Mechanical drilling rate prediction and drilling parameter optimization method, system, medium and equipment based on cloud edge collaboration architecture

Abstract

The present application relates to the field of oil and gas exploration, and discloses a mechanical drilling speed prediction and drilling parameter optimization method, system, medium and equipment based on a cloud-edge collaborative architecture, comprising: obtaining drilling data, training a neural base model using the drilling data on a cloud server, obtaining a pre-trained mechanical drilling speed prediction model and issuing it to an edge server; freezing the pre-trained mechanical drilling speed prediction model on the edge server, training a fine-tuned neural base model using real-time drilling data of the current well to learn the residual error between the predicted value of the pre-trained mechanical drilling speed prediction model and the measured value of the mechanical drilling speed; superimposing the predicted output of the pre-trained mechanical drilling speed prediction model and the residual correction value output by the fine-tuned neural base model to obtain the final mechanical drilling speed prediction value; extracting the nonlinear shape function inside the pre-trained mechanical drilling speed prediction model to generate a contribution curve of each drilling parameter to the mechanical drilling speed prediction value, determine the optimization interval of the drilling parameter, and output the drilling parameter optimization suggestion.

Description

Technical Field

[0001] This invention relates to the field of oil and gas exploration technology, and in particular to a method, system, medium, and equipment for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture. Background Technology

[0002] Accurate prediction of mechanical drilling rate is crucial for real-time optimization of drilling parameters, shortening well construction cycles, and reducing exploration and development costs. With the development of big data and artificial intelligence technologies, data-driven drilling rate prediction models have become an important component of intelligent drilling systems. However, existing drilling rate prediction technologies still face two major technical bottlenecks in practical applications: (1) The contradiction between data partitioning and learning paradigms. Currently, there are two main machine learning modes for data-driven mechanical drilling rate prediction. Adjacent well prediction: This mode constructs a machine learning model by aggregating historical drilling data from multiple drilled wells within a block. Although it can extract certain patterns of drilling rate changes, due to the intra-layer heterogeneity of the geological environment, i.e., significant differences in rock mechanical properties at different well locations within the same formation, the model has low prediction accuracy when applied to newly drilled wells. Same-well prediction: This mode uses real-time data from the drilled sections of newly drilled wells for model training. Although this mode can capture the local characteristics of newly drilled wells, it is limited by the inter-layer heterogeneity of the formation, and the model often overfits the characteristics of drilled formations, lacking the ability to predict deep, un-drilled formations; in addition, when drilling into the formation transition zone, the model prediction often exhibits severe lag and cannot predict changes in mechanical drilling rate in a timely manner.

[0003] (2) The problem of the uninterpretability of model predictions. In pursuit of higher prediction accuracy, data-driven modeling methods are used, including complex AI algorithms such as deep learning and ensemble learning, which are essentially "black box" models. Although these models can fit complex nonlinear relationships, they cannot provide a clear predictive logic and specific contribution curves of the impact of various drilling parameters (such as drilling pressure, rotational speed, and displacement) on the mechanical drilling rate. In high-risk operation scenarios such as drilling engineering, incorrect parameter adjustment may lead to downhole accidents. The lack of internal logic in the model makes it difficult for on-site engineers to understand and trust the prediction suggestions given by AI, which greatly hinders the actual implementation and widespread application of artificial intelligence technology in drilling operations. Summary of the Invention

[0004] To address the aforementioned issues, the present invention aims to provide a method, system, medium, and equipment for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture. This method can improve the prediction accuracy of mechanical drilling rate, reduce the prediction lag in the formation transition section, and enhance the interpretability of the prediction model, thereby achieving accurate and real-time prediction of mechanical drilling rate and optimization of drilling parameters.

[0005] To achieve the above objectives, in a first aspect, the technical solution adopted by the present invention is as follows: a method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture, employing a cloud-edge collaborative architecture including a cloud server and an edge server, comprising: a data acquisition step: acquiring real-time drilling data and stored historical drilling data, and performing preprocessing: the cloud server preprocesses the historical drilling data, and the edge server preprocesses the real-time drilling data of the current well; the drilling data includes at least drilling parameters and corresponding measured mechanical drilling rate values; a cloud-side pre-training step: training a neural network model on the cloud server using preprocessed historical drilling data from neighboring wells within the block to obtain a pre-trained mechanical drilling rate prediction model, and distributing the pre-trained mechanical drilling rate prediction model to the edge server; and an edge correction step: freezing the pre-trained mechanical drilling rate prediction model on the edge server. The model assigns weights to the pre-trained NMR prediction model and initializes a fine-tuned neural base model with reduced shared basis function count, hidden layer width, and network layer count. It is then trained using real-time drilling parameters of the current drilled section and corresponding measured mechanical drilling speeds (MRS). The model learns the residual between the predicted values ​​of the pre-trained MRS prediction model and the measured MRS. The collaborative prediction step involves superimposing the predicted output of the pre-trained MRS prediction model on the current input with the residual correction value of the fine-tuned NMR model output to obtain the final MRS prediction value. The interpretability analysis step extracts the nonlinear shape function within the pre-trained MRS prediction model to generate contribution curves for each drilling parameter to the MRS prediction value. The parameter optimization step determines the optimization range of the drilling parameters based on the contribution curves, while satisfying equipment operation constraints and drilling safety constraints, and outputs drilling parameter optimization suggestions.

