Downhole parameter prediction method and device

Abstract

The invention discloses an underground parameter prediction method and device. The method comprises the following steps: acquiring an underground parameter prediction model which is constructed by utilizing multi-source data and is used for predicting underground parameters and a loss function of the underground parameter prediction model; constructing a mechanism model for the downhole parameters, and embedding the mechanism model into a loss function of the downhole parameter prediction model to obtain a physical constraint-based loss function; performing back propagation updating on the model parameters through optimization training to obtain a mechanism data fused underground parameter intelligent prediction model; and using the underground parameter intelligent prediction model to predict underground parameters to obtain a prediction result. By using the scheme of the invention, the dependence of the intelligent model on data driving can be reduced, and the stability and accuracy of a prediction result can be improved.

Description

Technical Field

[0001] The present invention relates to the field of exploration technology, and in particular to a method and device for predicting downhole parameters. Background Art

[0002] With the continuous deepening of oil and gas exploration and development in my country, the focus and types of oil and gas exploration and development in my country are gradually extending to unconventional oil and gas resources such as deep, ultra-deep and low permeability. Due to the high temperature and pressure in deep formations, the safety density window is extremely narrow, and complex situations such as overflow and gas invasion are very likely to occur. The accurate prediction and real-time monitoring of downhole engineering parameters are the key links to ensure the safety of deep high-temperature and high-pressure drilling. The existing mechanism model for downhole parameter prediction has high model complexity, long calculation time and insufficient accuracy under complex geological and engineering conditions; in addition, under high-temperature and high-pressure working environments, the durability and reliability of the high-cost downhole measurement device have become a problem.

[0003] In recent years, the advantages of artificial intelligence algorithms and technologies in the oil and gas field have gradually emerged, and have been applied to multiple scenarios such as working condition diagnosis and parameter optimization. However, the current intelligent model belongs to a data-driven modeling method. The establishment of the model is extremely dependent on the quality and quantity of sample data. It is necessary to obtain a large amount of high-quality and representative data and conduct multiple training to obtain an accurate model. Due to the limited sample data in the drilling process and the large differences in data from different drilling processes, the existing data-driven intelligent model has poor stability in predicting downhole parameters, and its application in the drilling process also faces problems of real-time and generalization. Summary of the invention

[0004] The present invention provides a method and device for predicting downhole parameters, so as to reduce the dependence of intelligent models on data driving and improve the stability and accuracy of prediction results.

[0005] To this end, the present invention provides the following technical solutions:

[0006] A method for predicting downhole parameters, the method comprising:

[0007] Obtaining a downhole parameter prediction model and its loss function constructed by using multi-source data for predicting downhole parameters;

[0008] Constructing a mechanism model for the downhole parameters, embedding the mechanism model into the loss function of the downhole parameter prediction model, and obtaining a loss function based on physical constraints;

[0009] The model parameters are back-propagated and updated through optimization training to obtain an intelligent prediction model of downhole parameters with mechanism data fusion;

[0010] The downhole parameter intelligent prediction model is used to predict the downhole parameters to obtain prediction results.

[0011] Optionally, the method further comprises constructing the downhole parameter prediction model in the following manner:

[0012] Acquire multi-source data and generate a multi-source data set; the multi-source data includes any one or more of the following data: logging data, well logging data, and drilling data;

[0013] Preprocessing the data in the multi-source data set to obtain a downhole key parameter feature set;

[0014] The downhole key parameter feature set is used to train a downhole parameter prediction model and its loss function.

[0015] Optionally, the logging data includes any one or more of the following data: well depth, drilling pressure, rotary table torque, rotation speed, mechanical drilling speed, pump pressure, total pool volume, inlet density, outlet density, hook load, pump stroke, equivalent density;

[0016] The logging data includes any one or more of the following data: natural gamma, natural potential, acoustic time difference, deep lateral resistivity curve, shallow lateral resistivity, density or neutron;

[0017] The drilling data includes any one or more of the following data: natural gamma, natural potential, well diameter, apparent resistivity, deep lateral resistivity curve, shallow lateral resistivity curve, acoustic wave, density, neutron, etc.

