Method and device for predicting underground parameters through dynamically updated ground parameters

Abstract

The invention provides a method and a device for predicting underground parameters by dynamically updating ground parameters. Obtaining target first ground feature data corresponding to the first ground feature in ground feature data of a target well of the to-be-detected area; determining target feature data corresponding to the underground features by using a multi-target network prediction model according to target first ground feature data of a target well of the to-be-detected area; wherein the multi-target network prediction model is obtained by training based on second ground feature data in historical sample data, underground feature data in the historical sample data, a second ground feature prediction value and an underground feature prediction value; the first ground feature and the second ground feature are obtained by splitting the ground features according to the correlation between the ground features and the underground features determined according to historical sample data; and according to the target feature data, a drilling strategy for the target well is determined. Therefore, based on prediction of the underground characteristic data, the efficiency and accuracy of determining the drilling strategy are improved.

Description

Technical Field

[0001] This specification belongs to the technical field of oil extraction, and particularly relates to a method and device for predicting downhole parameters based on dynamically updated surface parameters. Background Art

[0002] Currently, oil and gas exploration and development are moving towards complex oil and gas resources such as deep and ultra-deep layers from conventional oil and gas resources. Under complex geological conditions, due to the fact that downhole parameter data is easily interfered by downhole temperature and pressure, there is a lot of noise and distortion, and the cost of downhole measurement equipment is high and the maintenance difficulty is large. Therefore, existing downhole characteristic data measurement equipment and methods are difficult to be popularized on a large scale, and it is impossible to accurately obtain downhole parameter data, resulting in problems of poor efficiency and accuracy in determining the drilling strategy of oil and gas wells.

[0003] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0004] This specification provides a method and device for predicting downhole parameters based on dynamically updated surface parameters, which avoid the problems of poor efficiency and accuracy in determining the drilling strategy caused by the inability to accurately obtain downhole characteristic data, thereby optimizing the drilling operation and improving the drilling speed and efficiency.

[0005] This specification provides a method for predicting downhole parameters based on dynamically updated surface parameters, including:

[0006] Receiving a processing request for the drilling strategy of a target well in a to-be-detected area;

[0007] Responding to the processing request, and obtaining target first surface characteristic data corresponding to a first surface characteristic in the surface characteristic data of the target well in the to-be-detected area;

[0008] Using a multi-objective network prediction model, based on the target first surface characteristic data of the target well in the to-be-detected area, determining target characteristic data corresponding to downhole characteristics; wherein, the multi-objective network prediction model is trained based on second surface characteristic data corresponding to a second surface characteristic in historical sample data, downhole characteristic data corresponding to the downhole characteristics in the historical sample data, a second surface characteristic prediction value, and a downhole characteristic prediction value, the second surface characteristic prediction value and the downhole characteristic prediction value are processed by the multi-objective network prediction model according to first surface characteristic data corresponding to the first surface characteristic in the historical sample data, and the first surface characteristic and the second surface characteristic are obtained by splitting the surface characteristic according to the correlation degree between the surface characteristic and the downhole characteristic determined from the historical sample data; and determining the drilling strategy for the target well according to the target characteristic data.

[0009] In one embodiment, before determining the target feature data corresponding to the downhole feature according to the target first surface feature data of the target well in the area to be detected by using the multi-target network prediction model, the method further includes:

[0010] Obtain the historical sample data; wherein, the historical sample data includes the surface feature data corresponding to the surface feature and the downhole feature data corresponding to the downhole feature;

[0011] According to the correlation between the surface feature and the downhole feature, split the surface feature into the first surface feature and the second surface feature;

[0012] Use the initial multi-target network prediction model to determine the second surface feature prediction value and the downhole feature prediction value according to the first surface feature data;

[0013] According to the second surface feature prediction value, the second surface feature data, the downhole feature data, and the downhole feature prediction value, determine whether the initial multi-target network prediction model converges;

[0014] In the case where the initial multi-target network prediction model converges, obtain the multi-target network prediction model; wherein, the multi-target network prediction model is used to determine the downhole feature data of the target well in the area to be detected for formulating a drilling strategy according to the target first surface feature data of the target well in the area to be detected.

[0015] In one embodiment, determining whether the initial multi-target network prediction model converges according to the second surface feature prediction value, the second surface feature data, the downhole feature data, and the downhole feature prediction value includes:

[0016] According to a preset loss function, the loss weight corresponding to the second surface feature, the loss weight corresponding to the downhole feature, the second surface feature prediction value, the second surface feature data, the downhole feature data, and the downhole feature prediction value, determine a loss value, and determine whether the initial multi-target network prediction model converges according to the loss value, wherein the loss weight corresponding to the downhole feature is greater than the loss weight corresponding to the second surface feature.

[0017] In one embodiment, the method further includes:

[0018] In the case of reaching the model update period, obtain the incremental sample data within the model update period;

[0019] Perform model update processing on the multi-target network prediction model according to the incremental sample data.

[0020] In one embodiment, the downhole feature data corresponding to the downhole feature in the incremental sample data is not acquired, and the surface feature data corresponding to the second surface feature has been acquired. The model update process for the multi-target network prediction model according to the incremental sample data includes:

[0021] Update the parameters of the hidden layer in the multi-target network prediction model according to the surface feature data corresponding to the first surface feature and the second surface feature in the incremental sample data.

[0022] In one embodiment, the splitting of the surface feature into a first surface feature and a second surface feature according to the correlation between the surface feature and the downhole feature includes:

[0023] Determine the second surface feature as the surface features with a correlation greater than a preset correlation threshold among the surface features according to the correlation between the surface feature and the downhole feature, and determine the first surface feature as the surface features with a correlation not greater than the preset correlation threshold among the surface features.

[0024] In one embodiment, before splitting the surface feature into a first surface feature and a second surface feature according to the correlation between the surface feature and the downhole feature, it further includes:

[0025] Determine the correlation between the surface feature and the downhole feature according to the surface feature data and the downhole feature data according to the following formula:

[0026]

[0027] where r xy is the correlation between the Xth surface feature and the Yth downhole feature, Cov(X,Y) is the covariance between the Xth surface feature and the Yth downhole feature, and σ X and σ Y are the standard deviations of the Xth surface feature and the Yth downhole feature respectively.

[0028] This specification provides a device for predicting downhole parameters with dynamically updated surface parameters, including:

[0029] A request receiving module, configured to receive a processing request for the drilling strategy of a target well in a to-be-detected area;

[0030] A request response module, configured to obtain the target first surface feature data corresponding to the first surface feature in the surface feature data of the target well in the to-be-detected area in response to the processing request;

[0031] A model prediction module, which is used to utilize a multi-objective network prediction model to determine target feature data corresponding to downhole features according to target first surface feature data of a target well in a to-be-detected area; wherein, the multi-objective network prediction model is trained based on second surface feature data corresponding to second surface features in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as second surface feature prediction values and downhole feature prediction values, and the second surface feature prediction values and the downhole feature prediction values are obtained by the multi-objective network prediction model through processing first surface feature data corresponding to the first surface features in the historical sample data, and the first surface features and the second surface features are obtained by splitting the surface features according to the relevance between the surface features and the downhole features determined from the historical sample data;

[0032] A strategy determination module, which is used to determine a drilling strategy for the target well according to the target feature data.

