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

Abstract

The invention discloses a formation pressure prediction method and device, electronic equipment and a storage medium. The method comprises the steps that logging-while-drilling data of a target area are acquired; performing variable sequence characteristic quantification on the continuous time sequence logging-while-drilling data by using a recurrent neural network to obtain sequence characteristics; on the basis of the sequential features, weighted values of all logging parameters in the logging-while-drilling data are constructed through an attention mechanism, and then stratum deposition features are constructed through a preset weight parameter matrix; inputting the stratum deposition characteristics into a prediction model to predict and obtain a stratum pressure prediction value of the target area; wherein the prediction model is obtained based on a full-connection network through training of formation feature representation marked with actual formation pressure. The method can efficiently and accurately realize formation pressure prediction, and 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 formation pressure prediction method, device, electronic equipment and storage medium. Background Art

[0002] Existing technologies typically use fixed weights or simple weighting methods to fuse multi-parameter logging data. This lacks the ability to dynamically focus on key geological parameters, making it difficult to adaptively reflect the differential contributions of various parameters under different formation conditions. Furthermore, the construction of formation sedimentary characteristics often relies on expert prior knowledge or statistical models, which are subject to strong subjectivity and poor generalization, and are particularly prone to introducing human bias in areas with complex geological structures. Traditional prediction models (such as linear regression and empirical formulas) have limited ability to fit nonlinear formation pressure relationships and are unable to effectively associate high-dimensional features with target outputs, resulting in prediction accuracy and stability that are difficult to meet real-time drilling requirements. Summary of the Invention

[0003] The present invention aims to at least partially address the limitations of related technologies. To this end, the present invention provides a formation pressure prediction method, device, electronic device, and storage medium that can accurately predict formation pressure.

[0004] In one aspect, an embodiment of the present invention provides a formation pressure prediction method, comprising the following steps:

[0005] Acquire LWD data in target areas;

[0006] The recurrent neural network is used to quantify the variable sequence characteristics of continuous time series logging while drilling data to obtain the sequence characteristics;

[0007] Based on the sequential characteristics, the attention mechanism is used to construct the weighted values of each logging parameter in the LWD data, and then the stratigraphic sedimentary characteristics are obtained through the preset weight parameter matrix.

[0008] Inputting the formation sedimentary characteristics into the prediction model to predict the formation pressure of the target area;

[0009] Among them, the prediction model is trained based on a fully connected network through formation feature representation marked with actual formation pressure.

[0010] Optionally, after the step of acquiring the logging while drilling data of the target area, the method further comprises the following steps:

[0011] Perform data preprocessing on LWD data;

[0012] Among them, data preprocessing includes outlier removal, missing value filling and data normalization.

[0013] Optionally, before the step of quantifying variable sequence features of continuous time series logging while drilling data using a recurrent neural network, the method further includes the following steps:

[0014] Use the preset convolutional neural network to perceive multiple local spatial features of logging while drilling data;

[0015] Among them, the recurrent neural network quantifies the variable sequential features based on the results corresponding to multiple local spatial feature perceptions.

[0016] Optionally, using a recurrent neural network to quantify variable sequence characteristics of continuous time series logging while drilling data to obtain sequence characteristics includes the following steps:

[0017] The long short-term memory network is used to quantify the variable sequence characteristics of continuous time series logging while drilling data through memory units and gating mechanisms to obtain sequential features.

[0018] Among them, the gating mechanism includes input gate, forget gate and output gate.

[0019] Optionally, a long short-term memory network is used to quantify variable sequence characteristics of continuous time series logging while drilling data through memory units and a gating mechanism to obtain sequence characteristics, including the following steps:

[0020] Obtain forgetting information using a forget gate based on the hidden state output at the first moment and the input data at the second moment; wherein the second moment represents the moment after the first moment, and the input data is determined based on the logging while drilling data;

[0021] According to the hidden state output at the first moment and the input data at the second moment, the input gate is used to process the added information;

[0022] According to the hidden state output at the first moment and the input data at the second moment, the hidden state is obtained by processing with the output gate;

[0023] Based on the hidden state output at the first moment and the input data at the second moment, the memory unit is used to process the new information, and then the state of the memory unit at the second moment is updated based on the state of the memory unit at the first moment combined with the forgotten information, added information and new information;

[0024] The hidden state output at the second moment is obtained according to the state of the memory unit at the second moment combined with the hidden state;

[0025] The second moment is taken as the first moment, and the process returns to the step of using the forget gate to process the forgotten information based on the hidden state output of the first moment and the input data of the second moment, until all the LWD data in the continuous time series are processed and the sequential features are obtained.

[0026] Optionally, based on the sequential features, a weighted value of each logging parameter in the logging while drilling data is constructed through an attention mechanism, including the following steps:

[0027] Based on the sequential features corresponding to the logging parameters and the preset learnable weight matrix, the attention mechanism is used to construct the query vector, key vector, and value vector corresponding to the logging parameters;

[0028] Based on the query vector and key vector combined with the preset scaling factor, the attention weight corresponding to the logging parameter is constructed through a normalized exponential function;

[0029] The weighted value corresponding to the logging parameter is obtained according to the product of the attention weight and the value vector.

[0030] Optionally, the method further comprises the following steps:

[0031] The formation feature representation is input into the fully connected network for processing to obtain the formation pressure training value; wherein the data structure of the formation feature representation is the same as the formation sedimentary feature;

[0032] A loss function is constructed based on the formation pressure training value corresponding to the formation characteristic representation and the actual formation pressure. Then, the network parameters of the fully connected network are adjusted through feedback using the error result of the loss function to obtain a prediction model.

[0033] In another aspect, an embodiment of the present invention provides a formation pressure prediction device, comprising:

[0034] The first module is used to obtain the logging while drilling data of the target area;

[0035] The second module is used to quantify the variable sequence characteristics of continuous time series logging while drilling data using a recurrent neural network to obtain sequence characteristics;

[0036] The third module is used to construct the weighted values of each logging parameter in the LWD data based on the sequential characteristics through the attention mechanism, and then obtain the stratigraphic sedimentary characteristics through the preset weight parameter matrix;

[0037] The fourth module is used to input the formation sedimentary characteristics into the prediction model to predict the formation pressure prediction value of the target area;

[0038] Among them, the prediction model is trained based on a fully connected network through formation feature representation marked with actual formation pressure.