[0006] Furthermore, the neural basis model is a subclass of the generalized additive model, which decomposes the task of predicting the mechanical drilling rate into the sum of independent nonlinear contributions of multiple input features. The contribution function of each input feature is represented as a linear combination of a set of shared basis functions invoked by each input feature; the shape function is a univariate mapping function between each input feature and the predicted mechanical drilling rate.

[0007] Furthermore, the specific implementation process of the cloud-side pre-training step includes: The cloud server obtains historical drilling data of neighboring wells within the block. The drilling data includes well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter and well inclination angle, as well as the measured mechanical drilling rate corresponding to each sampling point. After preprocessing the historical drilling data, it is divided into training set, validation set and test set, which are then input into the neural base model to be trained. With minimizing the global mean square error as the optimization objective, the Adam optimization algorithm is adopted and the model parameters are updated with a preset learning rate. After training, a pre-trained mechanical drilling rate prediction model is obtained.

[0008] Furthermore, the specific implementation process of the edge correction step includes: The edge server receives and freezes the weights of the pre-trained mechanical drilling rate prediction model sent by the cloud server, so that the parameters of the pre-trained mechanical drilling rate prediction model remain unchanged during the current well drilling process. The edge server initializes a fine-tuned neural base model with reduced number of shared base functions, hidden layer width, and network layers. The architecture of the fine-tuned neural base model is the same as that of the pre-trained mechanical drilling rate prediction model, but the number of model parameters is smaller than that of the pre-trained mechanical drilling rate prediction model. The edge server obtains real-time drilling data of the current well and calculates the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. The fine-tuned neural base model is trained using real-time drilling data as input and residuals as supervision signals, and its parameters are updated with the goal of minimizing local mean square error.

[0009] Furthermore, the specific implementation process of the collaborative prediction step includes: The edge server obtains the real-time drilling parameters of the current well and inputs them into the frozen pre-trained mechanical drilling rate prediction model and the trained fine-tuned neural network model, respectively. The drilling parameters include well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter, and well inclination angle. The pre-trained mechanical drilling rate prediction model outputs a baseline mechanical drilling rate prediction value based on real-time drilling parameters. The fine-tuned neural base model outputs residual correction values ​​based on real-time drilling parameters; The edge server adds the baseline mechanical drilling rate prediction value to the residual correction value, and outputs the superimposed result as the final mechanical drilling rate prediction value.

[0010] Furthermore, the specific implementation process of the interpretability analysis step includes: The edge server extracts the univariate shape function of each input feature generated during the forward propagation of the pre-trained mechanical drilling rate prediction model. The univariate shape function is the independent mathematical mapping relationship between each input feature and the mechanical drilling rate prediction value in the pre-trained mechanical drilling rate prediction model. The edge server converts each univariate shape function into a continuous contribution curve and verifies it according to pre-stored monotonicity rules, parameter upper and lower limits, and engineering experience rules. The edge server pushes the verified continuous contribution curves to the field terminal for visualization.

[0011] Furthermore, the specific implementation process of the parameter optimization step includes: The edge server acquires the contribution curves of each controllable drilling parameter, including drilling pressure, rotation speed, and displacement. The edge server provides a continuous mapping function based on a pre-trained mechanical drilling rate prediction model trained from a neural base model. By searching for the extreme points or points of gentle slope in the contribution curves of each controllable drilling parameter, the optimal value range for each controllable drilling parameter corresponding to maximizing the mechanical drilling rate is determined; among which, This represents the independent contribution of the i-th input feature to the mechanical drilling rate; For bias terms; The edge server pushes the optimal value range as drilling parameter optimization suggestions to the field terminal.

[0012] Secondly, the technical solution adopted by this invention is: a mechanical drilling rate prediction and drilling parameter optimization system based on a cloud-edge collaborative architecture, which adopts a cloud-edge collaborative architecture including cloud servers and edge servers, comprising: Data acquisition module: Acquires real-time drilling data and stored historical drilling data, and performs preprocessing: The cloud server preprocesses historical drilling data, and the edge server preprocesses real-time drilling data of the current well. The drilling data includes at least drilling parameters and corresponding measured mechanical drilling rates. Cloud-based pre-training module: The neural network model is trained on the cloud server using historical drilling data from neighboring wells within the block after preprocessing, resulting in a pre-trained mechanical drilling rate prediction model, which is then distributed to the edge server. Edge correction module: On the edge server, the weights of the pre-trained mechanical drilling rate prediction model are frozen, and a fine-tuned neural base model with reduced number of shared basis functions, hidden layer width and network layers is initialized. The model is trained using real-time drilling parameters of the current well section and the corresponding measured mechanical drilling rate to learn the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. Collaborative prediction module: The prediction output of the pre-trained mechanical drilling rate prediction model for the current input is superimposed with the residual correction value of the fine-tuned neural base model output to obtain the final mechanical drilling rate prediction value. Interpretability analysis module: Extracts the nonlinear shape function inside the pre-trained mechanical drilling rate prediction model and generates the contribution curve of each drilling parameter to the mechanical drilling rate prediction value; Parameter optimization module: Based on the contribution curve, determine the optimization range of drilling parameters under the conditions of meeting equipment operation constraints and drilling safety constraints, and output drilling parameter optimization suggestions.