[0033] This specification also provides an electronic device, which includes a processor and a memory for storing processor-executable instructions, and when the processor executes the instructions, a method for predicting downhole parameters with dynamically updated surface parameters is implemented.

[0034] This specification also provides a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed, a method for predicting downhole parameters with dynamically updated surface parameters is implemented.

[0035] A method for predicting downhole parameters based on ground parameters with dynamic update provided in this specification, which receives a processing request for a drilling strategy of a target well in a to-be-detected area; in response to the processing request, obtains target first ground feature data corresponding to a first ground feature in the ground feature data of the target well in the to-be-detected area; uses a multi-objective network prediction model to determine target feature data corresponding to downhole features according to the target first ground feature data of the target well in the to-be-detected area; wherein, the multi-objective network prediction model is trained based on second ground feature data corresponding to a second ground feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as a second ground feature prediction value and a downhole feature prediction value, and the second ground feature prediction value and the downhole feature prediction value are processed by the multi-objective network prediction model according to first ground feature data corresponding to the first ground feature in the historical sample data, and the first ground feature and the second ground feature are obtained by splitting the ground feature according to the correlation degree between the ground feature and the downhole feature determined from the historical sample data; determines a drilling strategy for the target well according to the target feature data. In this way, since the multi-objective network prediction model is trained based on second ground feature data corresponding to a second ground feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as a second ground feature prediction value and a downhole feature prediction value, and the first ground feature and the second ground feature are obtained by splitting the ground feature according to the correlation degree between the ground feature and the downhole feature determined from the historical sample data, therefore, in the case where the return frequency of the downhole feature data corresponding to the downhole features in the historical sample data is not high, the second ground feature data corresponding to the second ground feature in the historical sample data can be used to assist in training the multi-objective network prediction model, improving the training effect of the model, making the prediction of the model more accurate and real-time, and then the target feature data corresponding to the downhole features of the target well can be quickly and accurately determined through the multi-objective network prediction model, so as to determine a drilling strategy for the target well according to the target feature data, avoiding the problems of poor determination efficiency and determination accuracy of the drilling strategy caused by the inability to accurately obtain downhole feature data, and optimizing the drilling operation, improving the drilling speed and efficiency. Description of the Drawings

[0036] To more clearly illustrate the embodiments of this specification, the drawings required for use in the embodiments will be briefly introduced below. The drawings described below are only some of the embodiments recorded in this specification. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative efforts.

[0037] Figure 1It is a schematic flowchart of a method for predicting downhole parameters with dynamically updated surface parameters provided by an embodiment of this specification;

[0038] Figure 2 It is a schematic diagram of the prediction effect of downhole characteristics provided by an embodiment of this specification;

[0039] Figure 3 It is a schematic diagram of the correlation analysis between surface characteristics and downhole characteristics provided by an embodiment of this specification;

[0040] Figure 4 It is a schematic diagram of the structural composition of an electronic device provided by an embodiment of this specification;

[0041] Figure 5 It is a schematic diagram of the structural composition of a device for predicting downhole parameters with dynamically updated surface parameters provided by an embodiment of this specification;

[0042] Figure 6 It is a schematic diagram of the idea of using surface parameters to update the downhole parameter prediction model in real time provided by an embodiment of this specification;

[0043] Figure 7 It is a schematic diagram of another idea of using surface parameters to update the downhole parameter prediction model in real time provided by an embodiment of this specification;

[0044] Figure 8 It is a schematic diagram of the update of a multi - target network prediction model provided by an embodiment of this specification;

[0045] Figure 9 It is a structural diagram of a long - short - term memory network provided by an embodiment of this specification. Detailed implementation manners

[0046] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of this specification.

[0047] By real - time monitoring and analyzing downhole parameters, the drilling operation can be optimized, the drilling speed and efficiency can be improved, and the drilling cost can be reduced. At the same time, accurate downhole characteristic data helps to formulate more scientific drilling strategies, improve the drilling success rate, and ensure the efficient development and utilization of oil and gas resources. Therefore, downhole characteristic data is not only crucial for ensuring the safety of drilling operations, but also plays an indispensable role in improving drilling efficiency and economic benefits.

[0048] At present, certain progress has been made in the prediction research of downhole characteristics. A large number of scholars have used data-driven models based on drilling big data to predict downhole characteristics, achieving good prediction results. Common parameter prediction methods include time series analysis based on historical data, multivariate regression models using surface parameters, and comprehensive evaluation models combining expert experience, etc. These methods have improved the accuracy of downhole characteristic prediction to a certain extent and provided strong support for drilling operations.

[0049] However, there are still some main problems in the existing downhole characteristic prediction models. First of all, most models are static models and lack adaptability, unable to adjust according to the real-time changing environment. This results in the prediction results of the models in actual applications often lagging behind the actual situation and being difficult to cope with complex and changeable geological conditions. Secondly, the accuracy of the existing models is insufficient and cannot meet the requirements of efficient and safe drilling operations. Especially in complex environments such as deep and ultra-deep wells and deepwater in the ocean, the prediction errors of the models will increase significantly, affecting the safety and efficiency of drilling operations. Therefore, how to realize the dynamic update of the downhole characteristic model with the drilling conditions and wellbore status is the key way to accurately characterize downhole characteristics.

[0050] Aiming at the root causes of the above problems, this specification considers that according to the correlation between surface characteristics and downhole characteristics, the surface characteristics can be quickly and accurately split into the first surface characteristics and the second surface characteristics with a high correlation with downhole characteristics. Secondly, using a multi-objective network prediction model, based on the first surface characteristic data, the predicted values of the second surface characteristics and the downhole characteristic predicted values are determined. On the one hand, the prediction of downhole characteristics can be quickly realized based on surface characteristics in real time and accurately, thereby improving the characterization accuracy of downhole characteristics and helping on-site personnel to understand the downhole situation in real time and accurately. On the other hand, according to the predicted values of the second surface characteristics, the multi-objective network prediction model can be assisted to be optimized, improving the training effect of the model, making the prediction of the model more accurate and real-time. Finally, according to the downhole characteristic predicted values, the drilling strategy for the target well is determined, avoiding the problems of poor determination efficiency and determination accuracy of the drilling strategy caused by the inability to accurately obtain downhole characteristic data, and can optimize drilling operations and improve drilling speed and efficiency.