[0018] Optionally, the preprocessing of the data to obtain a downhole key parameter feature set includes:

[0019] Use any one or more methods of outlier removal, linear regression interpolation, and sliding average filtering to screen out abnormal and noisy data in the well;

[0020] Select feature parameters using variance selection method;

[0021] Calculating correlation coefficients between different feature parameters, and determining candidate feature parameters according to the correlation coefficients;

[0022] Determine effective characteristic parameters among the candidate characteristic parameters by using non-parametric methods in statistics;

[0023] Perform one-hot encoding on the non-time series feature parameters to obtain the encoded non-time series feature parameters;

[0024] The encoded non-time-series characteristic parameters are fused with the effective characteristic parameters to generate a downhole key parameter characteristic set.

[0025] Optionally, the training of the downhole key parameter feature set to obtain a downhole parameter prediction model includes:

[0026] Using the downhole key parameter feature set to train a candidate downhole parameter prediction model;

[0027] A model parameter optimization test is performed on the candidate downhole parameter model to determine the downhole parameter prediction model.

[0028] Optionally, the performing a model parameter optimization test on the candidate downhole parameter prediction model includes:

[0029] Select any one or more of the following as evaluation indicators of model prediction effect: root mean square error RMSE, average relative error δ, maximum relative error δ max , training time.

[0030] Optionally, the mechanism model includes any one or more of the following: a direct and inverse proportional response relationship model, a tubing force balance model, and a multiphase flow model.

[0031] Optionally, embedding the mechanism model into the loss function of the downhole parameter prediction model to obtain a loss function based on physical constraints includes:

[0032] The input parameters of the mechanism model are expressed as physical functions of downhole parameters to obtain a loss function based on physical constraints.

[0033] Optionally, the downhole parameters include any one or more of the following: bottom hole torque, bit pressure, equivalent density.

[0034] A downhole parameter prediction device, comprising:

[0035] A model determination module, used to obtain a downhole parameter prediction model and its loss function constructed by using multi-source data for predicting downhole parameters;

[0036] A correction module, used for constructing a mechanism model for the downhole parameters, embedding the mechanism model into the loss function of the downhole parameter prediction model, and obtaining a loss function based on physical constraints;

[0037] Model optimization module, used to back-propagate and update model parameters through optimization training to obtain an intelligent prediction model of downhole parameters with mechanism data fusion;

[0038] The prediction module is used to predict the downhole parameters using the downhole parameter intelligent prediction model to obtain prediction results.

[0039] Optionally, the device further comprises: a model building module, used to build the downhole parameter prediction model; the model building module comprises:

[0040] A data acquisition unit, used to acquire multi-source data and generate a multi-source data set; the multi-source data includes any one or more of the following data: logging data, well logging data, drilling data;

[0041] A data processing unit, used for preprocessing the data in the multi-source data set to obtain a downhole key parameter feature set;

[0042] A training unit is used to train the downhole key parameter feature set to obtain a downhole parameter prediction model and its loss function.

[0043] A computer-readable storage medium stores a computer program, which executes the steps of the above-mentioned downhole parameter prediction method when executed by a processor.

[0044] The downhole parameter prediction method and device provided by the present invention pre-build a downhole parameter prediction model to predict downhole parameters, and by adding mechanism constraints, deeply fuse data and mechanism, train to obtain a downhole parameter intelligent prediction model with mechanism data fusion, and use the model to realize downhole parameter prediction under mechanism constraints, thereby effectively improving the accuracy, stability and generalization of the model, and improving the stability and accuracy of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a flow chart of constructing a downhole parameter prediction model in the implementation of the present invention;

[0046] Figure 2 is a histogram of distance correlation coefficients of 14 time series feature parameters in an embodiment of the present invention;

[0047] Figure 3 It is a structural schematic diagram of a downhole parameter prediction model in an embodiment of the present invention;

[0048] Figure 4 is a flow chart of a downhole parameter prediction method provided by an embodiment of the present invention;

[0049] Figure 5 Schematic diagram of the training process of the intelligent prediction model of downhole engineering parameters with mechanism data fusion in an embodiment of the present invention;

[0050] Figure 6 It is a structural schematic diagram of a downhole parameter prediction device provided by an embodiment of the present invention;

[0051] Figure 7 It is a structural diagram of a model building module in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0053] The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments cannot be described one by one here, but the embodiments of the present invention are not therefore limited to the following embodiments.

