it can compress the range of large values, reduce the interference of extreme values on model training, and smooth the data distribution, making the data more consistent with the model training requirements. The principle is as follows:

[0148] C=lg(G+1)

[0149] Where G is the original abnormal formation pressure factor, and C is the normalized abnormal formation pressure factor. The addition of 1 (G+1) is to avoid infinity or mathematical errors when the logarithmic function processes zero or negative values. To ensure that the model can effectively capture the key characteristics of the parameters while avoiding distortion of outliers during data processing, the present invention uses the lg function as the function for normalizing the logarithmic function.

[0150] Based on the aforementioned improvement of the model's calculation accuracy by increasing the characteristic parameters of the causal mechanism of abnormal formation pressure, the present invention addresses the problem that the distribution of the vertical effective stress of the formation skeleton in the depth direction is different, which leads to differences in depth and accuracy between the model prediction value and the actual value. Drawing on the idea of vertical effective stress matching in the equivalent depth method, the present invention uses the transfer learning method to "match" the vertical effective stress of adjacent wells in the region in depth, so as to improve the applicability of the model to adjacent wells in the region.

[0151] like Figure 7As shown in FIG, the goal of the present invention is to train a formation pressure prediction model based on the labeled data of the source domain (exploratory well) and migrate the model to the target domain (neighboring well) to predict the formation pressure. Due to the differences in feature space and data distribution between the source domain and the target domain, directly applying the source domain model to the target domain may lead to a decrease in prediction performance. Therefore, a transfer learning strategy is adopted to improve the performance of the target domain model through fine-tuning. First, the source domain and the target domain are defined. The data in the source domain are denoted as O s ={(X s ,Y s )}, where X s is the feature of the source domain sample, Y s is the label of the source domain sample, the number of samples is N s The data of the target domain is recorded as O t ={X t}, where X t It is the characteristic of the target domain sample and does not contain label data. The number of samples is N t Assume that the source domain and target domain share the same feature space, namely X s =X t And the label space Y s =Y t Specifically, the source domain label Y s and the target domain label Y t They all represent the calculated value or abnormal state of formation pressure. The task is to predict the relationship between depth and formation pressure, or to identify the abnormal state of formation pressure.

[0152] In the specific migration process of the present invention, first use the source domain data O s ={(X s ,Y s )} for training, the goal of the model is to minimize the loss function on the source domain training data

[0153]

[0154] in, It represents the prediction (regression) results of the real-time calculation model of formation pressure on the source domain data. is the loss function, θ s is the parameter of the source domain model, N s is the amount of source domain data. The training process is performed by minimizing To learn the parameters θ of the model s , so that the model can effectively predict the formation pressure on the source domain data. In the target domain fine-tuning stage, the present invention only uses a part of the target domain data D t ={(X t ,Y t)}, the goal of fine-tuning is to adjust the model trained in the source domain so that the model can better adapt to the target domain data. Assume that the model parameters after source domain training are θ s , in the process of fine-tuning in the target domain, we hope to optimize the model parameters θ t , so that the loss of the model on the target domain data is minimized. The objective function in the fine-tuning process can be expressed as:

[0155]

[0156] in, represents the predicted output of the fine-tuned model on the target domain, is the loss function, θ t is the model parameter after fine-tuning in the target domain, N t is the number of samples in the target domain. In addition, in order to ensure that the prior knowledge of the source domain model is not excessively lost, the following strategies are usually adopted during fine-tuning:

[0157] (1) Freezing some layers: During fine-tuning, some layers of the source domain model (multiple local spatial feature perception layers) are frozen because the features learned by this module are generally more universal. Only the stratigraphic sedimentary feature extraction model is fine-tuned to adapt to the sequential characteristics of the target domain data.

[0158] (2) Smaller learning rate: To avoid drastically modifying the parameters learned in the source domain, a smaller learning rate α is usually set for fine-tuning:

[0159]

[0160] in, is the target loss function for the parameter θ t gradient.