[0039] Optionally, the apparatus further includes a fifth module configured to perform the following operations:

[0040] Perform data preprocessing on LWD data;

[0041] Among them, data preprocessing includes outlier removal, missing value filling and data normalization.

[0042] Optionally, the apparatus further includes a sixth module configured to perform the following operations:

[0043] Use the preset convolutional neural network to perceive multiple local spatial features of logging while drilling data;

[0044] Among them, the recurrent neural network quantifies the variable sequential features based on the results corresponding to multiple local spatial feature perceptions.

[0045] Optionally, the apparatus further includes a seventh module configured to perform the following operations:

[0046] The formation feature representation is input into the fully connected network for processing to obtain the formation pressure training value; wherein the data structure of the formation feature representation is the same as the formation sedimentary feature;

[0047] A loss function is constructed based on the formation pressure training value corresponding to the formation characteristic representation and the actual formation pressure. Then, the network parameters of the fully connected network are adjusted through feedback using the error result of the loss function to obtain a prediction model.

[0048] 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 formation pressure prediction method.

[0049] On the other hand, an embodiment of the present invention provides a computer storage medium storing a program executable by a processor. The program executable by the processor is used to implement the above-mentioned formation pressure prediction method when executed by the processor.

[0050] The embodiment of the present invention obtains LWD data of a target area; uses a recurrent neural network to quantify the variable sequential characteristics of the continuous time series LWD data to obtain sequential characteristics; based on the sequential characteristics, constructs weighted values of each logging parameter in the LWD data through an attention mechanism, and then obtains formation sedimentary characteristics through a preset weight parameter matrix; inputs the formation sedimentary characteristics into a prediction model to predict the formation pressure of the target area; wherein the prediction model is trained based on a fully connected network through formation feature representations marked with actual formation pressure. The present invention uses a recurrent neural network to dynamically model the continuous time series characteristics of logging-while-drilling data, accurately capturing the longitudinal evolution of the formation sedimentation process and enhancing the physical expression of sequential features. Furthermore, the present invention adaptively constructs weighted values for each logging parameter through an attention mechanism, achieving focus on key geological parameters and noise suppression, significantly improving the pertinence and interpretability of feature fusion. Sedimentary features are then generated based on a preset weight parameter matrix, eliminating reliance on manual experience. Complex geological structures are objectively characterized through a data-driven approach, reducing the impact of subjective bias on prediction results. Finally, the present invention constructs a prediction model using a fully connected network and performs end-to-end training in conjunction with annotated actual formation pressure data. This model is capable of nonlinearly mapping the complex relationship between high-dimensional features and formation pressure, improving prediction accuracy and generalization while ensuring real-time performance. The present invention can efficiently and accurately predict formation pressure. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] 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.

[0052] Figure 1 This is a schematic diagram of an implementation environment for a formation pressure prediction method provided by an embodiment of the present invention;

[0053] Figure 2 1 is a flow chart of a formation pressure prediction method provided by an embodiment of the present invention;

[0054] Figure 3 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;

[0055] Figure 4 A schematic diagram illustrating an example of the overall framework principle for formation pressure prediction provided by an embodiment of the present invention;

[0056] Figure 5 A schematic diagram of a variable sequence feature quantization unit structure based on a recurrent neural network provided in an embodiment of the present invention;

[0057] Figure 6A schematic diagram of the thermal correlation between relevant logging parameters and between logging parameters and formation pressure provided by an embodiment of the present invention;

[0058] Figure 7 A schematic diagram of an example parameter data distribution box provided in an embodiment of the present invention;

[0059] Figure 8 A schematic diagram of comparative experimental results of goodness of fit of various algorithm models provided in an embodiment of the present invention;

[0060] Figure 9 A schematic diagram of a curve showing changes in model pressure calculated values and actual values for multiple well examples provided in an embodiment of the present invention;

[0061] Figure 10 A schematic diagram of the distribution box lines of the running time of each model provided in an embodiment of the present invention;

[0062] Figure 11 A schematic structural diagram of a formation pressure prediction device provided by an embodiment of the present invention;

[0063] Figure 12 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] 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.

[0065] 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.

[0066] 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.

[0067] It is understood that the formation pressure prediction method 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.

[0068] 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:

[0069] 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.

[0070] 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.

[0071] 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.

[0072] 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.

[0073] Based on the example Figure 1In the implementation environment shown, an embodiment of the present invention provides a formation pressure prediction method. The following is an example of the formation pressure prediction method being applied in the server 101. It can be understood that the formation pressure prediction method can also be applied in the terminal 102.

[0074] Reference Figure 2 , Figure 2 The flowchart of the formation pressure prediction method applied to the server provided in the embodiment of the present invention, the execution subject of the formation pressure prediction method can be any of the aforementioned computer devices (including servers or terminals). Figure 2 , the method comprises the following steps:

[0075] S100, acquiring logging while drilling data of a target area;

[0076] For example, in some specific implementations, logging while drilling (LWD) is a necessary step in oil and gas field exploration and development, requiring post-drilling logging to understand the formation's oil and gas content. Specifically, a logging instrument can be placed on the drill bit, giving it "eyes" and allowing it to acquire various formation data as it drills. This is LWD. This not only allows logging for wells of any condition, particularly horizontal wells, but also utilizes the measured drilling and formation parameters to promptly adjust the drill bit trajectory to align drilling with the target formation. Because the formation parameters obtained by LWD are those immediately after drilling, they most closely resemble the formation's original state and are therefore more useful for evaluating oil and gas content in complex formations than conventional wireline logging. The LWD instrument, placed within the drill collar, measures conventional logging parameters such as resistivity, sound velocity, neutron porosity, and density, as well as certain imaging logs. It also measures drilling parameters such as weight on bit, torque, rotational speed, annular pressure, temperature, and chemical composition.

[0077] In some embodiments, the method may further include the following steps: performing data preprocessing on the LWD data; wherein the data preprocessing includes outlier removal, missing value filling and data normalization.