[0054] In order to solve the problem that the existing intelligent models are overly dependent on data-driven, resulting in poor stability of prediction results in applications during drilling, an embodiment of the present invention provides a downhole parameter prediction method and device, which predicts downhole parameters based on a pre-constructed downhole parameter prediction model, and then constructs a loss function based on the downhole prediction parameters and the corresponding downhole measured parameters, and embeds physical constraints in the loss function, and obtains a downhole parameter intelligent prediction model with mechanism data fusion through optimization training, and uses the downhole parameter intelligent prediction model with mechanism data fusion to predict downhole parameters and obtain prediction results.

[0055] It should be noted that the downhole parameter prediction model may be an existing model or a model constructed by collecting corresponding data.

[0056] First, a process of constructing a downhole parameter prediction model provided in an embodiment of the present invention is described in detail below.

[0057] like Figure 1 FIG. 1 is a flowchart of constructing a downhole parameter prediction model in the implementation of the present invention, which includes the following steps:

[0058] Step 101, acquiring multi-source data and generating a multi-source data set.

[0059] The multi-source data may come from multiple wells, and may include but is not limited to drilling data, well logging data, mud logging data, etc. Among them:

[0060] The logging data mainly include: well inclination data table, engineering parameter table, drilling fluid performance table, drill bit record table, cuttings description record table, etc., wherein the data include but are not limited to any one or more of the following data: well depth, drilling pressure, rotary table torque, rotation speed, mechanical drilling speed, pump pressure, total pool volume, inlet density, outlet density, hook load, pump stroke, equivalent density, etc.

[0061] The logging data include but are not limited to any one or more of the following data: natural gamma, natural potential, well diameter, apparent resistivity, deep lateral resistivity curve, shallow lateral resistivity curve, acoustic wave, density, neutron, etc.

[0062] The drilling data include but are not limited to any one or more of the following data: well deviation data, drilling fluid performance data, drill bit record data, cuttings description record data, etc. Among them:

[0063] The drilling fluid performance data include but are not limited to any one or more of the following data: funnel viscosity (unit s), plastic viscosity (unit mPa·s), dynamic shear force (unit Pa), drilling fluid system, etc.;

[0064] The drill bit record data includes but is not limited to any one or more of the following data: drill bit size, drill bit model, freshness at entry into the well (unit %), freshness at exit from the well (unit %), drilling pressure (unit kN), rotation speed (unit r / min), drill bit water hole, drilling tool assembly information, etc.;

[0065] The rock cuttings description record data includes but is not limited to any one or more of the following data: information such as layer position, well section, lithology and oil, gas and water description, and the lithology and oil, gas and water description information includes numerical and text information such as rock cuttings shape, maximum particle size, minimum particle size, average particle size, sorting, roundness, etc.

[0066] The multi-source data may include structured data and unstructured data; moreover, the data may be of numerical type or text type, which is not limited in the embodiment of the present invention.

[0067] Therefore, it is also necessary to pre-process these data, which may include but is not limited to unifying the sampling frequency, format conversion and other processing.

[0068] For example, the sampling frequency of the well deviation data, drilling fluid performance data, drill bit record data, and cuttings description record data is converted into sampling every 1m. For example, the well deviation data can be interpolated with the well trajectory data using the cubic spline curve interpolation method, and converted into sampling one point every 1m.

[0069] After unifying the sampling frequencies of various types of data, due to the inconsistency of different data types, it is necessary to establish corresponding data extraction and processing methods for specific data types.

[0070] For example, well deviation data and engineering parameters are all numerical data. After the sampling frequency is unified to 1m, they can be directly read using the Pandas module in Python, which facilitates subsequent data splicing operations.