[0013] Thirdly, the technical solution adopted by the present invention is: a computer-readable storage medium for storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any of the methods described above.

[0014] Fourthly, the technical solution adopted by the present invention is: a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods described above.

[0015] The present invention has the following advantages due to the adoption of the above technical solutions: 1. This invention achieves high-precision prediction of drilling machinery speed through a cloud-edge collaborative architecture. This mechanism helps to improve the problems of insufficient prediction accuracy and response lag in traditional prediction methods.

[0016] 2. The neural basis model used in this invention can decompose the prediction logic into intuitive and independent shape functions, which helps field engineers understand the model reasoning process, makes the parameter optimization suggestions scientific and verifiable, and greatly improves the safety and trustworthiness of decision-making.

[0017] 3. The cloud-edge system framework constructed by this invention can reduce the computational resource requirements for well site-side model training and improve the feasibility of on-site deployment. Attached Figure Description

[0018] Figure 1 This is a flowchart of the mechanical drilling rate prediction and drilling parameter optimization method based on cloud-edge collaborative architecture in an embodiment of the present invention; Figure 2 This is a flowchart of the data preprocessing process in an embodiment of the present invention; Figure 3 This is a neural base model framework diagram in an embodiment of the present invention; Figure 4 This is a comparison diagram of the predicted and measured values ​​of mechanical drilling speed in an embodiment of the present invention; Figure 5 This is a diagram illustrating the effect of mechanical drilling speed prediction in an embodiment of the present invention. Detailed Implementation

[0019] To address the issues that a single prediction paradigm in oil drilling cannot take into account the heterogeneity within and between layers, and that machine learning models suffer from lag in response in geological transition zones, high computational overhead, and lack of interpretability, making them difficult for field engineers to trust, this invention provides a method, system, medium, and equipment for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture. This method can efficiently solve the problems of poor prediction performance and delayed prediction information in traditional prediction methods.

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0021] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0022] In one embodiment of the present invention, a method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture is provided. This method employs a cloud-edge collaborative architecture consisting of a cloud server and edge servers, and uses a Neural Basis Model (NBM) model as the algorithm carrier. Global prediction and local correction are achieved through transfer learning and residual learning of a pre-trained mechanical drilling rate prediction model. In this embodiment, a pre-trained mechanical drilling rate prediction model is deployed in the cloud, and a fine-tuned NBM model is deployed at the edge. During model usage, the fine-tuned NBM model corrects the prediction results of the pre-trained mechanical drilling rate prediction model through residual learning, thereby improving the accuracy of mechanical drilling rate prediction. Figure 1 As shown, the method includes the following steps: 1) Data acquisition steps: Acquire real-time drilling data and stored historical drilling data, and perform preprocessing. The cloud server preprocesses the historical drilling data, and the edge server preprocesses the real-time drilling data of the current well. The drilling data includes at least the drilling parameters and the corresponding measured mechanical drilling rate.

[0023] 2) Cloud-side pre-training steps: The neural base model is trained on the cloud server using historical drilling data from neighboring wells within the block after preprocessing, resulting in a pre-trained mechanical drilling rate prediction model, which is then distributed to the edge server. 3) Edge correction step: On the edge server, freeze the weights of the pre-trained mechanical drilling rate prediction model and initialize a fine-tuned neural base model with reduced number of shared basis functions, hidden layer width and network layer number. Train the model using real-time drilling parameters of the current well section and the corresponding measured mechanical drilling rate to learn the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. 4) Collaborative prediction step: The predicted output of the pre-trained mechanical drilling rate prediction model for the current input is superimposed with the residual correction value of the fine-tuned neural base model output to obtain the final mechanical drilling rate prediction value. 5) Interpretability analysis steps: Extract the nonlinear shape function inside the pre-trained mechanical drilling rate prediction model and generate the contribution curve of each drilling parameter to the mechanical drilling rate prediction value; 6) Parameter optimization steps: Based on the contribution curve, determine the optimization range of drilling parameters under the conditions of meeting equipment operation constraints and drilling safety constraints, and output drilling parameter optimization suggestions.

[0024] In step 1) above, noise caused by sensor instability and drilling interference is eliminated by preprocessing the drilling data. For example... Figure 2 As shown, the preprocessing includes outlier removal, missing value completion, data denoising, and normalization of the acquired drilling data. Outlier removal uses the interquartile range detection method; missing value completion uses linear interpolation; data denoising uses a weighted adaptive median filtering algorithm; and normalization uses the Min-Max normalization method.

[0025] Specifically, 1.1) Outlier identification and removal (IQR method): The cloud server and edge server perform interquartile range (IQR) detection on the collected drilling features (such as drill pressure and rotation speed). First, the upper quartile (Q3) and lower quartile (Q1) of the dataset are calculated, and IQR is defined as Q3 - Q1. The outlier judgment interval is set to [Q1 - 1.5 * IQR, Q3 + 1.5 * IQR]. Data points outside this range are identified as pulse anomalies caused by sensor jitter or drilling interference and are removed, with the data at that position set to null.