[0051] Refer to Figure 1 As shown, the embodiment of this specification provides a method for predicting downhole parameters with dynamically updated surface parameters, where this method is specifically applied to the server side. Specifically implemented, this method may include the following content:

[0052] S101: Receive a processing request for the drilling strategy of the target well in the area to be detected;

[0053] S102: In response to the processing request, obtain the target first surface feature data corresponding to the first surface feature in the surface feature data of the target well in the area to be detected;

[0054] S103: Use a multi-target network prediction model to determine the target feature data corresponding to the downhole features based on the target first surface feature data of the target well in the area to be detected; wherein, the multi-target network prediction model is trained based on the second surface feature data corresponding to the second surface feature in the historical sample data, the downhole feature data corresponding to the downhole features in the historical sample data, and the second surface feature prediction value and the downhole feature prediction value. The second surface feature prediction value and the downhole feature prediction value are obtained by the multi-target network prediction model processing the first surface feature data corresponding to the first surface feature in the historical sample data. The first surface feature and the second surface feature are obtained by splitting the surface features according to the correlation degree between the surface features and the downhole features determined from the historical sample data;

[0055] S104: Determine a drilling strategy for the target well according to the target feature data.

[0056] Among them, the above drilling strategy may refer to a series of technical measures and operation plans formulated to achieve safe, economic, and efficient drilling goals during oil and gas exploration and development.

[0057] The above surface feature data may include logging parameters, drilling fluid parameters, and bit parameters and other feature data. Among them, the drilling fluid parameters may include parameters such as density and temperature; the logging parameters may include parameters such as standpipe pressure, flow rate, pump strokes, bit position, and drill pressure; the bit parameters may include parameters such as nozzle size, nozzle combination, and number of cutter wings.

[0058] Downhole features may include downhole equivalent circulating density (Equivalent Circulating Density, ECD), downhole pressure and other features. Among them, the downhole equivalent circulating density can represent the equivalent density at a certain depth position downhole in the case of circulating drilling fluid, usually expressed as the static density of the drilling fluid plus the additional pressure generated by circulation.

[0059] Multiple parameters that affect the liquid column stability, pressure distribution, and circulating pressure loss process of the target well and affect downhole features among the above three parameters (i.e., logging parameters, drilling fluid parameters, and bit parameters) can be selected as the surface features of the target well. For example, the surface features may include measured depth, torsion, bending moment, bit position, well depth, drilling time, hook load, surface drill pressure, surface torque, surface rotation speed, pump strokes, temperature, standpipe pressure, and the surface feature data corresponding to these multiple surface features can be obtained.

[0060] Among them, the first surface feature and the second surface feature can be obtained by splitting the surface feature according to the correlation between the surface feature determined according to the historical sample data and the downhole feature. For example, according to the correlation between the surface feature determined according to the historical sample data and the downhole feature, the riser pressure and the temperature in the above surface features can be used as the second surface feature data, and the remaining multiple parameters (i.e., measured depth, torsion, bending moment, bit position, well depth, drilling time, hook load, surface drilling pressure, surface torque, surface rotation speed, pump strokes) can be used as the first surface parameters.

[0061] Based on the above embodiments, since the multi-objective network prediction model is trained based on the second surface feature data corresponding to the second surface feature in the historical sample data, the downhole feature data corresponding to the downhole feature in the historical sample data, as well as the predicted values of the second surface feature and the predicted values of the downhole feature, and the first surface feature and the second surface feature are obtained by splitting the surface feature according to the correlation between the surface feature determined according to the historical sample data and the downhole feature. Therefore, when the return frequency of the downhole feature data corresponding to the downhole feature in the historical sample data is not high, the second surface feature data corresponding to the second surface feature in the historical sample data can be used to assist in training the multi-objective network prediction model, improving the training effect of the model, making the prediction of the model more accurate and real-time. Furthermore, through the multi-objective network prediction model, the target feature data corresponding to the downhole feature of the target well can be quickly and accurately determined, and according to this target feature data, the drilling strategy for the target well can be determined, avoiding the problems of poor determination efficiency and determination accuracy of the drilling strategy caused by the inability to accurately obtain the downhole feature data, optimizing the drilling operation, and improving the drilling speed and efficiency.

[0062] In some embodiments, before using the multi-objective network prediction model to determine the target feature data corresponding to the downhole feature according to the target first surface feature data of the target well in the area to be detected, when the method is specifically implemented, the following content may further be included:

[0063] S1: Obtain the historical sample data; among them, the historical sample data includes the surface feature data corresponding to the surface feature and the downhole feature data corresponding to the downhole feature;

[0064] S2: Split the surface feature into the first surface feature and the second surface feature according to the correlation between the surface feature and the downhole feature;

[0065] S3: Use the initial multi-objective network prediction model to determine the predicted value of the second surface feature and the predicted value of the downhole feature according to the first surface feature data;

[0066] S4: Determine whether the initial multi-objective network prediction model converges according to the predicted second ground feature value, the second ground feature data, the downhole feature data, and the predicted downhole feature value.

[0067] S5: Obtain the multi-objective network prediction model when the initial multi-objective network prediction model converges.

[0068] In some embodiments, before splitting the ground features into the first ground features and the second ground features according to the correlation between the ground features and the downhole features, specifically in implementation, it may include:

[0069] Obtain initial historical sample data and perform anomaly detection processing on the initial historical sample data.

[0070] When there is abnormal data in the initial historical sample data, perform preprocessing on the initial historical sample data to obtain the historical sample data.

[0071] Among them, when performing anomaly detection processing on the initial historical sample data, the 3σ method (3SigmaRule) can be used to detect outliers based on the standard deviation of the initial historical sample data. Specifically, assuming that the feature data (i.e., ground feature data and downhole feature data) in the initial historical sample data follows a normal distribution, if the mean difference of the feature data exceeds 3 times the standard deviation, it is determined that there is abnormal data in the initial historical sample data.

[0072] Among them, when performing preprocessing on the initial historical sample data, for the continuous feature data in the initial historical sample data, the mean imputation method can be used to fill the missing values with the mean in the feature data. In addition, the local feature data can be polynomially fitted by the smoothing filtering method to estimate the trend of the feature data, so as to achieve a smoothing effect, which can effectively remove noise and retain the trend of the signal.

[0073] Finally, after the above preprocessing operations, the initial multi-objective network prediction model can be trained and tested using the historical sample data. For example, the historical sample data can be split according to a ratio of 70%:30%. Specifically, the first 140,000 samples in the historical sample data can be used as training data, and the last 60,000 samples in the historical sample data can be used as test data to test the performance of the trained multi-objective network prediction model.

[0074] Among them, the above-mentioned initial multi-objective network prediction model can be constructed by using the long short-term memory network algorithm (Long Short Term Memory, LSTM), gated recurrent unit (Gated Recurrent Unit, GRU), or other deep learning algorithms suitable for time series data prediction according to the temporal characteristics of the downhole feature data.