[0161] During fine-tuning, fine-tuning of the variable sequence feature quantification unit based on a recurrent neural network and the formation sedimentary characteristics extraction unit based on an improved attention mechanism is particularly important. The former is used to capture the temporal dependencies in formation pressure calculations. By fine-tuning the network parameters in this unit (the weights of the forget gate, input gate, and output gate), the unit can be better adapted to the temporal characteristics of the target domain data. The latter helps the model assign different attention weights to different depth segments. By fine-tuning these weights, the model can focus more on the key features of specific areas in the target well (such as high-pressure or low-pressure areas), thereby improving prediction accuracy.

[0162] Based on the above theoretical analysis, the present invention uses three wells, A, B, and C, as examples. Specifically, Well A is used as the source domain dataset, and Wells B and C are used as target domain datasets. Considering the requirements for safety, real-time performance, and computational efficiency in engineering applications, the sample data in the target domain focuses on the formation compaction section below the mudline of Wells B and C, and this section of data is used to fine-tune the pre-trained model. The experimental results show that the prediction accuracy of the optimized model on adjacent wells is significantly improved. The specific performance indicators are shown in Table 1. The prediction results are compared. Figure 8 shown.

[0163] Table 1

[0164] Well A Well B Well C The accuracy of abnormal pressure causes is not considered 0.0493 1.1726 0.1537 Accuracy in considering causes of abnormal pressure 0.1028 0.9674 0.1456 Accuracy of transfer learning strategies based on equivalent depth 0.0472 0.1868 0.0804

[0165] Based on the test result distribution diagram of Well C and Well B, the curve change diagram and the accuracy comparison data in Table 1, the transfer learning strategy based on equivalent depth proposed in the present invention shows significant superiority in formation pressure calculation. First of all, in terms of the degree of consistency between the calculated value and the true value, the performance of the model optimized by transfer learning is particularly outstanding. In the test results of Well C, the calculated value curve of the transfer learning model is highly consistent with the true value in the 1500-4000 meter well section, especially in the abnormally high-pressure layer (2450 meter well section), the calculation error is only 0.05MPa, which is significantly lower than the 0.15MPa of the model considering the cause of abnormal pressure. Figure 8 As shown in Table 1, the transfer learning model's calculated values at the high-pressure gradient abrupt change point show a smooth transition, without the "step-like" deviations seen in previous methods. This demonstrates its ability to adapt to complex pressure mechanisms. This indicates that transfer learning, by learning from the common features of the source domain (Well A), effectively captures the local geological anomalies of Well C. In the test results for Well B, the pressure curve calculated by the transfer learning model is highly consistent with the true values in the 2000-4000 m well section, particularly in the compacted section below the mudline (2500-3000 m). Methods that fail to consider the causes of the abnormal pressures result in systematically higher predicted values (RMSE = 1.1726). However, the transfer learning strategy, by fine-tuning the target domain data, reduces the RMSE to 0.1868. The chart shows that the transfer learning model's calculated values at the high-pressure gradient abrupt change point (2800 m well section) show a smooth transition.

[0166] Secondly, the transfer learning strategy shows strong adaptability to complex geological conditions. Compared with the model optimization method that only considers the causes of abnormal formation pressure, the pressure calculation values of the transfer learning strategy model optimization method proposed in this invention are almost consistent with the true values in the abnormal high-pressure layers of Wells B and C (at the 2800-meter section of Well B and the 2450-meter section of Well C). This shows that the strategy can effectively capture the local geological anomaly characteristics of the target domain by learning the universal features of the source domain (Well A). In addition, in the three wells, the pressure gradient change trend of the entire well section of the transfer learning model is completely synchronized with the true value. For example, at the inflection point of the transition from normal pressure to overpressure at 2100 meters in Well B, the prediction curve of the transfer learning model transitions smoothly, without the "step-by-step" deviation of the traditional method. This shows that transfer learning can adapt to complex geological conditions and break through the strong dependence of the traditional equivalent depth theory on "similar geological background".

[0167] In terms of engineering practicality and generalization ability, the transfer learning strategy also shows significant advantages. By focusing on the compaction section data below the mudline for fine-tuning, the calculation time of the transfer learning model in Wells B and C was reduced by 40% compared with the traditional method, while maintaining high accuracy (RMSE < 0.2MPa). The smoothness of the calculated value curve in the chart further verifies the stability of the model in actual engineering. Although the geological structure of Well B is quite different from that of Well A in the source domain, the model optimized by the transfer learning strategy still achieved an 81% reduction in RMSE (from 1.1726 to 0.1868) through parameter migration and feature alignment. This shows that this method can break through the traditional equivalent depth theory's strong dependence on "similar geological background" and has a wider applicability.