[0078] For example, in some specific implementations, due to the dual influence of the drilling process and the marine-geological environment, the monitoring data of logging-while-drilling sensors often contains a large number of outliers and missing values, which can affect the training effect and computational efficiency of the model. Although the model design takes into account the issues of outlier and missing value processing, to ensure model accuracy during model training and testing, the present invention can perform outlier removal, missing value filling, and data normalization on the logging-while-drilling data.

[0079] In some embodiments, the method may further include the following steps: using a preset convolutional neural network to perform multiple local spatial feature perception on the logging while drilling data; wherein the recurrent neural network quantifies the variable sequential features based on the results corresponding to the multiple local spatial feature perceptions.

[0080] For example, in some specific implementations, the LWD data is multi-dimensional, and these high-dimensional data have autocorrelation, which can reflect the state changes of the current formation (local), and then characterize the changing trend of effective stress, that is, the input high-dimensional LWD data contains rich spatiotemporal features. Therefore, in some preferred implementations, the extraction of these high-dimensional, autocorrelation features can also help the model better perceive the spatial information of the local formation. In some specific application scenarios, convolutional neural networks (CNNs), as a powerful deep learning model, are widely used in many fields such as image processing, speech recognition, and time series data analysis. Its structure includes three layers, namely: convolution layer, activation layer, and maximum pooling layer. In the present invention, since the input LWD data is sequential data, in order to ensure that the input feature sequence has sufficient local features when further sequential features are extracted, the embodiment of the present invention can pre-use convolutional neural networks to perform multiple local spatial feature perceptions on the LWD data.

[0081] S200, using a recurrent neural network to quantify variable sequence characteristics of continuous time series logging while drilling data to obtain sequence characteristics;

[0082] It should be noted that, in some embodiments, step S200 may include the following steps: using a long short-term memory network to quantify the variable sequential characteristics of continuous time series logging while drilling data through memory units and a gating mechanism to obtain sequential characteristics; wherein the gating mechanism includes an input gate, a forgetting gate, and an output gate.

[0083] For example, in some specific implementations, LSTM is a type of recurrent neural network specifically designed to address the difficulty of traditional RNNs in capturing long-range dependencies in long sequences of data. By introducing a "memory unit" and a "gating mechanism" (including an input gate, a forget gate, and an output gate), LSTM can selectively remember or forget information during training, effectively retaining important historical data while filtering out irrelevant information.

[0084] In some embodiments, a long short-term memory network is used to quantify the variable sequential features of continuous time series while drilling logging data through memory units and a gating mechanism to obtain sequential features, which can include the following steps: using a forget gate to obtain forgotten information based on the hidden state output at the first moment and the input data at the second moment; wherein the second moment represents the next moment after the first moment, and the input data is determined based on the while drilling logging data; using an input gate to obtain added information based on the hidden state output at the first moment and the input data at the second moment; using an output gate to obtain a hidden state based on the hidden state output at the first moment and the input data at the second moment; using a memory unit to obtain new information based on the hidden state output at the first moment and the input data at the second moment, and then updating the state of the memory unit at the second moment based on the state of the memory unit at the first moment in combination with the forgotten information, the added information and the new information; constructing a hidden state output at the second moment based on the state of the memory unit at the second moment in combination with the hidden state; taking the second moment as the first moment, returning to execute the step of obtaining forgotten information based on the hidden state output at the first moment and the input data at the second moment using a forget gate, until all the while drilling logging data in the continuous time series are processed and the sequential features are obtained.

[0085] For example, in some specific implementations, the forget gate determines the information to be forgotten at this moment based on the state of the memory unit at the previous moment:

[0086] f t =σ(W f [h t-1 ;x t ]+b f )

[0087] The input gate needs to add relevant information based on the state of the memory unit at the previous moment and the input at this moment:

[0088] i t =σ(W i [h t-1 ;x t ]+b i )

[0089] At the same time, new information is generated:

[0090]

[0091] The memory unit updates the state of this moment by using the information generated at the previous moment:

[0092]

[0093] The output gate combines the state of the memory cell at the previous moment with the input at the current moment, and the output gate also determines the next hidden state:

[0094] O t =σ(W o [h t-1 ;x t ]+b o )

[0095] Output of Cain state:

[0096] h t =O t tanh(C t )

[0097] In the above formula, the input at this moment is x t ; Output at the previous moment—h t-1 ; Weight matrix in the forget gate—W f ; Neuron bias vector in forget gate—b f ; Weight matrix in the input gate—W i ; Neuron bias vector in the input gate—b i ; Weight matrix in the output gate—W o ; Neuron bias vector in the output gate—b o ; Weight matrix in memory unit—W c ; The neuron bias vector in the memory unit—b c .

[0098] S300, based on the sequential characteristics, constructing the weighted values of each logging parameter in the logging while drilling data through the attention mechanism, and then obtaining the formation sedimentary characteristics through the preset weight parameter matrix;

[0099] It should be noted that in some embodiments, based on the sequential features, the weighted values of each logging parameter in the logging while drilling data are constructed through the attention mechanism, which can include the following steps: based on the sequential features corresponding to the logging parameters combined with a preset learnable weight matrix, the query vector, key vector and value vector corresponding to the logging parameters are constructed using the attention mechanism; based on the query vector and the key vector combined with a preset scaling factor, the attention weight corresponding to the logging parameter is constructed through a normalized exponential function; and the weighted value corresponding to the logging parameter is obtained according to the product of the attention weight and the value vector.

[0100] For example, in some specific embodiments, 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 3 As shown, the specific implementation can be as follows:

[0101] 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:

[0102]

[0103] 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:

[0104] Q = x n W q n

[0105] K=x n W k n

[0106] 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:

[0107] O=Z n ×V n

[0108]

[0109] 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.

[0110] 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:

[0111] Output=O×W

[0112] S400: Inputting formation sedimentation characteristics into a prediction model to obtain a predicted formation pressure value of a target area.

[0113] Among them, the prediction model is trained based on a fully connected network through formation feature representation marked with actual formation pressure.

[0114] In some embodiments, the method may further include the following steps: inputting the formation characteristic representation into a fully connected network for processing to obtain a formation pressure training value; wherein the data structure of the formation characteristic representation is the same as the formation sedimentation characteristic; constructing a loss function based on the formation pressure training value corresponding to the formation characteristic representation and the actual formation pressure, and then feedback-adjusting the network parameters of the fully connected network through the error result of the loss function to obtain a prediction model.