[0075] In some embodiments, to determine whether the initial multi-objective network prediction model converges based on the second surface feature prediction value, the second surface feature data, the downhole feature data, and the downhole feature prediction value, when the method is specifically implemented, the following content may further be included:

[0076] Determine a loss value according to a preset loss function, a loss weight corresponding to the second surface feature, a loss weight corresponding to the downhole feature, the second surface feature prediction value, the second surface feature data, the downhole feature data, and the downhole feature prediction value, and determine whether the initial multi-objective network prediction model converges according to the loss value, where the loss weight corresponding to the downhole feature is greater than the loss weight corresponding to the second surface feature.

[0077] Specifically, the preset loss function can be expressed as a weighted sum of the mean square errors (MSE) of each feature data. Taking the downhole feature as ECD, and the second feature as temperature and riser pressure as examples, the preset loss function can be expressed according to the following formula:

[0078] loss = loss(ECD) + mse(temperature) + mse(riser pressure)

[0079] where loss is the loss value.

[0080] To ensure that the weight distribution among different feature data is the optimal value, different weight ratios can be set for different feature data. Specifically, it can be expressed according to the following formula:

[0081] loss = σ 1 loss(ECD) + σ 2 mse(temperature) + σ 3 mse(riser pressure)

[0082] where σ 1 represents the loss weight corresponding to the downhole feature, σ 2 represents the loss weight corresponding to the second surface feature temperature, and σ 3 represents the loss weight corresponding to the second surface feature riser pressure.

[0083] Furthermore, to ensure that the multi-objective network prediction model can ultimately prioritize the prediction performance of the downhole feature, the loss weight of the downhole feature can be greater than the loss weight corresponding to the second ground feature. In addition, the penalty factor can be set to ensure a balanced weight distribution between the downhole feature and the second ground feature to avoid excessive influence of a certain feature on the model performance.

[0084] Furthermore, the selection of loss weights can be performed by searching for optimal hyperparameters through a Bayesian optimization algorithm. The Bayesian optimization algorithm can construct a Gaussian process model based on known hyperparameter combinations and their corresponding training loss values ​​in each iteration, and select the next set of hyperparameter combinations that may perform better for trial under the guidance of the Gaussian process model. This process will iterate multiple times in the specified hyperparameter search space until a hyperparameter combination that optimizes the model performance is found (i.e., the loss weight corresponding to the downhole feature, and the loss weight corresponding to the second ground feature).

[0085] Based on the above embodiment, after the optimized multi-objective network prediction model is introduced into the second ground feature, the prediction accuracy of the underground feature can be significantly improved. Figure 2 As shown in the figure, the addition of the second surface feature (such as standpipe pressure and temperature) not only effectively improves the prediction accuracy of downhole characteristics (such as ECD), but also can correct the prediction fluctuation of the multi-objective network prediction model, making the model more robust and reliable.

[0086] In some embodiments, the method may further include the following when implemented:

[0087] S1: When a model update cycle is reached, obtaining incremental sample data within the model update cycle;

[0088] S2: Performing model update processing on the multi-objective network prediction model according to the incremental sample data.

[0089] In some embodiments, the model updating process of the multi-objective network prediction model according to the incremental sample data may include:

[0090] The incremental update technology is used to perform model update processing on the multi-objective network prediction model according to the incremental sample data.

[0091] Among them, the above-mentioned incremental update technology can be a data-driven model optimization method. Its core concept is to use the newly input incremental sample data to make small updates to the multi-objective network prediction model without restarting the entire multi-objective network prediction model training.

[0092] Specifically, the above incremental update technology is applied to the dynamic adjustment of the multi-objective network prediction model and the real-time verification of the relationship between downhole features and second surface features. By continuously monitoring the measured feature data of surface features and downhole features, the multi-objective network prediction model can continuously receive the latest feedback of surface feature data. Furthermore, based on the incremental sample data newly added in real time on site, the loss weights of the neuron state, downhole features, and second surface features of the multi-objective network prediction model are timely fine-tuned.

[0093] Further, when the incremental sample data enters the multi-objective network prediction model, the first thing that can be updated is the state of the neurons in the hidden layer of the multi-objective network prediction model. The multi-objective network prediction model can use the incremental sample data to adjust the weights of the neurons, so that the expression of the multi-objective network prediction model in the internal feature space is more in line with the dynamic changes of the current environment. During the incremental update process, the multi-objective network prediction model can only update the feature data that has an impact to reduce the computational overhead, and at the same time, it can also reduce the resource consumption of retraining the entire multi-objective network prediction model.

[0094] In addition, the shared layer and the exclusive layer of each feature in the model can also be synchronously adjusted through the incremental sample data to ensure that after each update, the understanding of the relationship between the downhole features and the second surface features by the multi-objective network prediction model can accurately fit the current downhole environment. Among them, the second surface feature can not only play an auxiliary prediction role, but also provide more accurate data support for the downhole feature through the incremental sample data. Especially in the complex and changeable downhole environment, the effective auxiliary correction of the second surface feature can help the model more sensitively perceive the downhole conditions and improve the adaptability of the multi-objective network prediction model and the accuracy of the prediction results.

[0095] Based on the above method, compared with the traditional batch training method, the incremental update technology can gradually absorb new data features and dynamically optimize the weights of the multi-objective network prediction model, thereby avoiding the sudden decline in the performance of the multi-objective network prediction model or "overfitting". In addition, it can quickly reflect data changes and ensure that the multi-objective network prediction model always has efficient and accurate prediction capabilities.

[0096] In some embodiments, the downhole feature data corresponding to the downhole feature in the incremental sample data is not obtained, and the surface feature data corresponding to the second surface feature has been obtained. When performing model update processing on the multi-objective network prediction model according to the incremental sample data, the specific implementation of the method may further include the following content:

[0097] Update the parameters of the hidden layer in the multi-objective network prediction model according to the surface feature data corresponding to the first surface feature and the second surface feature in the incremental sample data.

[0098] Specifically, when the downhole feature data corresponding to the downhole features and the surface feature data corresponding to the second surface feature in the incremental sample data have been obtained, according to the surface feature data corresponding to the first surface feature and the second surface feature in the incremental sample data, it may further include:

[0099] Dynamically adjust the relationship ratio between the downhole feature and the second surface feature according to the incremental sample data.

[0100] Furthermore, when the data feature change of the second surface feature is greater than a preset threshold (for example, the sudden fluctuation of the temperature of the second surface feature causes the difference between the true value of the temperature at the current moment and the true value of the temperature at the next moment to be greater than the preset threshold), the multi-objective network prediction model will automatically increase the loss weight of the second surface feature related to the temperature to ensure that the downhole feature can be corrected in real time during each update, improving the overall prediction accuracy of the multi-objective network prediction model.

[0101] In some embodiments, when splitting the surface feature into a first surface feature and a second surface feature according to the correlation degree between the surface feature and the downhole feature, when the method is specifically implemented, it may further include the following:

[0102] According to the correlation degree between the surface feature and the downhole feature, determine the surface feature with a correlation degree greater than a preset correlation degree threshold in the surface feature as the second surface feature, and determine the surface feature with a correlation degree not greater than the preset correlation degree threshold in the surface feature as the first surface feature.