[0026] 1.2) Missing value completion (linear interpolation method): To address missing values ​​in data sequences caused by packet loss, short-term sensor failure, or being removed as outliers, cloud servers and edge servers employ linear interpolation for repair. This method establishes a linear function relationship between known measurement points before and after the missing point, and the calculation formula is shown in equation (1):

[0027] In the formula, This indicates that the previous known measurement point before the missing point is at... The measured value at the location; Indicates that after the missing point, a known measurement point is located at... The measured value at the location; This indicates the sequence position corresponding to a known measurement point following a missing point. This indicates the sequence position corresponding to the known measurement point preceding the missing point.

[0028] If the length of consecutive missing data exceeds the preset number of sampling points or the preset time threshold, a whole segment is discarded to ensure the physical authenticity of the time-series signal.

[0029] 1.3) Data Denoising (Weighted Adaptive Median Filtering): The weighted adaptive median filtering algorithm includes two mechanisms. First, it performs a weighted fusion of the median and the mean to balance the robustness and smoothness of the filter, as shown in equation (2):

[0030] in, The weighted median, and These represent the original median and mean within the filter window, respectively.

[0031] Then, through two-layer logic identification and replacement of pulse interference, the first layer of judgment is used to determine whether the current window value is contaminated by noise, as shown in formula (3):

[0032] in, , Let T represent the minimum and maximum values ​​within the window, and let T be a preset threshold, which is 3 in this embodiment. This indicates the first level of judgment. If this condition is met, it means the window center value is within the normal fluctuation range, and the process proceeds directly to the next level of judgment. If this condition is not met, it means the window is contaminated by noise, and the window length is increased before re-judging.

[0033] Second-level determination: used to identify the center point to be processed. Whether it is impulse noise can be determined by formula (4):

[0034] in, This is the value of the current sampling point. This indicates the second level of judgment. If the second level of judgment is not satisfied, then a judgment is made. For noise points, use the weighted center value. Replacement is performed. This mechanism can effectively filter out abnormal fluctuations caused by sensor jitter. The filtering window gradually increases according to each filtering decision. In this embodiment, the filtering window length starts with 10 data points and increases by 3 data points each time, with a maximum value of 30.

[0035] 1.4) Normalization (Min-Max): To eliminate the order-of-magnitude differences between different drilling parameters, the Min-Max normalization method is used to scale all feature vectors to the [0,1] interval. The minimum and maximum values ​​are determined only using the training set, and the same parameters are used for the validation set, test set, and marginal data to avoid information leakage. The formula is shown in equation (5):

[0036] in, Represents the original eigenvalues ​​before normalization; Represents the normalized eigenvalues; and These represent the minimum and maximum values ​​of the feature in the dataset, respectively. The pre-trained mechanical drilling rate prediction model uses the same normalization parameters as the fine-tuned neural base model.

[0037] In the above embodiments, such as Figure 3 As shown, the neural basis model is a subclass of the generalized additive model, which decomposes the task of predicting the mechanical drilling rate into the sum of independent nonlinear contributions of multiple input features. The contribution function of each input feature is represented as a linear combination of a set of shared basis functions called by each input feature; the shape function is a univariate mapping function between each input feature and the predicted mechanical drilling rate.

[0038] In this embodiment, the powerful computing resources of the cloud server are used to extract general drilling patterns from historical drilling data to obtain a global drilling rate prediction model. The cloud server processes the drilling data of neighboring wells within the block according to a data preprocessing and standardization process, and then uses the data to train the pre-trained mechanical drilling rate prediction model. The drilling parameters used in this embodiment include well depth, drilling pressure, rotational speed, displacement, bit torque, pump pressure, mud density, bit diameter, and well inclination angle.

[0039] Specifically, the neural basis model achieves high lightweightness and interpretability by learning a set of shared basis functions. The model decomposes the task of predicting mechanical drilling rate into the sum of nonlinear contributions from D input features (such as drilling pressure, rotational speed, etc.), the basic form of which is shown in Equation (6):

[0040] in, For bias terms, This represents the independent contribution of the i-th input feature to the mechanical drilling rate. To address the problem of parameter explosion caused by the increase of features in traditional models, the neural base introduces a shared basis function layer. The contribution function of each feature is represented as a linear combination of basis functions, as shown in equation (7):

[0041] in, It is the k-th basis function generated by the shared multilayer perceptron. It is the fixed weight corresponding to the k-th feature of the i-th feature, and B represents the number of shared basis functions. The final output formula is as shown in equation (8):

[0042] In step 2) above, the cloud server is responsible for integrating historical drilling data from drilled wells and using high-performance computing power to perform pre-training of the neural network-based model. The cloud server has established a centralized database to store drilling parameters and mechanical drilling rate records.

[0043] Specifically, the implementation process of the cloud-side pre-training step includes the following steps: 2.1) The cloud server obtains historical drilling data of all neighboring wells in the block where the current well is located. The drilling data includes well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter and well inclination angle, as well as the measured mechanical drilling rate corresponding to each sampling point.

[0044] 2.2) After preprocessing the historical drilling data, it is divided into training set, validation set and test set, which are then input into the neural base model to be trained.

[0045] In this embodiment, the collected data is preprocessed, including outlier removal, missing value completion, data denoising, and normalization. The neural network is then trained on a cloud server using the preprocessed drilling data to predict the rate of penetration (ROP).

[0046] In this embodiment, 70% of the wells are used as the training set, 20% as the validation set, and 10% as the test set. This division by well ensures that data from the same well only enters one subset of the data. The training set is used to train the model, the validation set is used for hyperparameter selection and early stopping, and the test set is only used for final performance evaluation.