[0071] For another example, if the drilling fluid system data is in text type, the drilling fluid system can be roughly classified into three major systems: water-based, oil-based, and gas-based. Among them, the water-based includes non-dispersible, dispersed, polymer, low solid phase, calcium-treated, and saturated brine drilling fluids; the gas-based includes air and foam drilling fluids. For different categories, they can be mapped to corresponding values, that is, mapping text-based data to numerical data.

[0072] For another example, drill bit model data is stored in the drill tool assembly information in the form of text, and information is extracted according to the actual drill bit type, such as a PDC drill bit.

[0073] After preprocessing the above multi-source data, a multi-source data set can be obtained. For example, the well inclination, engineering parameters, drilling fluid properties, drill bit records, cuttings description records and other data can be spliced ​​according to the well depth as the index to form a comprehensive logging parameter data set.

[0074] Step 102, preprocessing the data in the multi-source data set to obtain a downhole key parameter feature set.

[0075] The preprocessing of the data mainly includes:

[0076] (1) The data is processed by outlier elimination, linear regression interpolation, sliding average filtering, etc. to screen out abnormal data and noise data in the well, reduce the impact of the well noise on the collected data, and improve the density and quality of the data samples.

[0077] (2) Use variance selection method to select feature parameters.

[0078] The smaller the variance is, the less feature information there is. To this end, a variance threshold may be set to calculate the variance of each feature parameter, and feature parameters that are smaller than the variance threshold may be deleted.

[0079] (3) Calculate the correlation coefficients between different feature parameters, and determine candidate feature parameters based on the correlation coefficients.

[0080] For example, we can combine the three correlation coefficients, namely the Pearson correlation coefficient, the Spearman correlation coefficient, and the distance correlation coefficient, to analyze the linear relationship between features, and select features based on the theoretical connection between features and a correlation coefficient of 0.6 as the threshold.

[0081] In a non-limiting embodiment, the distance correlation coefficient between different feature parameters may be calculated using the following calculation formula:

[0082]

[0083] in, is the distance covariance of X and Y; is the distance variance of X; is the distance variance of Y; R n (X,Y) is the distance correlation coefficient of X and Y.

[0084] Of course, other methods may also be used to calculate the correlation between different characteristic parameters, which is not limited in this embodiment of the present invention.

[0085] If the correlation coefficient of two characteristic parameters is less than a set correlation coefficient threshold (for example, set to 0.6), it is considered that the two characteristic parameters are unrelated.

[0086] The correlation coefficients between all different feature parameters are calculated in turn, and then the feature parameters with higher correlation can be selected as candidate feature parameters, and irrelevant feature parameters can be deleted.

[0087] Figure 2 A histogram of distance correlation coefficients of 14 time series feature parameters in an embodiment of the present invention is shown.

[0088] (4) Using non-parametric methods in statistics, effective feature parameters among the candidate feature parameters are determined, so as to further screen the candidate feature parameters.

[0089] By using non-parametric methods in statistics, using chi-square test and mutual information method, the independence of the target feature and other features is calculated, so as to determine each effective feature in the candidate feature parameters. Among them, the mutual information method is a filtering feature selection algorithm. In feature selection, mutual information is a measure of the degree of mutual dependence between features and label classes. The larger the mutual information value, the greater the degree of mutual dependence between features and label classes.

[0090] (5) Normalize the effective feature parameters.

[0091] Considering that the scale differences of various features of the original data may be large, which may easily cause gradient problems, the maximum and minimum values ​​of the selected effective feature parameters can be further normalized to eliminate the influence of the dimension, avoid gradient problems, and speed up the model training.

[0092] The specific formula for normalization is as follows:

[0093]

[0094] Among them, x′ i is the result of normalization of the i-th variable; x i is the i-th variable; min(x i ) is the minimum value of the i-th variable; max(x i ) is the maximum value of the i-th variable.

[0095] It should be noted that the effective feature parameters screened out by (2)-(4) above are time series data feature parameters.

[0096] (6) Perform one-hot encoding on the non-time series data feature parameters to obtain encoded non-time series feature parameters.