[0168] In summary, through Figure 9 Comparing the prediction curve with the accuracy data in Table 1, the proposed transfer learning strategy demonstrates significant advantages in computational accuracy, adaptability to complex geology, engineering practicality, and generalization. These results fully demonstrate the advanced nature of the transfer learning strategy in formation pressure calculation and its engineering application value.

[0169] The present invention addresses the problems of insufficient accuracy and poor generalization performance of the real-time calculation model for formation pressure in adjacent well applications, and innovatively proposes a method and strategy to significantly improve the accuracy, generalization, and engineering applicability of the model. First, the causes of abnormal pressure are introduced as input parameters for the model for different formations, breaking through the limitations of traditional methods that rely on well logging data and effective stress theory. By quantifying the correlation between the causes of abnormal pressure (fluid expansion, undercompaction, etc.) and formation pressure, the model's adaptability to complex geological conditions is enhanced. Experimental results show that in Wells B and C, after the introduction of the abnormal pressure cause factor, the calculation error is significantly reduced (RMSE of Well B is reduced from 1.1726 to 0.9674, and that of Well C is reduced from 0.1537 to 0.1456). In particular, in abnormally high-pressure layers (such as 2800 meters in Well B and 2450 meters in Well C), the deviation between the calculated value and the true value is reduced by more than 60%.

[0170] Secondly, the present invention proposes an optimization strategy for model migration based on transfer learning of equivalent depth, which solves the problems of insufficient accuracy and poor adaptability of traditional equivalent depth methods in cross-well migration. The strategy is divided into two stages: pre-training and fine-tuning: the pre-training stage uses the source domain (well A) data to learn the general formation pressure characteristics; the fine-tuning stage focuses on the compaction section data below the mudline in the target domain (well B and well C), and adapts the target domain characteristics through parameter optimization. Experimental verification shows that the transfer learning strategy performs excellently in the target domain: the RMSE of well B is reduced from 1.1726 to 0.1868 (accuracy increased by 84%), and the RMSE of well C is reduced from 0.1537 to 0.0804 (accuracy increased by 48%). The calculated value curve shows that the calculated value of the model transitions smoothly at the high-pressure gradient mutation point (such as the inflection point at 2100 meters in well B), avoiding the "step-by-step" deviation of the traditional method.

[0171] The combined advantages of this method-based strategy for migrating adjacent wells are reflected in three key aspects: First, prediction accuracy is significantly improved, with the error in key horizons in the target domain less than one-fifth of the pre-optimization method. Second, adaptability to complex geology is enhanced, with the model able to capture local features such as abnormal pressures and lithologic interfaces, breaking through the traditional theory's strong reliance on "similar geological settings." Third, engineering practicality is improved. By focusing on compaction section data for fine-tuning, computational time is reduced by 40%, while maintaining high accuracy (RMSE < 0.2 MPa), meeting real-time requirements. Furthermore, the model demonstrates outstanding cross-well generalization capabilities. Despite the differences in fault development between Well B and Well A in the source domain, transfer learning still achieves an 81% reduction in RMSE through feature alignment.

[0172] This invention contributes both theoretically and practically. Theoretically, by integrating the causes of abnormal pressure with a transfer learning strategy, it expands the modeling framework for real-time formation pressure calculation. Practically, it demonstrates the effectiveness and engineering applicability of this method under complex geological conditions, providing a high-precision tool for oil and gas exploration and development. Future research could further explore multi-source transfer learning strategies, incorporate physical constraints to enhance model interpretability, and extend this to areas such as geothermal resource assessment. This invention provides an important technical reference for the research and practice of formation pressure prediction.