[0115] For example, in some specific embodiments, the present invention constructs a prediction model by adopting a fully connected network and performs end-to-end training in combination with labeled actual formation pressure data. It can nonlinearly map the complex relationship between high-dimensional features and formation pressure, thereby improving prediction accuracy and generalization ability while ensuring real-time performance.

[0116] 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.

[0117] First of all, it should be noted that formation pressure, as one of the important indicators of formation stability, is closely related to the safety and efficiency of deep-sea resource exploration and production. However, deep-sea oil and gas resource exploration and production generally faces the problem of abnormal formation pressure. If timely response is not made, it will lead to formation instability problems such as lost circulation and well kick during drilling and completion. Therefore, timely and accurate calculation of formation pressure is crucial to ensure the safety and efficiency of deep-sea oil and gas and even efficient development.

[0118] In light of this, this paper addresses the challenge of high-precision, real-time pressure calculation. It proposes a real-time pressure calculation method based on formation sedimentary characteristics, breaking through the traditional static modeling paradigm to achieve high-precision, real-time calculation of formation pressure at drilling points. The proposed method and model are validated using logging-while-drilling data from actual wells.

[0119] The sedimentation process of deep-sea strata is often accompanied by the formation of abnormal pressure, and the problem of abnormal high pressure is inevitable during the exploitation of oil and gas resources. At present, the high-precision calculation methods (models) for formation pressure are mostly post-drilling calculations (fitting), which cannot provide real-time guidance for the operating parameters of the exploration process. Therefore, the present invention analyzes the characteristics of formation pressure to clarify the requirements for real-time and high precision. Considering that the logging while drilling parameters are real-time and can characterize the mechanical properties of the relevant formations. Therefore, the present invention uses the logging while drilling data to propose a formation intelligent calculation model to meet the real-time requirements. However, the real-time logging while drilling parameters are diverse, some data are noisy and have autocorrelation. Therefore, how to mine the local characteristics of the formation space in the data and eliminate noise points is a specific problem description of the real-time calculation model.

[0120] Based on the above analysis of formation pressure characteristics, 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 perspectives 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 the figure. Figure 4 Specifically, the embodiment of the present invention can be implemented through the following process:

[0121] 1. Multiple local spatial feature perception:

[0122] LWD data is multidimensional, and this high-dimensional data exhibits autocorrelation, which can reflect changes in the current (local) formation state and, in turn, characterize the changing trends in effective stress. This high-dimensional LWD input data contains rich spatiotemporal features. Therefore, in formation pressure calculation models, extracting these high-dimensional, autocorrelated features helps the model better perceive the spatial information of the local formation.

[0123] Convolutional neural networks (CNNs), as powerful deep learning models, are widely used in fields such as image processing, speech recognition, and time series data analysis. Their architecture consists of three layers: convolutional, activation, and max pooling. In this paper, since the input LWD data is sequential, the input feature sequence must have sufficient local features for subsequent sequential feature extraction.

[0124] When performing the convolution operation, the present invention uses a one-dimensional convolution operation, and the calculation process is as follows:

[0125]

[0126] in, Represents the input vector x i ={x1,x2,…,x n}The feature vector obtained after the convolution operation; for The corresponding weight; r is the step size; represent The corresponding threshold; σ is the sigmoid activation function.

[0127] After the convolution operation, a large number of feature vectors are obtained. The maximum pooling layer is used to reduce the dimension of the features and thus the number of parameters. At the same time, the use of the pooling layer not only improves the computational efficiency to achieve the purpose of real-time calculation, but also prevents the computational model from overfitting. The calculation process is as follows:

[0128]

[0129] In addition, the structure also takes advantage of the position invariance and denoising performance of CNN. CNN has translation invariance, that is, the feature extraction results will not be significantly affected by small translations of the input data. In the present invention, the stratigraphic features often vary in space, that is, these changes can be effectively captured during the feature extraction process without being affected by their specific positions, which can effectively improve the applicability of the model; in terms of noise reduction: the model designed by the present invention is a real-time calculation model, and the input data may be subject to noise interference. CNN can perform preliminary denoising on the input data through convolution and pooling operations, thereby extracting cleaner signal features and ensuring the robustness of the model.

[0130] In summary, the theoretical analysis of CNN combined with actual working conditions shows that CNN can improve the accuracy and reliability of predictions in formation pressure calculations by extracting local spatial features of the formation and improving the model's applicability and robustness. Therefore, the present invention selects the CNN model as the local spatial feature extraction unit in the formation pressure calculation process.

[0131] 2. Stratigraphic sedimentary characteristics extraction model:

[0132] (1) Variable Sequential Feature Quantization Unit Based on Recurrent Neural Network:

[0133] Through the mechanism analysis of the formation pressure formation process, it can be seen that the formation pressure is continuous, that is, it has a certain degree of sequentiality. In addition, the drilling process and the sedimentation process show that the logging data also has a certain degree of sequentiality, that is, it has autocorrelation in the same layer. Therefore, in order to ensure accuracy, it is crucial to extract this sequential feature in the formation pressure calculation process. Although the autocorrelation of each parameter is extracted in the local feature extraction unit of the formation space, the size of the convolution kernel is 3 when performing the convolution operation, and the logging parameters are set to evenly distribute 6 measuring points along the length of the well. Therefore, this part can only characterize the local (0.5m) feature changes of the formation and cannot capture long-range dependencies well. Therefore, the present invention uses LSTM to capture the long-range dependencies of each logging data.