[0047] 2.3) With minimizing the global mean square error as the optimization objective, the Adam optimization algorithm is adopted and the model parameters are updated with a preset learning rate. After training, a pre-trained mechanical drilling speed prediction model is obtained.

[0048] In this embodiment, the Adam optimization algorithm is used during model training, with a learning rate of 0.0001. The training process aims to minimize the global mean square error, as shown in equation (9):

[0049] in, Historical measured mechanical drilling speed The values ​​represent the predicted values ​​of the pre-trained mechanical drilling rate prediction model, where N represents the total number of historical samples used in model training. The trained model parameters are stored in a cloud server, ready to be distributed to edge servers.

[0050] In step 3) above, the edge server is deployed at the current well site and collects drilling data (including drilling parameters and measured mechanical drilling rates at each sampling point) in real time through a distributed sensor system. Drilling parameters include well depth, drilling pressure, rotational speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter, and well inclination angle. The edge server uploads the field data to the cloud server in real time for iterative updates of the pre-trained mechanical drilling rate prediction model. Its task is to perform residual learning under limited computing power, quickly correcting the global prediction value using local data from the current well location. Specifically, the edge correction step includes the following steps: 3.1) During the current well drilling process, the edge server receives and freezes the weights of the pre-trained mechanical drilling rate prediction model sent by the cloud server, including the model structure, model weights, feature order, normalization parameters and version information, so that the parameters of the pre-trained mechanical drilling rate prediction model remain unchanged during the current well drilling process, in order to prevent catastrophic forgetting or overfitting when facing local sparse data.

[0051] 3.2) Initialize a fine-tuned neural base model on the edge server with reduced number of shared base functions, hidden layer width and network layers. The architecture of the fine-tuned neural base model is the same as the model type of the pre-trained mechanical drilling rate prediction model, but the number of model parameters is smaller than that of the pre-trained mechanical drilling rate prediction model.

[0052] 3.3) The edge server obtains the real-time drilling parameters of the current well and the corresponding measured mechanical drilling rate, and calculates the predicted value of the pre-trained mechanical drilling rate prediction model. Compared with the measured value of mechanical drilling speed residuals between .

[0053] residual for:

[0054] The optimization objective of residual learning is to minimize the local mean square error, as shown in equation (11):

[0055] in, The output of the edge-end fine-tuning neural base model. Let represent the input feature vector of the m-th training sample. Let m represent the residual of the m-th training sample, and M represent the total number of local samples participating in the training of the edge residual model. This represents the mean squared error loss of the fine-tuned neural base model.

[0056] 3.4) The fine-tuned neural base model is trained using real-time drilling data as input and residuals as supervision signals, and the parameters of the fine-tuned neural base model are updated with minimizing local mean square error as the optimization objective.

[0057] In step 4) above, the collaborative mechanism employed involves the cloud server pre-training the mechanical drilling rate prediction model using historical drilling data, and the edge server downloading the pre-trained mechanical drilling rate prediction model from the cloud and correcting the model's prediction results. Specifically, the collaborative prediction step includes the following steps: 4.1) The edge server obtains the real-time drilling parameters of the current well and inputs these parameters into the frozen pre-trained mechanical drilling rate prediction model and the trained fine-tuned neural network model, respectively. The drilling parameters include well depth, drilling pressure, rotational speed, displacement, bit torque, pump pressure, mud density, bit diameter, and well inclination angle.

[0058] 4.2) The pre-trained mechanical drilling rate prediction model outputs a baseline mechanical drilling rate prediction value based on real-time drilling parameters. .

[0059] 4.3) Fine-tuning the neural base model outputs residual correction values ​​based on real-time drilling parameters. .

[0060] 4.4) The edge server will predict the baseline mechanical drilling rate. With residual correction value The summation results are used as the final predicted mechanical drilling rate.

[0061] 4.5) When the edge server has insufficient samples, only the predicted value of the pre-trained mechanical drilling rate prediction model is output.

[0062] In this embodiment, based on the length of a single drill pipe on site or a preset update distance, for example, approximately every 10 meters of drilling, the edge server is set to perform a model update every 10 meters of drilling. Model updates are automatically executed during the connection of a single drill pipe (typically about 5 minutes), ensuring real-time model iteration without interrupting drilling operations. Each iteration cycle selects data from the most recent 30 meters of drilled section: the first 20 meters are used for residual model training, the middle 10 meters for validation, and the newly drilled 10 meters are used for real-time performance testing.

[0063] Final mechanical drilling rate prediction output for:

[0064] like Figure 4 The image shows the model's prediction performance; the blue curve represents the measured values, and the red curve represents the model's predicted values.

[0065] In step 5) above, the nonlinear shape function inside the pre-trained mechanical drilling rate prediction model is extracted to make the drilling rate prediction logic transparent, and an optimization scheme for drilling parameters is constructed accordingly. The neural base model used in this invention is a self-explanatory model.

[0066] Specifically, the implementation process of the interpretability analysis step includes the following steps: 5.1) The edge server extracts the univariate shape functions of each input feature generated during the forward propagation of the pre-trained mechanical drilling rate prediction model. The univariate shape function represents the independent mathematical mapping relationship between each input feature and the predicted mechanical drilling rate in the pre-trained mechanical drilling rate prediction model.