[0173] On the other hand, an embodiment of the present invention provides a model optimization device for formation pressure prediction, which may include:

[0174] The first module is used to obtain the logging while drilling data of the first exploration well and the abnormal pressure factors in each depth range, and then organize them to obtain source domain samples;

[0175] The second module is configured to use the measured pressure value or abnormal state corresponding to the source domain sample of the first exploration well as the source domain label;

[0176] The third module is configured to perform a first training on a preset prediction model using source domain samples and source domain labels to obtain a first model;

[0177] Among them, the prediction models include convolutional neural networks, recurrent neural networks, and fully connected networks;

[0178] The fourth module is used to obtain target domain samples of the second exploration well, and fine-tune the parameters of the first model using transfer learning based on the target domain samples to obtain a second model;

[0179] The second model is used to predict the formation pressure of the second exploration well.

[0180] In some embodiments, the apparatus may further include a fifth module configured to perform the following operations:

[0181] Normalize abnormal stress factors.

[0182] The contents of the method embodiments of the present invention are all applicable to the device embodiments. The functions specifically implemented by the device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0183] In another aspect, an embodiment of the present invention further provides an electronic device comprising a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned model optimization method for formation pressure prediction. The electronic device can be any smart terminal, including a tablet computer and an in-vehicle computer.

[0184] It can be understood that the contents of the above method embodiments are applicable to the present device embodiments, the functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0185] like Figure 10 As shown, Figure 10 The hardware structure of an electronic device 1000 according to another embodiment is shown. The electronic device 1000 includes:

[0186] The processor 1001 may be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute relevant programs to implement the technical solutions provided by the embodiments of the present invention.

[0187] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called by the processor 1001 to execute the network node population optimization method of the embodiment of the present invention.

[0188] Input / output interface 1003, used to implement information input and output;

[0189] Communication interface 1004, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0190] Bus 1005 , which transmits information between various components of the device (e.g., processor 1001 , memory 1002 , input / output interface 1003 , and communication interface 1004 );

[0191] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0192] The electronic device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one location or distributed across multiple network units. Some or all of these modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0193] The contents of the method embodiments of the present invention are all applicable to the electronic device embodiments. The functions specifically implemented by the electronic device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0194] Another aspect of an embodiment of the present invention further provides a computer-readable storage medium, wherein the storage medium stores a program, and the program is executed by a processor to implement the above method.

[0195] It should be noted that the computer-readable medium shown in the embodiments of the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present invention, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0196] The contents of the method embodiments of the present invention are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0197] The present invention also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform the above method.

[0198] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0199] It should be noted that although several modules of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0200] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD to ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present invention.

[0201] In some optional embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the present invention is provided in an exemplary manner for the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operation and logic flow presented by the present invention. Optional embodiments are contemplated in which the order of the various operations is changed and the sub-operations described as a part of a larger operation are performed independently.

[0202] Furthermore, while the present invention has been described in the context of functional modules, it should be understood that, unless otherwise indicated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It will also be understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present invention. More specifically, given the properties, functions, and internal relationships of the various functional modules in the devices disclosed in the present invention, the actual implementation of the module will be understood within the ordinary skill of an engineer. Therefore, a person skilled in the art will be able to implement the present invention as set forth in the claims using ordinary skill without undue experimentation. It will also be understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present invention, which is determined by the full scope of the appended claims and their equivalents.

[0203] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0204] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution apparatus, device, or apparatus (e.g., a computer-based apparatus, a device including a processor, or other apparatus that can fetch instructions from and execute instructions on an instruction execution apparatus, device, or apparatus). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution apparatus, device, or apparatus.

[0205] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0206] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0207] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0208] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

[0209] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. A model optimization method for formation pressure prediction, characterized in that: The following steps are involved: Obtain the logging while drilling data of the first exploration well and the abnormal pressure factors in each depth range, and then organize them to obtain source domain samples; Using the measured pressure value or abnormal state corresponding to the source domain sample of the first exploration well as a source domain label; Performing a first training on a preset prediction model using the source domain samples and the source domain labels to obtain a first model; Wherein, the prediction model includes convolutional neural network, recurrent neural network and fully connected network; Obtaining a target domain sample of a second exploration well, and fine-tuning parameters of the first model using transfer learning based on the target domain sample to obtain a second model; The second model is used to predict the formation pressure of the second exploration well.