[0134] LSTM is a type of recurrent neural network, specifically designed to solve the problem that traditional RNNs have difficulty capturing long-range dependencies in long sequence data. By introducing "memory units" and "gating mechanisms" (including input gates, forget gates, and output gates), LSTM can selectively remember or forget information during the training process, thereby effectively maintaining the memory of important historical data while filtering out irrelevant information. This structure gives LSTM a significant advantage in processing long time series and data with complex dependencies. At the same time, compared to traditional RNNs, LSTM can better solve the problems of gradient disappearance and gradient explosion, enabling it to capture the temporal pattern of data over a longer time span. Its structure is as follows: Figure 5 As shown, specifically:

[0135] The forget gate determines the information that needs to be forgotten at this moment based on the state of the memory unit at the previous moment:

[0136] f t =σ(W f [h t-1 ;x t ]+b f )

[0137] The input gate needs to add relevant information based on the state of the memory unit at the previous moment and the input at this moment:

[0138] i t =σ(W i [h t-1 ;x t ]+b i )

[0139] At the same time, new information is generated:

[0140]

[0141] The memory unit updates the state of this moment by using the information generated at the previous moment:

[0142]

[0143] The output gate combines the state of the memory cell at the previous moment with the input at the current moment, and the output gate also determines the next hidden state:

[0144] O t =σ(W o [h t-1 ;x t ]+b o )

[0145] Output of Cain state:

[0146] ht =O t tanh(C t )

[0147] In the above formula, the input at this moment is x t ; Output at the previous moment—h t-1 ; Weight matrix in the forget gate—W f ; Neuron bias vector in forget gate—b f ; Weight matrix in the input gate—W i ; Neuron bias vector in the input gate—b i ; Weight matrix in the output gate—W o ; Neuron bias vector in the output gate—b o ; Weight matrix in memory unit—W c ; The neuron bias vector in the memory unit—b c ;

[0148] The gating mechanism of LSTM can automatically ignore irrelevant information in the formation pressure prediction by dynamically adjusting the impact of data at different times and depths, thereby improving the training efficiency and prediction accuracy of the formation pressure calculation model. At the same time, for complex stratigraphic structures (faults, changes in lithology, etc.), the flexibility of the input gate and the forget gate allows LSTM to update the memory state under specific conditions and adapt to different geological environments, especially when the pressure patterns at different depths change dramatically, it can be adjusted more accurately to improve the applicability of the model. Therefore, the present invention selects the LSTM model as the basic unit for sequential feature extraction in the formation pressure calculation process, laying the foundation for subsequent formation pressure calculation (prediction).

[0149] (2) Stratigraphic sedimentary characteristics extraction unit based on improved attention mechanism:

[0150] When calculating formation pressure, the input logging data is diverse, and different parameters have varying effects on formation pressure. Selecting the appropriate input from a multitude of input features based on the spatial characteristics of the formation is a key step in ensuring calculation accuracy and applicability.

[0151] The attention mechanism is a technology that has the ability to dynamically adjust weights in deep learning and is often used in conjunction with convolutional neural networks (CNNs) and long short-term memory networks (LSTMs). In formation pressure prediction, the attention mechanism dynamically assigns weights to well logging data at different depths, adaptively adjusting the model's focus based on changes in formation characteristics. Similar to humans' selective attention to important information, the model can flexibly identify and focus on depths or features that have a greater impact on prediction results. Its core mechanism relies on a query vector Q, a key vector K, and a numerical vector V. Specifically, the key vector K represents the features of each depth point, and the query vector Q is used to map out the information that needs to be focused on in the current task. The attention mechanism dynamically adjusts the weight of each feature by calculating the relevance of this information. Ultimately, the weighted important features are input as a numerical vector V, ensuring that the model can flexibly and accurately predict formation pressure.

[0152] This ability to dynamically adjust weights prevents the model from treating all depth points equally, thereby reducing unnecessary computational burden and improving the model's prediction accuracy. In complex geological conditions, the attention mechanism can effectively focus on the information that is most valuable for calculations. In addition, changes in formation pressure are often related to the interaction between multiple depth layers. Pressure changes at certain depth points may be driven by features of distant depth layers. The attention mechanism can dynamically calculate the dependencies between these depths based on the relationship between each depth point, thereby capturing global and local pressure change characteristics. In terms of model interpretability, by assigning weights to different features, we can better understand the key parameters that affect formation pressure and help identify structural changes in the formation.

[0153] 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, whose structure is as follows: Figure 3 As shown:

[0154] 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:

[0155]

[0156] 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:

[0157] Q = x n W q n

[0158] K=x n W k n

[0159] 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:

[0160] O=Z n ×V n

[0161]

[0162] 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.

[0163] 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:

[0164] Output=O×W

[0165] 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.

[0166] In some specific application scenarios, the present invention integrates actual drilling data from the North Carnarvon Basin, along with mechanistic knowledge and expert experience, to determine the model's input variables. After removing outliers and normalizing the LWD data, the proposed real-time formation pressure calculation model based on formation sedimentary characteristics is used for practical verification. Furthermore, by comparing the proposed method with existing intelligent computing (prediction) methods, the superiority of the proposed method in terms of accuracy and real-time performance is verified, thereby demonstrating its effectiveness:

[0167] ① Multi-source LWD data selection:

[0168] In the formation pressure prediction model, multi-source logging while drilling data provides a wealth of formation information, including parameters such as porosity, resistivity, gamma rays, and acoustic wave transit time. How to effectively select the most relevant features from these multi-source logging data directly affects the accuracy and computational efficiency of the model. First, the present invention relies on empirical methods to select logging while drilling parameters that are correlated with formation mechanical characteristics (see Table 1 below for details). This process is mainly qualitatively determined through theoretical foundations such as stress state and deformation characteristics analysis, geological background and sedimentary environment analysis, historical cases and field experience, and laboratory test results, combined with engineering and laboratory experience.

[0169] Table 1

[0170]

[0171]

[0172] Secondly, after determining the LWD parameters that are correlated with formation pressure, the present invention takes into account the nonlinear correlation between the relevant logging data and formation pressure and adopts the mutual information method for feature screening. This method quantitatively evaluates the mutual dependence between each logging parameter and the target variable (formation pressure), which is defined as:

[0173]

[0174] Where p(x,y) is the joint probability distribution of X and Y, p(x) and p(y) are the marginal probability distributions of X and Y, respectively. The base of log can be 2 or e. If it is 2, the unit of mutual information is bits. According to the above formula, when I(X;Y) = 0, X and Y are independent; when I(X;Y) > 0, X and Y have a certain degree of dependence, and a larger value indicates a higher correlation between the relevant parameters.

[0175] like Figure 6The figure shows the correlation thermodynamic diagram between relevant logging parameters and between logging parameters and formation pressure. According to the mutual information value, the present invention selects natural gamma, depth, acoustic wave time difference, formation density and well diameter as the input parameters of the model to avoid redundant information in the input parameters and improve the model performance.