[0103] For example, the preset correlation degree threshold may be 0.8. The surface features with an absolute value of the correlation degree between the surface feature and the downhole feature greater than or equal to 0.8 and less than the maximum correlation degree value of 1 (such as temperature and riser pressure) are used as the second surface feature, while the surface features with an absolute value of the correlation degree between the surface feature and the downhole feature less than 0.8 and greater than the minimum correlation degree value of 0 are used as the first surface feature.

[0104] In some embodiments, before splitting the surface feature into a first surface feature and a second surface feature according to the correlation degree between the surface feature and the downhole feature, when the method is specifically implemented, it may further include the following:

[0105] Determine the correlation degree between the surface feature and the downhole feature according to the surface feature data and the downhole feature data according to the following formula:

[0106]

[0107] where r xyis the correlation between the X-th ground feature and the Y-th downhole feature, Cov(X,Y) is the covariance between the X-th ground feature and the Y-th downhole feature, and σ X and σ Y are the standard deviations of the X-th ground feature and the Y-th downhole feature, respectively.

[0108] Among them, the above Pearson correlation coefficient can be a statistical measure that can quantify the strength of the linear relationship between two variables, and its value (i.e., the correlation between the ground feature and the downhole feature) ranges from -1 to 1. Specifically, when it is 1, it indicates a perfect positive linear correlation between the two variables; when the value is -1, it means there is a perfect negative linear correlation between the two variables; and when the value is 0, it means there is no linear correlation between the two variables.

[0109] Further, referring to Figure 3 the shown correlation analysis heat map, the darker colors in the map represent higher absolute values of the Pearson correlation coefficient, indicating a higher correlation between the ground feature and the downhole feature, and there is a significant correlation. During the feature selection process, to maintain the independence between features, those features that show a high correlation with each other are excluded. For example, the correlation coefficient between the inlet density and the outlet density is 1, and the correlation between them is relatively large, so one of them needs to be excluded to avoid the problem of multicollinearity.

[0110] Based on the above embodiments, by evaluating the Pearson correlation coefficient between the ground feature and the downhole feature, features that have a significant impact on the prediction target can be effectively identified, and at the same time, those redundant or low-correlation features can be excluded, thereby optimizing the feature set of the multi-objective network prediction model and improving the performance and interpretability of the model.

[0111] As can be seen from the above, a method for predicting downhole parameters based on dynamically updated surface parameters provided by an embodiment of this specification receives a processing request for a drilling strategy of a target well in a to-be-detected area; in response to the processing request, obtains target first surface feature data corresponding to a first surface feature in the surface feature data of the target well in the to-be-detected area; uses a multi-objective network prediction model to determine target feature data corresponding to downhole features according to the target first surface feature data of the target well in the to-be-detected area; wherein, the multi-objective network prediction model is trained based on second surface feature data corresponding to a second surface feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as a second surface feature prediction value and a downhole feature prediction value, the second surface feature prediction value and the downhole feature prediction value are processed by the multi-objective network prediction model according to first surface feature data corresponding to the first surface feature in the historical sample data, the first surface feature and the second surface feature are obtained by splitting the surface features according to the correlation degree between the surface features and the downhole features determined from the historical sample data; determines a drilling strategy for the target well according to the target feature data. In this way, since the multi-objective network prediction model is trained based on second surface feature data corresponding to a second surface feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as a second surface feature prediction value and a downhole feature prediction value, and the first surface feature and the second surface feature are obtained by splitting the surface features according to the correlation degree between the surface features and the downhole features determined from the historical sample data, therefore, in the case where the frequency of return of downhole feature data corresponding to the downhole features in the historical sample data is not high, the second surface feature data corresponding to the second surface feature in the historical sample data can be used to assist in training the multi-objective network prediction model, improving the training effect of the model, making the prediction of the model more accurate and real-time, and then the target feature data corresponding to the downhole features of the target well can be quickly and accurately determined through the multi-objective network prediction model, so as to determine a drilling strategy for the target well according to the target feature data, avoiding the problems of poor determination efficiency and determination accuracy of the drilling strategy caused by the inability to accurately obtain downhole feature data, and can optimize the drilling operation, improving the drilling speed and efficiency.

[0112] Refer to Figure 4 As shown, an embodiment of this specification also provides a specific electronic device. Among them, the electronic device includes a network communication port 401, a processor 402, and a memory 403. The above structures are connected by internal cables so that each structure can perform specific data interaction.

[0113] Among them, the network communication port 401 can specifically be used to receive a processing request for a drilling strategy of a target well in a to-be-detected area.

[0114] The processor 402 can be specifically configured to, in response to the processing request, obtain target first ground feature data corresponding to a first ground feature from the ground feature data of the target well in the area to be detected; use a multi-target network prediction model to determine target feature data corresponding to downhole features according to the target first ground feature data of the target well in the area to be detected; wherein, the multi-target network prediction model is trained based on second ground feature data corresponding to a second ground feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as a second ground feature prediction value and a downhole feature prediction value, the second ground feature prediction value and the downhole feature prediction value are processed by the multi-target network prediction model according to first ground feature data corresponding to the first ground feature in the historical sample data, and the first ground feature and the second ground feature are obtained by splitting the ground feature according to the correlation degree between the ground feature and the downhole feature determined from the historical sample data; and determine a drilling strategy for the target well according to the target feature data.

[0115] The memory 403 can be specifically configured to store corresponding instruction programs.

[0116] Based on the above method, the related structural performance of the electronic device can be effectively utilized, the data processing speed of the electronic device can be improved, and the method for predicting downhole parameters based on dynamically updated ground parameters can be efficiently implemented.

[0117] In this embodiment, the network communication port 401 can be bound to different communication protocols, so as to send or receive different data. For example, the network communication port can be a port responsible for web data communication, or a port responsible for FTP data communication, or a port responsible for email data communication. In addition, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM, CDMA, etc.; it can also be a Wifi chip; it can also be a Bluetooth chip.

[0118] In this embodiment, the processor 402 can be implemented in any suitable manner. For example, the processor can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, etc. This specification does not make any limitations.

[0119] In this embodiment, the memory 403 may include multiple levels. In a digital system, anything that can store binary data can be a memory; in an integrated circuit, a circuit with a storage function without a physical form is also called a memory, such as RAM, FIFO, etc.; in a system, a storage device with a physical form is also called a memory, such as a memory stick, a TF card, etc.

[0120] The embodiments of this specification also provide a computer-readable storage medium for a method of predicting downhole parameters based on the above-mentioned dynamically updated surface parameters. The computer-readable storage medium stores computer program instructions, which when executed, implement: receiving a processing request for a drilling strategy of a target well in a to-be-detected area; in response to the processing request, obtaining target first surface feature data corresponding to a first surface feature in the surface feature data of the target well in the to-be-detected area; using a multi-target network prediction model, based on the target first surface feature data of the target well in the to-be-detected area, determining target feature data corresponding to downhole features; wherein, the multi-target network prediction model is trained based on second surface feature data corresponding to a second surface feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, as well as a second surface feature prediction value and a downhole feature prediction value, the second surface feature prediction value and the downhole feature prediction value are processed by the multi-target network prediction model according to first surface feature data corresponding to the first surface feature in the historical sample data, and the first surface feature and the second surface feature are obtained by splitting the surface features according to the correlation degree between the surface features and the downhole features determined from the historical sample data; determining a drilling strategy for the target well according to the target feature data.