2. The model optimization method for formation pressure prediction according to claim 1, characterized in that: The method of obtaining the abnormal pressure factors of each depth range of the first exploration well includes the following steps: Obtaining acoustic wave values and density values at various depth points of the first exploration well; Based on the acoustic wave value and the density value at each depth point, using a sliding window to calculate the slope of an interaction graph of acoustic wave and density at each depth range of the first exploration well; determining the abnormal pressure factor of each depth range of the first exploration well according to the slope of the interaction graph; The expression of the abnormal pressure factor is: Where V P(i) Represents the acoustic wave value at the i-th depth point; R HOB(i) represents the density value of the i-th depth point; W represents the width of the sliding window; G i It represents the abnormal pressure factor in the W depth range after the i-th depth point.

3. The model optimization method for formation pressure prediction according to claim 1, characterized in that: Before the step of arranging and obtaining source domain samples, the method further includes the following steps: The abnormal pressure factors are normalized.

4. The model optimization method for formation pressure prediction according to claim 3, characterized in that: The method of performing standardization on the abnormal pressure factor comprises the following steps: Using a logarithmic function to perform normalization on the abnormal pressure factor; The expression for the standardization process is: C=lg(G+1) Where G represents the original abnormal pressure factor; C represents the abnormal pressure factor after normalization; and lg is the applied logarithmic function.

5. The model optimization method for formation pressure prediction according to claim 1, characterized in that: The first training of a preset prediction model using the source domain samples and the source domain labels to obtain a first model includes the following steps: Inputting the source domain sample into the prediction model for processing to obtain a prediction result; Constructing a loss function based on the prediction results corresponding to the source domain samples and the source domain labels, and then adjusting the model parameters of the prediction model through convergence optimization feedback of the loss function until the loss function meets preset conditions, thereby training the prediction model; Among them, the expression of the loss function is: Where, L s represents the loss function; N s Indicates the number of source domain samples; represents the i-th source domain sample; Indicates the source domain label corresponding to the i-th source domain sample; Represents the prediction result corresponding to the i-th source domain sample; L represents the sub-loss function applied to the prediction result corresponding to the i-th source domain sample and the source domain label.

6. The model optimization method for formation pressure prediction according to claim 1, characterized in that: Fine-tuning the parameters of the first model using transfer learning based on the target domain sample includes the following steps: Obtaining a target domain label corresponding to the target domain sample; Inputting the target domain sample into the first model for processing to obtain a prediction output; constructing a target loss function based on the predicted output corresponding to the target domain sample and the target domain label, and fine-tuning model parameters of the first model based on the target loss function in combination with a preset training strategy to obtain the prediction model; Among them, the expression of the objective loss function is: Where, L t represents the loss function; N t Indicates the number of target domain samples; represents the i-th target domain sample; represents the target domain label corresponding to the i-th target domain sample; Represents the predicted output corresponding to the i-th target domain sample; L represents the sub-loss function applied to the predicted output corresponding to the i-th target domain sample and the target domain label.

7. The model optimization method for formation pressure prediction according to claim 6, characterized in that: Fine-tuning the model parameters of the first model based on the target loss function in combination with a preset training strategy includes the following steps: Freezing the convolutional neural network in the first model; Fine-tuning model parameters of the first model based on the loss function and a preset learning rate; The learning rate is preset by the following relationship: Where θ t represents the model parameters of the first model after fine-tuning with target domain samples; θ s represents the model parameters of the first model; α represents the learning rate; Represents the target loss function for the model parameter θ t gradient.

8. A model optimization device for formation pressure prediction, characterized in that: include: The first module is used to obtain the logging while drilling data of the first exploration well and the abnormal pressure factors in each depth range, and then organize them to obtain source domain samples; The second module is configured to use the measured pressure value or abnormal state corresponding to the source domain sample of the first exploration well as a source domain label; A third module is configured to perform a first training on a preset prediction model using the source domain samples and the source domain labels to obtain a first model; Wherein, the prediction model includes convolutional neural network, recurrent neural network and fully connected network; A fourth module is configured to obtain target domain samples of a second exploration well, and fine-tune parameters of the first model using transfer learning based on the target domain samples to obtain a second model; The second model is used to predict the formation pressure of the second exploration well.

9. An electronic device, characterized in that: including a processor and a memory; The memory is used to store programs; The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer storage medium storing a program executable by a processor, characterized in that: The program executable by the processor is used to implement the method according to any one of claims 1 to 7 when executed by the processor.

Images