[0097] For example, the drilling tool assembly, drilling fluid type, drill bit type and other non-time series data are encoded with one-hot encoding. For example, according to the classification of drilling fluid system into water-based, oil-based, and gas-based drilling fluids, they can be encoded as [0,0,1], [0,1,0], [1,0,0] respectively; according to the classification of drill bit type into PDC and roller, they can be encoded as [0,1], [1,0] respectively; according to the classification of drilling tool assembly into pendulum, full-eye, and tower drilling tool assembly, they can be encoded as [0,0,1], [0,1,0], [1,0,0] respectively.

[0098] (7) The encoded non-time series feature parameters are integrated with the normalized effective feature parameters to generate a downhole key parameter feature set.

[0099] For example, in a non-limiting embodiment, the encoded non-time-series characteristic parameters may be combined with normalized effective characteristic parameters to generate a downhole key parameter characteristic set.

[0100] Step 103, using the downhole key parameter feature set to train and obtain a downhole parameter prediction model and its loss function.

[0101] In an embodiment of the present invention, the downhole parameter prediction model may adopt but is not limited to any one of the following: a machine learning model, a deep learning model, a plurality of deep learning integrated models, etc.

[0102] Furthermore, a candidate downhole parameter prediction model may be first trained based on the above-mentioned downhole key parameter feature set, and then the candidate downhole parameter model may be subjected to a model parameter optimization test using a corresponding test set to determine a final downhole parameter prediction model.

[0103] In specific applications, you can select but not limited to any one or more of the following: root mean square error RMSE, average relative error δ, maximum relative error δ max , training time, etc., as evaluation indicators of the model prediction effect, so as to determine the final downhole parameter prediction model according to the corresponding evaluation indicators.

[0104] The calculation formulas for the root mean square error RMSE, the average relative error δ, and the maximum relative error δmax are as follows:

[0105]

[0106]

[0107] δ max=max{δ1,δ2,δ3,……δ N}

[0108] Among them, y itrue is the target true value of the i-th data; y ipre is the target prediction value of the i-th data; N is the total number of samples; δ i is the relative error of the ith value.

[0109] For example, in a specific application, 74 well data from a certain block were used to compare and analyze the three models of BP (Back Propagation) neural network, LSTM (Long Short-Term Memory), and BP-LSTM. The number of neural network layers, number of neurons, discard rate, and activation function were optimized. The experimental parameters were designed using the orthogonal test method. The network model was trained using a training set of 52 wells and a test set of 22 wells. Each model was trained 120 times. After model training and testing, each neural network can obtain 9 models for predicting bottom hole drilling pressure and 9 models for predicting bottom hole torque. The bottom hole drilling pressure and bottom hole torque calculation models were optimized based on the three indicators of relative error, root mean square error, and model complexity. The orthogonal experimental scheme is shown in Table 2. Finally, the BP-LSTM network was selected as the final intelligent prediction basic model.

[0110] Figure 3 A structural schematic diagram of a downhole parameter prediction model in an embodiment of the present invention is shown.

[0111] The downhole parameter prediction model uses a BP-LSTM network, in which the BP neural network is used to input non-time series data, for example, three non-time series variables: drilling tool assembly, drilling fluid type, and drill bit type. After unique hot encoding, there are 8 types. An input sample of the BP network is a one-dimensional vector, which can be expressed as:

[0112]

[0113] The LSTM network is used to input time series data. The historical time step is selected as 5. For example, a data sample input to the LSTM network is a 5×14 matrix, which can be expressed as:

[0114]

[0115] in, represents the i-th sample at the j-th time, and t is the current time.

[0116] The downhole parameter prediction method provided in an embodiment of the present invention utilizes the above-mentioned downhole parameter prediction model to predict the downhole parameters, and deeply integrates the data and the mechanism by adding mechanism constraints and other methods, trains the downhole parameter intelligent prediction model of the mechanism data fusion, and realizes the downhole parameter prediction under the mechanism constraints, which can effectively improve the accuracy, stability and generalization of the model.