[0121] In this embodiment, the above storage medium includes but is not limited to a random access memory (RAM), a read-only memory (ROM), a cache, a hard disk drive (HDD), or a memory card. The memory can be used to store computer program instructions. The network communication unit can be set according to the standards specified by the communication protocol and is used for an interface for network connection communication.

[0122] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer-readable storage medium can be explained by comparison with other embodiments and will not be elaborated here.

[0123] Refer to Figure 5, at the software level, the embodiments of this specification also provide a device for predicting downhole parameters with dynamically updated surface parameters. This device may specifically include the following structural modules:

[0124] A request receiving module 501, configured to receive a processing request for the drilling strategy of a target well in a to-be-detected area;

[0125] A request response module 502, configured to, in response to the processing request, obtain target first surface feature data corresponding to a first surface feature in the surface feature data of the target well in the to-be-detected area;

[0126] A model prediction module 503, configured to use a multi-objective network prediction model to determine target feature data corresponding to downhole features according to the target first surface feature data of the target well in the to-be-detected area; wherein, the multi-objective network prediction model is trained based on second surface feature data corresponding to a second surface feature in historical sample data, downhole feature data corresponding to the downhole features in the historical sample data, a second surface feature prediction value, and a downhole feature prediction value. The second surface feature prediction value and the downhole feature prediction value are obtained by the multi-objective network prediction model processing first surface feature data corresponding to the first surface feature in the historical sample data. The first surface feature and the second surface feature are obtained by splitting the surface feature according to the correlation degree between the surface feature and the downhole feature determined from the historical sample data;

[0127] A strategy determination module 504, configured to determine a drilling strategy for the target well according to the target feature data.

[0128] In some embodiments, before the above model prediction module 503, specifically in implementation, obtain the historical sample data; wherein, the historical sample data includes surface feature data corresponding to the surface feature and downhole feature data corresponding to the downhole features; a feature splitting module, configured to split the surface feature into the first surface feature and the second surface feature according to the correlation degree between the surface feature and the downhole feature; use an initial multi-objective network prediction model to determine the second surface feature prediction value and the downhole feature prediction value according to the first surface feature data; a convergence determination module, configured to determine whether the initial multi-objective network prediction model converges according to the second surface feature prediction value, the second surface feature data, the downhole feature data, and the downhole feature prediction value; in the case where the initial multi-objective network prediction model converges, obtain a multi-objective network prediction model; wherein, the multi-objective network prediction model is configured to determine downhole feature data of the target well in the to-be-detected area for formulating a drilling strategy according to the target first surface feature data of the target well in the to-be-detected area.

[0129] In some embodiments, when the above-mentioned convergence determination module is specifically implemented, it determines a loss value according to a preset loss function, a loss weight corresponding to the second ground feature, a loss weight corresponding to the downhole feature, a predicted value of the second ground feature, the second ground feature data, the downhole feature data, and the predicted value of the downhole feature, and determines whether the initial multi-objective network prediction model converges according to the loss value, where the loss weight corresponding to the downhole feature is greater than the loss weight corresponding to the second ground feature.

[0130] In some embodiments, the device further includes obtaining incremental sample data within the model update period when the model update period is reached; a model update module, configured to perform model update processing on the multi-objective network prediction model according to the incremental sample data.

[0131] In some embodiments, the downhole feature data corresponding to the downhole feature in the incremental sample data is not obtained, and the ground feature data corresponding to the second ground feature has been obtained. When the above-mentioned model update module is specifically implemented, it updates the parameters of the hidden layer in the multi-objective network prediction model according to the ground feature data corresponding to the first ground feature and the second ground feature in the incremental sample data.

[0132] In some embodiments, when the above-mentioned feature splitting module is specifically implemented, according to the correlation degree between the ground feature and the downhole feature, the ground feature with a correlation degree greater than a preset correlation degree threshold in the ground feature is determined as the second ground feature, and the ground feature with a correlation degree not greater than the preset correlation degree threshold in the ground feature is determined as the first ground feature.

[0133] In some embodiments, before the above-mentioned feature splitting module, when specifically implemented, according to the ground feature data and the downhole feature data, the correlation degree between the ground feature and the downhole feature is determined according to the following formula:

[0134]

[0135] where r xy is the correlation degree between the Xth ground feature and the Yth downhole feature, Cov(X,Y) is the covariance between the Xth ground feature and the Yth downhole feature, σ X and σ Y are the standard deviations of the Xth ground feature and the Yth downhole feature, respectively.

[0136] It should be noted that the units, devices, modules, etc. illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. For the convenience of description, when describing the above devices, they are divided into various modules according to functions for separate description. Of course, when implementing this specification, the functions of each module can be implemented in the same or multiple software and / or hardware, or the modules implementing the same function can be realized by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of the devices or units can be in electrical, mechanical or other forms.

[0137] As can be seen from the above, based on a device for predicting downhole parameters with dynamically updated surface parameters provided by the embodiments of this specification, first, according to the correlation between surface features and downhole features, the surface features can be quickly and accurately split into the first surface features and the second surface features with a high correlation with the downhole features. Secondly, using a multi-objective network prediction model, based on the first surface feature data, the predicted values of the second surface features and the downhole feature predicted values are determined. On the one hand, the prediction of downhole features can be quickly realized based on surface features in real time and accurately, thereby improving the characterization accuracy of downhole features and helping on-site personnel understand the downhole situation in real time and accurately. On the other hand, according to the predicted values of the second surface features, the multi-objective network prediction model can be assisted in optimization to improve the training effect of the model, making the prediction of the model more accurate and real-time.

[0138] In a specific scenario example, a method and device for predicting downhole parameters with dynamically updated surface parameters provided by this specification can be applied to solve the problem that most of the existing downhole prediction models in the drilling process adopt single-task learning, and the models cannot obtain timely feedback on the changes in downhole data for updating, making it difficult to adapt to the changing downhole complex environment, resulting in low prediction accuracy of downhole feature data. The aim is to ensure the timeliness and accuracy of the downhole feature data prediction model, improve the characterization accuracy of downhole feature data, help on-site personnel understand the downhole situation in real time and accurately, thereby avoiding the problems of poor determination efficiency and determination accuracy of the drilling strategy due to the inability to accurately obtain downhole feature data, and can optimize the drilling operation and improve the drilling speed and efficiency. The specific implementation process can include the following content.

