Intelligent well building safety management and control system

Abstract

The invention discloses an intelligent well building safety management and control system, which relates to the technical field of coal well building, and comprises a data acquisition and fusion module, a risk assessment module, a decision engine module, a control execution module, a man-machine interaction module and a decision knowledge base module, evaluating the potential risk of well building; the decision engine module constructs a decision strategy generation model based on a risk assessment result and a decision knowledge base query result, and generates a high-level decision instruction and a parameter adjustment sequence; and the control execution module converts the parameter adjustment sequence into a control instruction sequence, and issues the control instruction sequence to an actual well building control system for execution through a data communication interface. According to the method, based on a Transform encoder-decoder structure, an augmentation strategy and self-supervised contrast learning are introduced, the generalization ability and robustness of a risk assessment model are improved, and risk situation time sequence prediction and risk level classification are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of safety management and control, and in particular to an intelligent well construction safety management and control system. Background Art

[0002] Coal mining is a high-risk operation, in which the well construction process is one of the key links. Due to the complex and changeable geological conditions underground, huge rock pressure and surrounding rock squeezing force, there are many safety hazards in well construction, such as gas outburst, water inrush, well collapse, pipe column damage, etc. Once they occur, they will pose a great threat to the safety of operators and equipment, and cause serious economic losses.

[0003] Traditional well construction safety management and control methods mainly rely on manual monitoring and experience judgment, and there are some obvious defects. On the one hand, delayed response is a major problem. The well construction environment is extremely complex and changeable, and safety risks often come swiftly and ruthlessly. It is difficult to achieve real-time monitoring and rapid response by relying on manual judgment. On the other hand, the one-sided and subjective risk assessment caused by human factors is also a major drawback. Different people have different experiences and judgment abilities, and it is difficult for people to concentrate for a long time, so it is easy to miss key risk signals. The traditional method lacks real-time, accuracy and efficiency, which is not conducive to timely discovery and control of potential risks, and it is difficult to fundamentally guarantee the safety and reliability of well construction operations. Summary of the invention

[0004] In view of the fact that the existing well construction safety management and control methods mainly rely on manual monitoring and experience judgment, there are obvious defects such as delayed response and one-sided risk assessment. It is difficult to timely discover and control various safety risks and hidden dangers in well construction operations, thus affecting the safety and efficiency of the operations. The present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to achieve real-time monitoring, intelligent evaluation and rapid decision-making of well construction risks, so as to effectively improve the safety level and reliability of well construction operations.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: In the first aspect, an embodiment of the present invention provides an intelligent well construction safety management and control system, which includes a data acquisition and fusion module, which is used to collect various raw data in the well construction process, pre-process the raw data, construct a standardized time series data sample set, and input the time series data sample set into a risk assessment module; the risk assessment module builds a risk assessment model based on the encoder-decoder framework of Transformer, receives the time series data sample set output by the data acquisition and fusion module, evaluates the potential risk of well construction in real time, and inputs the evaluation results into a decision engine module and a human-computer interaction module, wherein the evaluation results include a risk situation sequence and a risk level; the decision engine module is used to send a query request to a decision knowledge base module, and based on the risk situation sequence and risk level output by the risk assessment module, build a decision strategy generation model in combination with the query results to generate high-level decision instructions and parameter adjustment sequence, input high-level decision instructions into the human-computer interaction module, and input the parameter adjustment sequence into the control execution module; the control execution module is used to receive the parameter adjustment sequence output by the decision engine module, convert the parameter adjustment sequence into a corresponding control instruction sequence, and send the control instruction sequence to the actual well construction control system for execution through the data communication interface; the machine interaction module includes a visualization interface and a natural language interaction interface. The visualization interface is used to display risk assessment results, decision results and control execution status. The natural language interaction interface supports manual intervention in decision-making, and feeds back manual intervention instructions to the risk assessment module, the decision engine module or the optimization control module; the decision knowledge base module is used to build a decision knowledge base, receive query requests from the decision engine module, retrieve relevant data from the knowledge base according to the query request, and input the retrieval results into the decision engine module.

[0007] As a preferred solution of the intelligent well construction safety management and control system described in the present invention, the operation process of the data acquisition and fusion module is as follows: various original data in the well construction process are collected to construct an original well construction process data set; from the original well construction process data set, feature data related to well construction risk assessment is selected; the selected feature data is fused using a self-attention mechanism, and adaptive weights are assigned according to the importance of each feature to risk assessment; missing value processing is performed on the fused feature data, features with missing values ​​are detected, and the missing values ​​are filled using the historical data interpolation method of similar features; feature encoding is performed on the processed feature data set, One-Hot encoding is performed on discrete features, and normalization encoding is performed on continuous features; according to the time attribute of the feature, the encoded feature data is divided into three categories: time series features, discrete features and continuous features; sliding window segmentation is performed on the time series features, the window size is set, and a time series feature sequence is obtained; the segmented time series feature sequence is spliced ​​with the discrete features and continuous features in the corresponding window to construct multivariate feature data; the spliced ​​multivariate feature data is formatted into a time series data sample set, and the formatted time series data sample set is input into the risk assessment module.