[0067] In this embodiment, the univariate shape function of each input feature is generated. This corresponds to the independent contribution of each input parameter to the baseline mechanical drilling rate prediction.

[0068] 5.2) The edge server converts each univariate shape function into a continuous contribution curve and verifies it according to the pre-stored monotonicity rules, parameter upper and lower limits and engineering experience rules.

[0069] In this embodiment, the edge server converts the extracted shape function into a continuous contribution curve and verifies it according to pre-stored monotonicity rules, parameter upper and lower limits, and engineering experience rules. For example... Figure 5 As shown, this is the contribution function of each feature to the predicted mechanical drilling rate.

[0070] 5.3) The edge server pushes the verified continuous contribution curves to the field terminal for visualization display.

[0071] Specifically, the edge server extracts the shape function of the neural network model in real time, and intuitively displays the contribution curves of each drilling parameter to the mechanical drilling rate on the field terminal, helping engineers understand the model's prediction logic. Based on the continuous response curves generated by the model and sensitivity analysis, the edge server pushes optimal drilling parameter combination suggestions to the terminal to maximize the mechanical drilling rate.

[0072] Engineers can visually observe the sensitivity of parameters such as drill pressure, rotational speed, and displacement within different value ranges under current formation conditions. By comparing the shape functions of different parameters, the system helps engineers identify key physical driving factors. For example, when drill pressure exhibits a negative statistical correlation while drill bit torque shows a strong linear positive correlation, the system will prompt engineers that torque is a more direct indicator of current rock breaking efficiency, thus avoiding mechanical efficiency losses caused by blindly increasing drill pressure.

[0073] In step 6) above, the specific implementation process of the parameter optimization step includes the following steps: 6.1) The edge server obtains the contribution curves of each controllable drilling parameter, including drilling pressure, rotation speed and displacement.

[0074] 6.2) The edge server provides a continuous mapping function based on the pre-trained mechanical drilling rate prediction model obtained from the neural base model. By searching for the extreme points or points of gentle slope in the contribution curves of each controllable drilling parameter, the optimal value range for each controllable drilling parameter corresponding to maximizing the mechanical drilling rate is determined; among which, This represents the independent contribution of the i-th input feature to the mechanical drilling rate; This is a bias term.

[0075] 6.3) The edge server pushes the optimal value range as drilling parameter optimization suggestions to the field terminal.

[0076] In this embodiment, the dynamic adjustment closed loop is as follows: the edge server obtains the predicted mechanical drilling rate and contribution function. The system calculates the optimal range for current formation drilling pressure, rotational speed, and displacement, and pushes the optimization plan to the engineer. After the engineer adjusts the parameters according to the suggestions, the edge server obtains new data through the next 10-meter sliding window for fine-tuning, realizing continuous closed-loop updating of the prediction model and optimization suggestions.

[0077] In one embodiment of the present invention, a mechanical drilling rate prediction and drilling parameter optimization system based on a cloud-edge collaborative architecture is provided, which adopts a cloud-edge collaborative architecture including cloud servers and edge servers, and includes: Data acquisition module: Acquires real-time drilling data and stored historical drilling data, and performs preprocessing. The cloud server preprocesses historical drilling data, and the edge server preprocesses real-time drilling data of the current well. The drilling data includes at least drilling parameters and corresponding measured mechanical drilling rates. Cloud-based pre-training module: The neural network model is trained on the cloud server using historical drilling data from neighboring wells within the block after preprocessing, resulting in a pre-trained mechanical drilling rate prediction model, which is then distributed to the edge server. Edge correction module: On the edge server, the weights of the pre-trained mechanical drilling rate prediction model are frozen, and a fine-tuned neural base model with reduced number of shared basis functions, hidden layer width and network layers is initialized. The model is trained using real-time drilling parameters of the current well section and the corresponding measured mechanical drilling rate to learn the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. Collaborative prediction module: The prediction output of the pre-trained mechanical drilling rate prediction model for the current input is superimposed with the residual correction value of the fine-tuned neural base model output to obtain the final mechanical drilling rate prediction value. Interpretability analysis module: Extracts the nonlinear shape function inside the pre-trained mechanical drilling rate prediction model and generates the contribution curve of each drilling parameter to the mechanical drilling rate prediction value; Parameter optimization module: Based on the contribution curve, determine the optimization range of drilling parameters under the conditions of meeting equipment operation constraints and drilling safety constraints, and output drilling parameter optimization suggestions.

[0078] In the above embodiments, the neural basis model is a subclass of the generalized additive model, which decomposes the task of predicting the mechanical drilling rate into the sum of independent nonlinear contributions of multiple input features. The contribution function of each input feature is represented as a linear combination of a set of shared basis functions jointly invoked by each input feature; the shape function is a univariate mapping function between each input feature and the predicted mechanical drilling rate value.

[0079] In the above embodiments, the specific implementation process of the cloud-side pre-training module includes: The cloud server obtains historical drilling data of neighboring wells within the block. The drilling data includes well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter and well inclination angle, as well as the measured mechanical drilling rate corresponding to each sampling point. After preprocessing the historical drilling data, it is divided into training set, validation set and test set, which are then input into the neural base model to be trained. With minimizing the global mean square error as the optimization objective, the Adam optimization algorithm is adopted and the model parameters are updated with a preset learning rate. After training, a pre-trained mechanical drilling rate prediction model is obtained.