[0139] Refer to Figure 6As shown, in some embodiments, a method flow for using surface parameters (i.e., surface features) to update downhole parameters (i.e., downhole features) in real time is used. Specifically, in combination with prior mechanism knowledge and data correlation analysis, a second surface feature that is highly correlated with the target downhole parameter (i.e., downhole feature) is selected from the surface parameters (i.e., surface features, such as logging parameters, drilling fluid parameters, and drill bit parameters, etc.) as a sub-task (i.e., riser pressure and temperature) for mechanism data driving, and real-time verification (i.e., incremental update technology) is performed to dynamically update the multi-objective network prediction model and predict the downhole parameters (i.e., downhole features).

[0140] Based on the above embodiments, firstly, combining prior mechanism knowledge with data correlation analysis, the ground parameters that are highly correlated with the target downhole parameters are selected as sub-tasks, which helps the model to adjust the weights in real time according to the changes in the ground parameters to make the target result prediction more accurate; then, a multi-objective network prediction model is established, and the difference between the predicted value and the true value of the sub-task is used to update the neural weights in the model's shared hidden layer in real time, so as to achieve high-precision dynamic prediction of downhole parameters. Finally, the incremental update technology is used to update the main-sub-task relationship in the multi-objective network prediction model online based on the returned measured values ​​of the downhole parameters, so as to ensure the real-time perception of the downhole environment by the multi-objective network prediction model. This method can accurately and quickly predict downhole parameters, improve the safety and efficiency of the drilling process, and timely warn of potential risks and prevent downhole accidents. It has broad application prospects.

[0141] See also Figure 7 As shown, in some embodiments, specifically, by performing correlation analysis on the surface parameters (i.e., surface features), according to the correlation between the surface parameters (i.e., surface features) and the downhole parameters (i.e., downhole features), the sub-task (i.e., the second surface feature) is screened out, and then combined with the main task (i.e., downhole features, such as downhole equivalent circulation density ECD), a multi-objective network prediction model is used to update the predicted downhole parameters in real time through the surface parameters.

[0142] Based on the above embodiments, firstly, the method of updating downhole parameters by real-time verification of the multi-objective network prediction model using ground parameters is used, so that the multi-objective network prediction model's ability to instantly perceive the downhole environment is improved. In this case, the prediction results of the downhole target parameters by the multi-objective network prediction model are the results after the weights are updated based on the real-time environment. Secondly, the prediction accuracy of the multi-objective network prediction model is improved: this method takes into account the shared information and common feature space between different parameters, and makes full use of the complementarity between different tasks. At the same time, the use of ground parameters not only effectively improves the prediction accuracy of the multi-objective network prediction model on downhole parameters, but also further ensures the generalizability of the multi-objective network prediction model.

[0143] See also Figure 8As shown, in some embodiments, the multi-objective network prediction model performs supervised learning on the secondary task by obtaining ground parameters in real time. The model continuously updates the prediction results of the secondary task, feeds back the prediction error to the shared neural parameters of the model for real-time correction, so that the model can continuously adjust itself in the change of ground parameters. Through the real-time nature of the ground data, the model can capture the changes in the wellbore environment, and then update the prediction of the downhole parameters of the primary task in real time, maintaining high accuracy and flexibility.

[0144] In some embodiments, the multi-objective network prediction model can be constructed based on the Long Short-Term Memory (LSTM) network. Specifically, the Long Short-Term Memory (LSTM) network is a variant of the recurrent neural network, and its structure is as Figure 9 shown, which is specifically designed to solve the problem of long sequence dependencies. Traditional recurrent neural networks are prone to the problems of gradient vanishing or gradient explosion when processing long sequences, making it difficult to learn long-term dependencies. To address this challenge, the LSTM network introduces a special structure that allows the network to selectively remember or forget information in long sequences. Each unit in the LSTM network includes three modules: the forget gate, the input gate, and the output gate.

[0145] Specifically, when the LSTM network unit operates, the state information h of the previous layer i-1 will first pass through the first forget gate, and the forget gate will control whether to forget the state information h from the previous layer with a certain probability i-1 , and its expression is as follows.

[0146] f t =σ(W f ·[h t-1 ,x t +b f )

[0147] where h i-1 is the output of the previous stage, x i is the input of the current stage, σ is the sigmoid activation function, W f is the weight matrix of the forget gate, and b f is the bias of the forget gate.

[0148] Next, the data comes to the input gate layer. The input gate has the same calculation input as the forget gate. By comprehensively considering the hidden layer state h of the previous stage i-1 and the current input x i , it further determines which information should enter the cell state; for the information found to be worth remembering through the forget gate, this information is recorded and converted into a format that can be used for cell state update.

[0149] i t = σ(W i ·[h t-1 , x t + b i )

[0150]

[0151] where W i is the weight matrix of the input gate, b i is the bias of the input gate, W C is the weight matrix of the candidate memory, b C is the bias of the candidate memory.

[0152] Subsequently, the newly obtained information of this unit is updated into the "backup" of the long short-term memory network, that is, the cell state C t-1 of the previous stage is updated to C t .

[0153]

[0154] Finally, the sigmoid activation function of the output gate is used to determine what result to output according to the current state of the LSTM unit.

[0155] o t = σ(W o *[h t-1 , x t + b o )

[0156] h t = o t *tanh(C t )

[0157] where W o represents the weight matrix of the output gate, b o is the bias of the output gate.

[0158] In some embodiments, by analyzing the relationship between downhole ECD parameters and various engineering parameters, the following two conclusions can be drawn: on the one hand, the downhole ECD parameters increase with the increase of backpressure pump flow rate, drilling fluid density, and drilling fluid viscosity. On the other hand, from the perspective of traditional momentum conservation, there is a partial derivative relationship between downhole ECD and well depth as an input parameter, which can be expressed by a mechanism formula.

[0159] Therefore, under the condition of mechanism constraints, the downhole ECD prediction problem can be regarded as a constrained nonlinear programming problem. Its expression is as follows:

[0160]

[0161] Among them, P pre represents the predicted pressure, and P pre represents the true pressure. n represents the number of samples, and represents the partial derivative of the variable.

[0162] The loss function plays a crucial role in the neural network. It is a key metric for measuring the difference between the predicted values and the true values of the model. During the training process of the neural network, the parameters of the model are adjusted by minimizing the loss function, making the predicted results of the model closer to the true values, thereby improving the accuracy and generalization ability of the model.

[0163] During the training process of the neural network, the value of the loss function is the goal of model optimization. By calculating the gradient of the loss function and using optimization algorithms such as gradient descent to update the parameters of the model, the loss function gradually decreases. When the loss function reaches the minimum value or converges to a small value, the training process of the model ends, and at this time, the model already has good prediction ability.

[0164] In summary, the loss function plays an important role in measuring the performance of the model and guiding the optimization of the model in the neural network. It directly affects the training effect of the model and the final prediction accuracy. Therefore, in the design and training process of the neural network, it is crucial to select an appropriate loss function and reasonably adjust its parameters.