[0117] like Figure 4 FIG. 1 is a flow chart of a downhole parameter prediction method provided by an embodiment of the present invention, comprising the following steps:

[0118] Step 401, obtaining a downhole parameter prediction model and its loss function for predicting downhole parameters constructed using multi-source data.

[0119] Step 402: construct a mechanism model for the downhole parameters, embed the mechanism model into the loss function of the downhole parameter prediction model, and obtain a loss function based on physical constraints.

[0120] The mechanism model is used as a physical constraint to ensure that the prediction results conform to physical laws, avoid the prediction results from violating actual engineering conditions, and make the predicted downhole parameters closer to the actual downhole engineering parameters. For example, physical knowledge constraints such as hydraulics and tubing mechanics can be embedded, and the established mechanism models include but are not limited to: positive and negative proportional response relationship models, tubing force balance models, multiphase flow models, etc. These models can be used for the mechanism calculation of downhole parameters such as bottom hole torque, drilling pressure, and equivalent density.

[0121] The input parameters of the mechanism model are expressed as physical functions of the downhole parameters to obtain a loss function based on physical constraints.

[0122] The physical constraints and their neural networks can be expressed as:

[0123]

[0124] Among them, w [n] represents the model weight matrix of the nth layer; b represents the model bias value; x represents the constrained function variable, such as well depth, drilling fluid density, bottom hole torque, etc.; T represents the downhole parameters to be predicted, such as bottom hole drilling pressure, torque, etc.

[0125] Furthermore, considering that the objective function of a constrained neural network is relatively complex, the penalty function method can be used to construct the objective function and the constraint conditions as auxiliary functions, thereby converting the constrained nonlinear programming problem into an unconstrained nonlinear programming problem. At the same time, the prediction accuracy and the fitting of the physical constraints can be balanced by reasonably setting the weight of the physical constraints in the loss function. For example, different penalty factors λ can be set and solved to make the model have different responsiveness when facing different degrees of constraints.

[0126]

[0127] Among them, F is the objective function to be optimized, T pre is the model prediction value, T true is the true value of the parameter, and variables such as x1 and x2 represent the network input parameters, such as well depth, drilling fluid density, bottom hole torque, etc.

[0128] Step 403, back-propagation update of model parameters is performed through optimization training to obtain a downhole parameter intelligent prediction model fused with mechanism data.

[0129] In the embodiment of the present invention, the model parameters can be back-propagated and updated by the optimizer (i.e., the correction coefficient balances the error size). Specifically, the predicted downhole prediction parameters are compared with the actual downhole measured parameters to analyze the prediction error. According to the error size, the prediction results are corrected using the correction coefficient to further improve the accuracy and reliability of the prediction. Through multiple corrections and adjustments, the prediction results are continuously optimized. When the set error threshold is met, the training is stopped to obtain the intelligent prediction model of downhole engineering parameters fused with mechanism data.

[0130] Figure 5 It is a schematic diagram of the training process of the intelligent prediction model of downhole engineering parameters fused with the above mechanism data.

[0131] Step 404: Use the downhole parameter intelligent prediction model to predict downhole engineering parameters to obtain prediction results.

[0132] Accordingly, the present invention also provides a downhole parameter prediction device, such as Figure 6 FIG. 1 is a schematic diagram of the structure of a downhole parameter prediction device provided by an embodiment of the present invention.

[0133] The downhole parameter prediction device of this embodiment includes the following modules:

[0134] A model determination module 601 is used to obtain a downhole parameter prediction model and its loss function constructed using multi-source data for predicting downhole parameters;

[0135] A correction module 602 is used to construct a mechanism model for the downhole parameters, embed the mechanism model into the loss function of the downhole parameter prediction model, and obtain a loss function based on physical constraints;

[0136] The model optimization module 603 is used to perform back propagation update on the model parameters through optimization training to obtain a downhole parameter intelligent prediction model fused with mechanism data;

[0137] The prediction module 604 is used to predict the downhole parameters using the downhole parameter intelligent prediction model to obtain prediction results.

[0138] It should be noted that the downhole parameter prediction model may be an existing model or a model constructed by collecting corresponding data.