[0080] In the above embodiments, the specific implementation process of the edge correction module includes: The edge server receives and freezes the weights of the pre-trained mechanical drilling rate prediction model sent by the cloud server, so that the parameters of the pre-trained mechanical drilling rate prediction model remain unchanged during the current well drilling process. The edge server initializes a fine-tuned neural base model with reduced number of shared base functions, hidden layer width, and network layers. The architecture of the fine-tuned neural base model is the same as that of the pre-trained mechanical drilling rate prediction model, but the number of model parameters is smaller than that of the pre-trained mechanical drilling rate prediction model. The edge server obtains real-time drilling data of the current well and calculates the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. The fine-tuned neural base model is trained using real-time drilling data as input and residuals as supervision signals, and its parameters are updated with the goal of minimizing local mean square error.

[0081] In the above embodiments, the specific implementation process of the collaborative prediction module includes: The edge server obtains the real-time drilling parameters of the current well and inputs them into the frozen pre-trained mechanical drilling rate prediction model and the trained fine-tuned neural network model, respectively. The drilling parameters include well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter, and well inclination angle. The pre-trained mechanical drilling rate prediction model outputs a baseline mechanical drilling rate prediction value based on real-time drilling parameters. The fine-tuned neural base model outputs residual correction values ​​based on real-time drilling parameters; The edge server adds the baseline mechanical drilling rate prediction value to the residual correction value, and outputs the superimposed result as the final mechanical drilling rate prediction value.

[0082] In the above embodiments, the specific implementation process of the interpretability analysis module includes: The edge server extracts the univariate shape function of each input feature generated during the forward propagation of the pre-trained mechanical drilling rate prediction model. The univariate shape function is the independent mathematical mapping relationship between each input feature and the mechanical drilling rate prediction value in the pre-trained mechanical drilling rate prediction model. The edge server converts each univariate shape function into a continuous contribution curve and verifies it according to pre-stored monotonicity rules, parameter upper and lower limits, and engineering experience rules. The edge server pushes the verified continuous contribution curves to the field terminal for visualization.

[0083] In the above embodiments, the specific implementation process of the parameter optimization module includes: The edge server acquires the contribution curves of each controllable drilling parameter, including drilling pressure, rotation speed, and displacement. The edge server provides a continuous mapping function based on a pre-trained mechanical drilling rate prediction model trained from a neural base model. By searching for the extreme points or points of gentle slope in the contribution curves of each controllable drilling parameter, the optimal value range for each controllable drilling parameter corresponding to maximizing the mechanical drilling rate is determined; among which, This represents the independent contribution of the i-th input feature to the mechanical drilling rate; For bias terms; The edge server pushes the optimal value range as drilling parameter optimization suggestions to the field terminal.

[0084] The system provided in this embodiment is used to execute the above-described method embodiments. For specific processes and details, please refer to the above embodiments, which will not be repeated here.

[0085] This invention provides a computing device, which may be a terminal, and may include one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any of the methods in the above embodiments.

[0086] Furthermore, when the aforementioned instructions can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0087] In one embodiment of the present invention, a computer program product is provided, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to perform the methods provided in the above-described method embodiments.

[0088] In one embodiment of the present invention, a non-transitory computer-readable storage medium is provided, which stores server instructions that cause a computer to perform the methods provided in the above embodiments.

[0089] The computer-readable storage medium provided in the above embodiments has a similar implementation principle and technical effect to the above method embodiments, and will not be described again here.

[0090] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0091] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0092] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0093] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture, characterized in that: include: Data acquisition steps: Acquire real-time drilling data and stored historical drilling data, and perform preprocessing: The cloud server preprocesses the historical drilling data, and the edge server preprocesses the real-time drilling data of the current well. The drilling data includes at least the drilling parameters and the corresponding measured mechanical drilling rate. Cloud-based pre-training steps: The neural network model is trained on the cloud server using historical drilling data from neighboring wells within the block after preprocessing, resulting in a pre-trained mechanical drilling rate prediction model, which is then distributed to the edge server. Edge correction steps: On the edge server, freeze the weights of the pre-trained mechanical drilling rate prediction model and initialize a fine-tuned neural base model with reduced number of shared basis functions, hidden layer width and network layers. Train the model using real-time drilling parameters of the current drilled section and the corresponding measured mechanical drilling rate to learn the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. Collaborative prediction step: The predicted output of the pre-trained mechanical drilling rate prediction model for the current input is superimposed with the residual correction value of the fine-tuned neural base model output to obtain the final mechanical drilling rate prediction value. Interpretability analysis steps: Extract the nonlinear shape function inside the pre-trained mechanical drilling rate prediction model and generate the contribution curves of each drilling parameter to the mechanical drilling rate prediction value; Parameter optimization steps: Based on the contribution curve, determine the optimization range of drilling parameters under the conditions of meeting equipment operation constraints and drilling safety constraints, and output drilling parameter optimization suggestions.

2. The method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture as described in claim 1, characterized in that, The neural basis model is a subclass of the generalized additive model. It decomposes the task of predicting the mechanical drilling rate into the sum of independent nonlinear contributions of multiple input features. The contribution function of each input feature is represented as a linear combination of a set of shared basis functions invoked by all input features. The shape function is a univariate mapping function between each input feature and the predicted mechanical drilling rate.