[0165] Therefore, in the process of integrating mechanism knowledge into the neural network, in this example, the mechanism knowledge is embedded into the neural network by changing the form of the loss function. The specific design of the loss function of the multi-objective network prediction model is as follows:

[0166]

[0167] Among them, P pre represents the predicted pressure, and P true represents the true pressure. n represents the number of samples, and λ 1 , λ 2 , λ 3 , λ 4 represent weight parameters, and represents the partial derivative of the variable.

[0168] Although this specification provides method operation steps as described in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way among many execution orders of steps and does not represent the only execution order. When the actual device or client product is executed, it can be executed in the method order shown in the embodiments or the drawings or executed in parallel (such as in a parallel processor or multi-threaded processing environment, or even in a distributed data processing environment). The terms "comprise", "include" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, product or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or also includes elements inherent to such process, method, product or device. Without further limitation, there are no additional identical or equivalent elements excluded in the process, method, product or device comprising the said elements. Words such as first, second, etc. are used to denote names and do not denote any particular order.

[0169] Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, the method steps can be logically programmed to enable the controller to implement the same functions in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Therefore, such a controller can be regarded as a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0170] From the description of the above embodiments, those skilled in the art can clearly understand that this specification can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of this specification can essentially be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, magnetic disk, optical disc, etc., and includes several instructions for causing a computer device (which can be a personal computer, mobile terminal, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this specification.

[0171] Although this specification is depicted through embodiments, those of ordinary skill in the art know that this specification has many variations and changes without departing from the spirit of this specification, and it is hoped that the appended claims will cover these variations and changes without departing from the spirit of this specification.

Claims

1. A method for predicting downhole parameters by dynamically updating surface parameters, characterized in that: include: receiving a processing request for a drilling strategy for a target well in the area to be inspected; In response to the processing request, obtaining target first ground feature data corresponding to the first ground feature in the ground feature data of the target well in the area to be detected; Utilizing a multi-objective network prediction model, target feature data corresponding to the downhole feature is determined according to target first ground feature data of the target well in the area to be detected; wherein the multi-objective network prediction model is obtained by training based on second ground feature data corresponding to the second ground feature in historical sample data, downhole feature data corresponding to the downhole feature in the historical sample data, and second ground feature prediction value and downhole feature prediction value, the second ground feature prediction value and the downhole feature prediction value are obtained by processing the multi-objective network prediction model according to the first ground feature data corresponding to the first ground feature in the historical sample data, the first ground feature and the second ground feature are obtained by splitting the ground feature according to the correlation between the ground feature and the downhole feature determined according to the historical sample data; A drilling strategy for the target well is determined based on the target characteristic data.

2. The method according to claim 1, characterized in that Before determining the target feature data corresponding to the downhole feature according to the target first surface feature data of the target well in the to-be-detected area by using the multi-target network prediction model, the method further includes: Acquire the historical sample data; wherein the historical sample data includes the ground feature data corresponding to the ground feature, and the downhole feature data corresponding to the downhole feature; According to the correlation between the surface feature and the downhole feature, splitting the surface feature into the first surface feature and the second surface feature; Determining the second ground feature prediction value and the downhole feature prediction value according to the first ground feature data using an initial multi-objective network prediction model; Determining whether the initial multi-objective network prediction model converges according to the second ground feature prediction value, the second ground feature data, the downhole feature data, and the downhole feature prediction value; When the initial multi-objective network prediction model converges, a multi-objective network prediction model is obtained.

3. The method according to claim 2, characterized in that The step of determining whether the initial multi-objective network prediction model converges according to the second ground feature prediction value, the second ground feature data, the downhole feature data, and the downhole feature prediction value comprises: A loss value is determined based on a preset loss function, a loss weight corresponding to the second ground feature, a loss weight corresponding to the downhole feature, a predicted value of the second ground feature, the second ground feature data, the downhole feature data and the predicted value of the downhole feature, and based on the loss value, it is determined whether the initial multi-objective network prediction model converges, wherein the loss weight corresponding to the downhole feature is greater than the loss weight corresponding to the second ground feature.

4. The method according to claim 3, characterized in that The method further comprises: When a model update cycle is reached, obtaining incremental sample data within the model update cycle; According to the incremental sample data, the multi-objective network prediction model is updated.

5. The method according to claim 4, characterized in that The downhole feature data corresponding to the downhole feature in the incremental sample data has not been acquired, and the ground feature data corresponding to the second ground feature has been acquired, and the multi-objective network prediction model is updated according to the incremental sample data, including: The parameters of the hidden layer in the multi-objective network prediction model are updated according to the ground feature data corresponding to the first ground feature and the second ground feature in the incremental sample data.

6. The method according to claim 2, characterized in that The step of splitting the surface feature into a first surface feature and a second surface feature according to the correlation between the surface feature and the downhole feature comprises: According to the correlation between the ground features and the downhole features, the ground features whose correlation is greater than a preset correlation threshold are determined as the second ground features, and the ground features whose correlation is not greater than the preset correlation threshold are determined as the first ground features.

7. The method according to claim 6, characterized in that Before splitting the surface feature into the first surface feature and the second surface feature according to the correlation between the surface feature and the downhole feature, the method further includes: According to the surface feature data and the downhole feature data, the correlation between the surface feature and the downhole feature is determined according to the following formula: Among them, r xy is the correlation between the Xth surface feature and the Yth downhole feature, Cov(X,Y) is the covariance between the Xth surface feature and the Yth downhole feature, σ X and σ Y are respectively the standard deviations of the Xth surface feature and the Yth downhole feature.

8. A device for predicting downhole parameters by dynamically updating surface parameters, characterized in that: include: A request receiving module, used for receiving a processing request for a drilling strategy for a target well in the area to be inspected; A request response module, configured to obtain, in response to the processing request, target first ground feature data corresponding to the first ground feature in the ground feature data of the target well in the area to be detected; A model prediction module is used to determine the target feature data corresponding to the downhole feature according to the target first ground feature data of the target well in the area to be detected by using a multi-target network prediction model; wherein the multi-target network prediction model is obtained by training based on the second ground feature data corresponding to the second ground feature in the historical sample data, the downhole feature data corresponding to the downhole feature in the historical sample data, and the second ground feature prediction value and the downhole feature prediction value, the second ground feature prediction value and the downhole feature prediction value are obtained by processing the multi-target network prediction model according to the first ground feature data corresponding to the first ground feature in the historical sample data, the first ground feature and the second ground feature are obtained by splitting the ground feature according to the correlation between the ground feature and the downhole feature determined according to the historical sample data; A strategy determination module is used to determine a drilling strategy for the target well according to the target characteristic data.

9. An electronic device, characterized in that: It comprises a processor and a memory for storing processor executable instructions, and when the processor executes the instructions, the steps of the method for predicting downhole parameters by dynamically updating ground parameters as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: Computer instructions are stored thereon, and when the instructions are executed by a processor, the steps of the method for predicting downhole parameters by dynamically updating ground parameters as described in any one of claims 1 to 7 are implemented.

Images