[0139] Accordingly, the downhole parameter prediction model may be constructed by the model construction module using a data-driven modeling method.

[0140] The model building module may be a part of the downhole parameter prediction device of the present invention, or may be independent of the device, which is not limited in the embodiment of the present invention.

[0141] like Figure 7 FIG. 1 is a schematic diagram of a structure of a model building module in an embodiment of the present invention.

[0142] The model building module includes the following units:

[0143] The data acquisition unit 701 is used to acquire multi-source data and generate a multi-source data set; the multi-source data includes any one or more of the following data: logging data, well logging data, and drilling data;

[0144] A data processing unit 702 is used to pre-process the data in the multi-source data set to obtain a downhole key parameter feature set;

[0145] The training unit 703 is used to train the downhole key parameter feature set to obtain a downhole parameter prediction model and its loss function.

[0146] The structure of the above-mentioned downhole parameter prediction model, as well as more functions and implementation methods of each module and unit in the downhole parameter prediction device of the present invention can be referred to the description in the previous embodiment of the method of the present invention, and will not be repeated here.

[0147] The downhole parameter prediction method and device provided by the present invention pre-build a downhole parameter prediction model to predict downhole parameters, and by adding mechanism constraints, deeply fuse data and mechanism, train to obtain a downhole parameter intelligent prediction model with mechanism data fusion, and use the model to realize downhole parameter prediction under mechanism constraints, thereby effectively improving the accuracy, stability and generalization of the model, and improving the stability and accuracy of the prediction results.

[0148] The solution of the present invention can effectively solve the problems existing in the existing downhole drilling parameter prediction, is applicable to different types of drilling operations and geological conditions, has a certain degree of portability, and provides a reliable and intelligent downhole engineering parameter prediction technology for the drilling industry, which has high practical value and broad application prospects.

[0149] The “plurality” mentioned in the embodiments of the present invention refers to two or more than two.

[0150] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or apparatus.

[0151] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. Moreover, the system embodiments described above are merely schematic, in which the modules and units described as separate components may or may not be physically separated, that is, they may be located on one network unit, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0152] In a specific implementation, each module / unit included in each device or product described in the above embodiments may be a software module / unit or a hardware module / unit, or may be partly a software module / unit and partly a hardware module / unit.

[0153] For example, for each device or product applied to or integrated in a chip, each module / unit contained therein may be implemented in the form of hardware such as circuits, or at least some of the modules / units may be implemented in the form of software programs, which run on a processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits; for each device or product applied to or integrated in a chip module, each module / unit contained therein may be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (such as a chip, circuit module, etc.) or different components of the chip module, or at least some of the modules / units may be implemented in the form of software programs. The element can be implemented in the form of a software program, which runs on a processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated in the terminal, the various modules / units contained therein can be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or in different components in the terminal, or, at least some modules / units can be implemented in the form of a software program, which runs on a processor integrated inside the terminal, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits.

[0154] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be physically arranged separately, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0155] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform some steps of the method described in each embodiment of the present application.

[0156] The embodiments of the present invention are described in detail above. The present invention is described in detail using specific implementation methods herein. The description of the above embodiments is only used to help understand the method and system of the present invention. It is only a part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should belong to the scope of protection of the present invention, and the content of this specification should not be understood as limiting the present invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A downhole parameter prediction method, characterized in that: The method comprises: Obtaining a downhole parameter prediction model and its loss function constructed by using multi-source data for predicting downhole parameters; Constructing a mechanism model for the downhole parameters, embedding the mechanism model into the loss function of the downhole parameter prediction model, and obtaining a loss function based on physical constraints; The model parameters are back-propagated and updated through optimization training to obtain an intelligent prediction model of downhole parameters with mechanism data fusion; The downhole parameter intelligent prediction model is used to predict the downhole parameters to obtain prediction results.