3. The method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture as described in claim 1, characterized in that, The specific implementation process of the cloud-side pre-training step includes: The cloud server obtains historical drilling data of neighboring wells within the block. The drilling data includes well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter and well inclination angle, as well as the measured mechanical drilling rate corresponding to each sampling point. After preprocessing the historical drilling data, it is divided into training set, validation set and test set, which are then input into the neural base model to be trained. With minimizing the global mean square error as the optimization objective, the Adam optimization algorithm is adopted and the model parameters are updated with a preset learning rate. After training, a pre-trained mechanical drilling rate prediction model is obtained.

4. The method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture as described in claim 1, characterized in that, The specific implementation process of the edge correction step includes: The edge server receives and freezes the weights of the pre-trained mechanical drilling rate prediction model sent by the cloud server, so that the parameters of the pre-trained mechanical drilling rate prediction model remain unchanged during the current well drilling process. The edge server initializes a fine-tuned neural base model with reduced number of shared base functions, hidden layer width, and network layers. The architecture of the fine-tuned neural base model is the same as that of the pre-trained mechanical drilling rate prediction model, but the number of model parameters is smaller than that of the pre-trained mechanical drilling rate prediction model. The edge server obtains real-time drilling data of the current well and calculates the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. The fine-tuned neural base model is trained using real-time drilling data as input and residuals as supervision signals, and its parameters are updated with the goal of minimizing local mean square error.

5. The method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture as described in claim 1, characterized in that, The specific implementation process of the collaborative prediction step includes: The edge server obtains the real-time drilling parameters of the current well and inputs them into the frozen pre-trained mechanical drilling rate prediction model and the trained fine-tuned neural network model, respectively. The drilling parameters include well depth, drilling pressure, rotation speed, displacement, drill bit torque, pump pressure, mud density, drill bit diameter, and well inclination angle. The pre-trained mechanical drilling rate prediction model outputs a baseline mechanical drilling rate prediction value based on real-time drilling parameters. The fine-tuned neural base model outputs residual correction values ​​based on real-time drilling parameters; The edge server adds the baseline mechanical drilling rate prediction value to the residual correction value, and outputs the superimposed result as the final mechanical drilling rate prediction value.

6. The method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture as described in claim 1, characterized in that, The specific implementation process of the interpretability analysis step includes: The edge server extracts the univariate shape function of each input feature generated during the forward propagation of the pre-trained mechanical drilling rate prediction model. The univariate shape function is the independent mathematical mapping relationship between each input feature and the mechanical drilling rate prediction value in the pre-trained mechanical drilling rate prediction model. The edge server converts each univariate shape function into a continuous contribution curve and verifies it according to pre-stored monotonicity rules, parameter upper and lower limits, and engineering experience rules. The edge server pushes the verified continuous contribution curves to the field terminal for visualization.

7. The method for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture as described in claim 1, characterized in that, The specific implementation process of the parameter optimization steps includes: The edge server acquires the contribution curves of each controllable drilling parameter, including drilling pressure, rotation speed, and displacement. The edge server provides a continuous mapping function based on a pre-trained mechanical drilling rate prediction model trained from a neural base model. By searching for the extreme points or points of gentle slope in the contribution curves of each controllable drilling parameter, the optimal value range for each controllable drilling parameter corresponding to maximizing the mechanical drilling rate is determined; among which, This represents the independent contribution of the i-th input feature to the mechanical drilling rate; For bias terms; The edge server pushes the optimal value range as drilling parameter optimization suggestions to the field terminal.

8. A system for predicting mechanical drilling rate and optimizing drilling parameters based on a cloud-edge collaborative architecture, characterized in that, include: Data acquisition module: Acquires real-time drilling data and stored historical drilling data, and performs preprocessing: The cloud server preprocesses historical drilling data, and the edge server preprocesses real-time drilling data of the current well. The drilling data includes at least drilling parameters and corresponding measured mechanical drilling rates. Cloud-based pre-training module: The neural network model is trained on the cloud server using historical drilling data from neighboring wells within the block after preprocessing, resulting in a pre-trained mechanical drilling rate prediction model, which is then distributed to the edge server. Edge correction module: On the edge server, the weights of the pre-trained mechanical drilling rate prediction model are frozen, and a fine-tuned neural base model with reduced number of shared basis functions, hidden layer width and network layers is initialized. The model is trained using real-time drilling parameters of the current well section and the corresponding measured mechanical drilling rate to learn the residual between the predicted value of the pre-trained mechanical drilling rate prediction model and the measured mechanical drilling rate. Collaborative prediction module: The prediction output of the pre-trained mechanical drilling rate prediction model for the current input is superimposed with the residual correction value of the fine-tuned neural base model output to obtain the final mechanical drilling rate prediction value. Interpretability analysis module: Extracts the nonlinear shape function inside the pre-trained mechanical drilling rate prediction model and generates the contribution curve of each drilling parameter to the mechanical drilling rate prediction value; Parameter optimization module: Based on the contribution curve, determine the optimization range of drilling parameters under the conditions of meeting equipment operation constraints and drilling safety constraints, and output drilling parameter optimization suggestions.

9. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 7.

10. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 7.

Images