[0176] ②Data preprocessing:

[0177] Due to the dual influence of the drilling process and the marine-geological environment, the monitoring data of logging-while-drilling sensors often contains many outliers and missing values, which can affect the training effect and computational efficiency of the model. Although the model design takes into account the issues of outlier and missing value processing, to ensure the accuracy of the model during the training and testing process, the present invention removes outliers, fills in missing values, and normalizes the raw data.

[0178] 1) Outlier removal:

[0179] The purpose of removing outliers is to improve the accuracy and stability of the model and avoid the interference of extreme values on the training process, especially when the outliers are far away from the normal data distribution, which may cause the model to misjudge or converge slowly. Common outlier removal methods are mainly statistical methods (such as Z-score and IQR). Among them, the Z-score method can effectively identify and remove outliers that deviate greatly from most of the data by calculating the deviation of the data point from the mean (in units of standard deviation). In addition, since drilling data often does not conform to the normal distribution, Z-score is particularly suitable for those situations where most of the data are relatively concentrated and a few data points deviate significantly from the mean. Therefore, the present invention selects the Z-score method to remove outliers in the original data. The principle is as follows:

[0180]

[0181] Where Z is the normalized data point, X is the original data point, μ is the mean of the data set, and σ is the standard deviation of the data set. In order to visualize the outliers, this paper uses box plots to show the distribution range of the data and uses scatter plots to intuitively represent the outliers, such as Figure 7 shown.

[0182] 2) Missing value filling:

[0183] Missing data of characteristic parameters often leads to instability or deviation in model training, thereby affecting the accuracy of prediction. Therefore, the present invention fills in the missing values of the original data during data preprocessing. Based on the theory of stratigraphic sedimentation: the stratigraphic sedimentation process is slow, that is, the stratigraphic related characteristics are continuous. The KNN method (K-nearest neighbor) can predict missing values based on the similarity of the data, and use the neighboring relationship of known data points to fill in the unknown values (by calculating the distance between the missing value sample and other samples, selecting the K most similar neighbors, and filling the missing values with the mean or mode of these neighbors). This method is suitable for situations where there are complex patterns or nonlinear relationships in the data, and does not rely on the distribution assumptions of the data, and can effectively restore the potential structure of the missing data. Therefore, the present invention selects the KNN method as the missing value filling method.

[0184] 3) Normalization:

[0185] Data normalization is a common preprocessing operation in machine learning and data mining. Its main purpose is to unify data of different scales and dimensions into a standard range, thereby reducing the negative impact of differences between different features on model performance. Taking the present invention as an example, the variables involved in the present invention (such as well depth, drilling pressure, etc.) have different dimensions and numerical ranges. If normalization is not performed, some features may have an excessive impact on model training, while other features are ignored. Therefore, normalization can not only accelerate the convergence rate of the model, but also improve the accuracy and stability of the model. Two commonly used data normalization methods are Z-Score normalization and Min-Max normalization. During the drilling process, the parameters involved, such as well depth, drilling pressure, and formation pressure, usually show an increasing trend with depth, and these parameters often do not follow a standard normal distribution. For such data, Z-Score normalization may not be applicable because it relies on the mean and standard deviation of the data, and the data distribution of features such as depth and drilling pressure is often biased or uneven. Therefore, the present invention selects the Min-Max method to normalize the relevant well logging data.

[0186]

[0187] Among them, max(X i ) and min(X i ) are the maximum and minimum values of different data, X i are the values of different variables, The results are normalized for different variables.

[0188] After data preprocessing, the data of well A were divided into training and test sets in a ratio of (9:1), and wells B and C were used as test wells to verify the accuracy, real-time performance, and applicability of the model.

[0189] ③Experimental verification and effect analysis:

[0190] For example, the present invention uses the actual drilling data of three wells A, B, and C in the North Carnarvon Basin to test the real-time and accuracy performance of the proposed calculation model, and compares its performance with the existing formation pressure intelligent calculation model to verify the effectiveness and advancement of the proposed model.

[0191] (1) Calculation accuracy:

[0192] To verify the reliability and advancement of the proposed model, the present invention analyzes and discusses the computational accuracy of the proposed model and the comparative models (MLP, RF, SVM, and XGBoost). Table 2 shows the results of training and testing the above algorithm models on three wells A, B, and C. When quantitatively evaluating the computational accuracy of each model, the present invention uses the root mean square error (RMSE) as the evaluation indicator instead of the coefficient of determination (R 2 ) to measure the degree of deviation of the calculation results. The reason is as follows: RMSE measures the error size with the same dimension (unit) as the target value, and intuitively reflects the absolute deviation between the predicted value and the true value. In the prediction of formation pressure, engineering practice pays more attention to the deviation between the predicted pressure and the true pressure. RMSE can directly reveal whether the prediction error is within an acceptable range. 2 It is a dimensionless index that only indicates the relative quality of the fit and does not provide information on the actual size of the error. 2 It measures the model's ability to explain the variance of the target value, and its range is [-∞,1]. 2 While suitable for evaluating overall fit, it may fail in scenarios where the true pressure values are distributed within a narrow range or the model prediction deviation is large. Because RMSE directly reflects the actual scale of the prediction error, provides a more intuitive explanation of model performance, and is more suitable for precise error control in engineering practice, it is selected as the core accuracy evaluation metric for the formation pressure calculation problem in this paper.

[0193] Table 2

[0194] Proposed MLP RF Support Vector Machine XGBoost Well A 0.0493 2.8057 0.7048 2.0027 0.9060 Well B 1.1726 6.3186 4.5863 4.6294 4.9264 Well C 0.1537 6.0736 5.9186 6.5841 6.7874

[0195] As can be seen from the RMSE evaluation index values of each model in Table 2 above, the RMSE value of the proposed model on multiple well data sets (Well A, Well B, Well C) is significantly lower than that of other models. The RMSE for Well A is 0.0493, showing excellent accuracy; while the RMSE for Wells B and C are 1.1726 and 0.1537, respectively, which still outperform other models. Compared with MLP, RF, SVM, and XGBoost, the latter have larger errors on Wells A, B, and C, respectively. In particular, the errors on Wells B and C are significantly increased, indicating that the other models have poor adaptability to data from different wells. This shows that the proposed model not only has obvious advantages in accuracy, but also has better generalization ability and stability on different well and formation data than traditional models, demonstrating its advanced nature and practical application potential in formation pressure prediction.