2. The downhole parameter prediction method according to claim 1, characterized in that: The method further comprises constructing the downhole parameter prediction model in the following manner: Acquire multi-source data and generate a multi-source data set; the multi-source data includes any one or more of the following data: logging data, well logging data, and drilling data; Preprocessing the data in the multi-source data set to obtain a downhole key parameter feature set; The downhole key parameter feature set is used to train a downhole parameter prediction model and its loss function.

3. The downhole parameter prediction method according to claim 2, characterized in that: The logging data includes any one or more of the following data: well depth, drilling pressure, rotary table torque, rotation speed, mechanical drilling speed, pump pressure, total pool volume, inlet density, outlet density, hook load, pump stroke, equivalent density; The logging data includes any one or more of the following data: natural gamma, natural potential, acoustic time difference, deep lateral resistivity curve, shallow lateral resistivity, density or neutron; The drilling data includes any one or more of the following data: natural gamma, natural potential, well diameter, apparent resistivity, deep lateral resistivity curve, shallow lateral resistivity curve, acoustic wave, density, neutron, etc.

4. The downhole parameter prediction method according to claim 2, characterized in that: The data is preprocessed to obtain a downhole key parameter feature set including: Use any one or more methods of outlier removal, linear regression interpolation, and sliding average filtering to screen out abnormal and noisy data in the well; Select feature parameters using variance selection method; Calculating correlation coefficients between different feature parameters, and determining candidate feature parameters according to the correlation coefficients; Determine effective characteristic parameters among the candidate characteristic parameters by using non-parametric methods in statistics; Perform one-hot encoding on the non-time series feature parameters to obtain the encoded non-time series feature parameters; The encoded non-time-series characteristic parameters are fused with the effective characteristic parameters to generate a downhole key parameter characteristic set.

5. The downhole parameter prediction method according to claim 2, characterized in that: The method of using the downhole key parameter feature set to train a downhole parameter prediction model comprises: Using the downhole key parameter feature set to train a candidate downhole parameter prediction model; A model parameter optimization test is performed on the candidate downhole parameter model to determine the downhole parameter prediction model.

6. The downhole parameter prediction method according to claim 5, characterized in that: The performing of a model parameter optimization test on the candidate downhole parameter prediction model comprises: Select any one or more of the following as evaluation indicators of model prediction effect: root mean square error RMSE, average relative error δ, maximum relative error δ max , training time.

7. The downhole parameter prediction method according to claim 1, characterized in that: The mechanism model includes any one or more of the following: a positive and negative proportional response relationship model, a pipe string force balance model, and a multiphase flow model.

8. The downhole parameter prediction method according to claim 1, characterized in that: The step of embedding the mechanism model into the loss function of the downhole parameter prediction model to obtain a loss function based on physical constraints includes: The input parameters of the mechanism model are expressed as physical functions of downhole parameters to obtain a loss function based on physical constraints.

9. The downhole parameter prediction method according to any one of claims 1 to 8, characterized in that: The downhole parameters include any one or more of the following: bottom hole torque, bit pressure, and equivalent density.

10. A downhole parameter prediction device, characterized in that: The device comprises: A model determination module, used to obtain a downhole parameter prediction model and its loss function constructed by using multi-source data for predicting downhole parameters; A correction module, used for constructing a mechanism model for the downhole parameters, embedding the mechanism model into the loss function of the downhole parameter prediction model, and obtaining a loss function based on physical constraints; Model optimization module, used to back-propagate and update model parameters through optimization training to obtain an intelligent prediction model of downhole parameters with mechanism data fusion; The prediction module is used to predict the downhole parameters using the downhole parameter intelligent prediction model to obtain prediction results.

11. The downhole parameter prediction device according to claim 10, characterized in that: The device further comprises: a model building module for building the downhole parameter prediction model; the model building module comprises: A data acquisition unit, used to acquire multi-source data and generate a multi-source data set; the multi-source data includes any one or more of the following data: logging data, well logging data, drilling data; A data processing unit, used for preprocessing the data in the multi-source data set to obtain a downhole key parameter feature set; A training unit is used to train the downhole key parameter feature set to obtain a downhole parameter prediction model and its loss function.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the downhole parameter prediction method according to any one of claims 1 to 9 are executed.

Images