[0196] like Figure 8 The figure shows the goodness-of-fit comparison results of each model on the training and test sets. The horizontal axis represents the label value (true formation pressure value), and the vertical axis represents the formation pressure value predicted by each model. As can be seen from the figure, although the proposed model far outperforms other models in terms of the RMSE evaluation metric, indicating that its calculation error is within an acceptable range and the model has good applicability, the analysis of the fitting plot of the true and predicted values reveals that the fitting effect of the proposed model on Wells B and C still needs to be further improved. Specifically, Well C and Well A (training well) have similar formation pressure anomalies (fluid expansion), so the proposed model performs well on Well C. In particular, the formation pressure values calculated by the model are close to the true values in the 3500-4300 m well section, further verifying the model's effectiveness in capturing local spatial characteristics of the formation and extracting sedimentary properties. However, in Well B, the model fit is relatively poor, especially in the 2400-3400 m well section, where the calculated values are mostly lower than the true values and have a large deviation. This phenomenon can be attributed to the fact that the cause of the pressure anomaly in Well B is different from that in Well A. The model fails to fully adapt to this difference, resulting in a large deviation in its prediction results.

[0197] Nevertheless, as Figure 9 As shown in the figure, the pressure calculation results for Well B still closely track the changes in the true pressure values. This demonstrates that the model maintains a certain degree of adaptability and stability when processing data that differs significantly from the training wells, further verifying the model's effectiveness in capturing local spatial characteristics of the formation and extracting sedimentary properties. Therefore, although the fitting effect for Well B is poor, the model still demonstrates strong ability to extract formation characteristics. Future optimization of the model can improve its applicability and prediction accuracy under different causes of abnormal pressure.

[0198] (2) Real-time performance verification:

[0199] In order to fully verify the real-time performance of the model, the present invention conducts a detailed evaluation of the average running time of the proposed model and the comparison models (MLP, RF, SVM and XGBoost). Each model is run 20 times under the same hardware environment, and the average running time of each calculation is calculated (see Table 3). The distribution and dispersion of the running time of each model are intuitively displayed through box plots (such as Figure 10 The results show that the proposed model is highly stable in terms of computational efficiency, with a single run time only slightly higher than that of traditional small-parameter machine learning models (RF and SVM), but it has a significant advantage in prediction accuracy compared to these models.

[0200] Table 3

[0201]

[0202] The box plot analysis further reveals the discrete characteristics and extreme value distribution of the running time of each model. In comparison, the running time of the proposed model is not only stable, but also has a smaller degree of discreteness, and the frequency of extreme values is significantly lower than that of other deep learning models, showing strong robustness. This stable and efficient characteristic enables the model proposed in this invention to better meet the needs of real-time calculation of formation pressure, and provides a solid guarantee for its application in actual engineering scenarios. Specifically, Figure 10 As shown in the figure, by comparing the running time distribution of MLB, RF, SVM, XGBoost and the proposed model (Proposed), it can be seen that there are significant differences in the running efficiency and stability of each model. MLB and XGBoost have the shortest running time and the lowest volatility, showing extremely high stability, which is suitable for scenarios with extremely high real-time requirements; RF and SVM have significantly longer running times, especially SVM, which has the largest degree of discreteness and the presence of extreme values, indicating that its running efficiency is low and unstable, and is not suitable for real-time tasks. The running time of the proposed model (Proposed) is slightly higher than that of MLB and XGBoost, but it still maintains a lower volatility, and its performance is better than that of the lightweight model, proving that it has good real-time and stability while ensuring high precision. Therefore, the model proposed in the present invention takes into account both real-time and accuracy in the task of formation pressure calculation. At the same time, these experimental results also provide data support for the optimization direction of future models.

[0203] In summary, this paper first analyzes the characteristics of formation pressure during deep-sea drilling. Leveraging the mechanistic properties of formation deposition and logging-while-drilling data, it proposes a real-time pressure calculation model based on formation deposition characteristics. The proposed model is then validated using real-time drilling data from the North Carnarvon oil and gas wells, demonstrating significant advantages across multiple well datasets. First, the proposed model achieves the lowest prediction error across three wells using the RMSE evaluation metric, demonstrating its high accuracy and stability in formation pressure calculation. Although the fitting results for Wells B and C are less than ideal—particularly for Well B, where the model's predicted values are often lower than the true values and exhibit significant deviations—this is primarily due to the different causes of the abnormal pressure in Well B compared to Well A, the model still captures the overall trend well, demonstrating its effectiveness in localizing formation spatial perception and extracting depositional characteristics. Furthermore, the proposed model exhibits excellent real-time performance. Compared to other traditional models, its computation time is stable and efficient, meeting the real-time computing requirements of practical engineering applications. Overall, the proposed model has strong generalization ability and good real-time performance while ensuring high accuracy, and can effectively cope with the pressure calculation tasks of different wells, proving its advanced nature and practical application value in formation pressure prediction.

[0204] On the other hand, Figure 11 As shown, an embodiment of the present invention provides a formation pressure prediction device 900, which may include:

[0205] The first module 901 is used to obtain logging while drilling data of the target area;

[0206] The second module 902 is used to quantify the variable sequence characteristics of the continuous time series logging while drilling data using a recurrent neural network to obtain sequence characteristics;

[0207] The third module 903 is used to construct weighted values of various logging parameters in the logging while drilling data based on the sequential characteristics through the attention mechanism, and then obtain the formation sedimentary characteristics through the preset weight parameter matrix;

[0208] The fourth module 904 is used to input the formation sedimentary characteristics into the prediction model to predict the formation pressure of the target area;

[0209] Among them, the prediction model is trained based on a fully connected network through formation feature representation marked with actual formation pressure.

[0210] In some embodiments, the apparatus further includes a fifth module configured to perform the following operations:

[0211] Perform data preprocessing on LWD data;

[0212] Among them, data preprocessing includes outlier removal, missing value filling and data normalization.

[0213] In some embodiments, the apparatus further includes a sixth module configured to perform the following operations:

[0214] Use the preset convolutional neural network to perceive multiple local spatial features of logging while drilling data;

[0215] Among them, the recurrent neural network quantifies the variable sequential features based on the results corresponding to multiple local spatial feature perceptions.

[0216] In some embodiments, the apparatus further includes a seventh module configured to perform the following operations:

[0217] The formation feature representation is input into the fully connected network for processing to obtain the formation pressure training value; wherein the data structure of the formation feature representation is the same as the formation sedimentary feature;

[0218] A loss function is constructed based on the formation pressure training value corresponding to the formation characteristic representation and the actual formation pressure. Then, the network parameters of the fully connected network are adjusted through feedback using the error result of the loss function to obtain a prediction model.

[0219] 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.

[0220] Another embodiment of the present invention 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 above-described formation pressure prediction method. The electronic device can be any smart terminal, including a tablet computer and an in-vehicle computer.

[0221] 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.

[0222] like Figure 12 As shown, Figure 12 The hardware structure of an electronic device 1000 according to another embodiment is shown. The electronic device 1000 includes:

[0223] 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.

[0224] 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.

[0225] Input / output interface 1003, used to implement information input and output;

[0226] 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.);

[0227] 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 );

[0228] 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 .

[0229] 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.

[0230] 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.

[0231] 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.

[0232] 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.

[0233] 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.

[0234] 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.

[0235] 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.

[0236] 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.

[0237] 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.

[0238] 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.

[0239] 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.

[0240] 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.

[0241] 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.

[0242] 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.

[0243] 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.

[0244] 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.

[0245] 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.

[0246] 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 formation pressure prediction method, characterized in that: The following steps are involved: Acquire LWD data in target areas; quantifying variable sequence characteristics of the continuous time series logging while drilling data using a recurrent neural network to obtain sequence characteristics; Based on the sequential characteristics, the weighted values of the logging parameters in the LWD data are constructed through an attention mechanism, and then the formation sedimentary characteristics are obtained through a preset weight parameter matrix. Inputting the formation sedimentary characteristics into a prediction model to predict a formation pressure prediction value of the target area; The prediction model is obtained by training the formation feature representation based on a fully connected network marked with actual formation pressure.

2. The formation pressure prediction method according to claim 1, characterized in that: After the step of acquiring the logging while drilling data of the target area, the method further comprises the following steps: performing data preprocessing on the logging while drilling data; The data preprocessing includes outlier removal, missing value filling and data normalization.

3. The formation pressure prediction method according to claim 1, characterized in that: Before the step of quantifying variable sequence characteristics of the continuous time series logging while drilling data using a recurrent neural network, the method further includes the following steps: Using a preset convolutional neural network to perform multiple local spatial feature perception on the logging while drilling data; The recurrent neural network quantifies the variable sequential features based on the results corresponding to the multiple local spatial feature perceptions.

4. The formation pressure prediction method according to claim 1, characterized in that: The method of using a recurrent neural network to quantify the variable sequence characteristics of the continuous time series logging while drilling data to obtain the sequence characteristics includes the following steps: quantifying variable sequence characteristics of the continuous time series LWD data using a long short-term memory network through memory units and a gating mechanism to obtain the sequence characteristics; The gating mechanism includes an input gate, a forget gate and an output gate.

5. The formation pressure prediction method according to claim 4, characterized in that: The method of quantifying the variable sequence characteristics of the continuous time series LWD data by using a long short-term memory network through memory units and a gating mechanism to obtain the sequence characteristics includes the following steps: Obtaining forgetting information using the forget gate according to the hidden state output at the first moment and the input data at the second moment; wherein the second moment represents the next moment after the first moment, and the input data is determined based on the logging while drilling data; According to the hidden state output at the first moment and the input data at the second moment, obtaining additional information by processing with the input gate; Obtaining a hidden state by processing the hidden state output at the first moment and the input data at the second moment using the output gate; According to the hidden state output at the first moment and the input data at the second moment, the memory unit is used to process new information, and then the state of the memory unit at the second moment is updated according to the state of the memory unit at the first moment in combination with the forgotten information, the added information, and the new information; Constructing a hidden state output at the second moment according to the state of the memory unit at the second moment in combination with the hidden state; The second moment is used as the first moment, and the step of using the forget gate to process the forgotten information based on the hidden state output of the first moment and the input data of the second moment is returned to be executed until all the LWD data in the continuous time series are processed and the sequential features are obtained.

6. The formation pressure prediction method according to claim 1, characterized in that: The method of constructing a weighted value of each logging parameter in the logging while drilling data based on the sequential features through an attention mechanism includes the following steps: Based on the sequential features corresponding to the well logging parameters and a preset learnable weight matrix, the attention mechanism is used to construct a query vector, a key vector, and a value vector corresponding to the well logging parameters; Based on the query vector and the key vector combined with a preset scaling factor, an attention weight corresponding to the logging parameter is constructed by a normalized exponential function; The weighted value corresponding to the logging parameter is obtained according to the product of the attention weight and the value vector.

7. The formation pressure prediction method according to claim 1, characterized in that: The method further comprises the following steps: Inputting the formation feature representation into the fully connected network for processing to obtain a formation pressure training value; wherein the data structure of the formation feature representation is the same as the formation sedimentary feature; A loss function is constructed based on the formation pressure training value corresponding to the formation characteristic representation and the actual formation pressure, and then the network parameters of the fully connected network are feedback-adjusted through the error result of the loss function to obtain the prediction model.

8. A formation pressure prediction device, characterized in that: include: The first module is used to obtain the logging while drilling data of the target area; The second module is used to quantify the variable sequence characteristics of the continuous time series logging while drilling data using a recurrent neural network to obtain sequence characteristics; The third module is used to construct weighted values of various logging parameters in the logging while drilling data based on the sequential features through an attention mechanism, and then obtain the formation sedimentary characteristics through a preset weight parameter matrix; The fourth module is configured to input the formation sedimentary characteristics into a prediction model to predict the formation pressure of the target area; The prediction model is obtained by training the formation feature representation based on a fully connected network marked with actual formation pressure.

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