[0008] As a preferred solution of the intelligent well construction safety management and control system described in the present invention, the operation process of the risk assessment module is as follows: receiving the time series data sample set output by the data acquisition and fusion module, and dividing the time series data sample set into a training set and a test set according to a preset ratio; augmenting the training set data to obtain an augmented data set, wherein the augmentation includes time series perturbation augmentation, feature mask augmentation and semantic preservation augmentation; designing a Transformer-based encoder-decoder model framework as a risk assessment model; adopting the self-supervised comparative learning strategy of the masked language model to train the risk assessment model, and optimizing the parameters of the encoder and decoder; during the training process, the cycle The current model performance is evaluated on the test set in a systematic way. If the performance meets the predefined conditions, the training is stopped in advance. Otherwise, the training is continued until the maximum number of training rounds is reached. The generalization performance of the risk assessment model after training is comprehensively evaluated on the test set. If it meets the preset evaluation indicators, it is deployed. The deployed risk assessment model is used to conduct risk assessment on the real-time well construction data stream to obtain the risk situation sequence and risk level. The risk situation sequence is input into the human-computer interaction module, and feedback instructions are received to determine whether the risk assessment model needs to be retrained. If necessary, new well construction process data is collected, and retraining and evaluation are performed. If not, the risk situation sequence and risk level are input into the decision engine module.

[0009] As a preferred solution of the intelligent well construction safety management and control system described in the present invention, the framework of the risk assessment model includes an encoder part and a decoder part, the encoder part includes a timing encoder, a risk relationship modeler and an attention fusion module, and the decoder part includes a first decoding head and a second decoding head.

[0010] As a preferred solution of the intelligent well construction safety management and control system of the present invention, the risk assessment model is trained including the following steps: obtaining the original training set and the augmented data set; sampling a batch of data X and X' from the original training data and the augmented data set respectively; randomly masking the partial positions of X and X' to obtain masked inputs X_masked and X'_masked; inputting X_masked and X'_masked into the encoder respectively, and the encoder outputs the corresponding time series feature representation and ; Calculate the contrast loss of the encoder output ;Will and Input the first decoding head and the second decoding head of the decoder respectively, the first decoding head outputs the predicted sequence Y_pred and Y'_pred, and calculates the generation loss of Y_pred and the original label Y ; The second decoding head outputs the classification probabilities P and P', and calculates the classification loss of P and the original label C ; The contrast loss , generate loss , classification loss Perform weighted summation to get the total loss ; Based on total loss , calculate the gradient of the training parameters, and use the gradient descent algorithm to back-propagate and optimize the training parameters; contrast loss The specific formula is as follows:

[0011] in, and Represent the output representation of the encoder for the original input X and the augmented input X', respectively. for The transpose of is the L2 norm, is a scalar.

[0012] As a preferred solution of the intelligent well construction safety management and control system described in the present invention, the operation process of the decision engine module is as follows: receiving the risk situation sequence and risk level output by the risk assessment module; sending a query request to the decision knowledge base module, the query conditions include the risk situation sequence, risk level and the current well construction stage, receiving the query results returned by the decision knowledge base module, and the query results include parameter data subsets, operation process templates, historical risk cases and safety constraints; constructing a decision strategy generation model, taking the query results as input, and generating a series of candidate decision strategy sets based on the machine learning algorithm; for each decision strategy in the candidate strategy set, evaluating its expected risk level, and selecting the decision strategy with the lowest risk as output; decomposing the decision strategy into two parts: a well construction parameter adjustment plan and a high-level decision instruction, and inputting the high-level decision instruction into the human-computer interaction module for manual review; if the manual feedback decision is unreasonable, inputting a modification instruction through the natural language interaction interface, feeding back the modification instruction to the decision engine module, and regenerating the decision strategy; if the manual feedback decision is reasonable, combining the well construction parameter adjustment plan with the real-time well construction parameters, using the model predictive control algorithm, calculating the optimized parameter adjustment sequence and inputting it into the control execution module.

[0013] As a preferred solution of the intelligent well construction safety management and control system described in the present invention, the operation process of the control execution module is as follows: receiving the optimized parameter adjustment sequence from the decision engine module; parsing the parameter adjustment sequence, identifying the adjustment object and its adjustment target value; obtaining the current state data of the real-time well construction control system, including the state feedback of each actuator; comparing the parameter adjustment target value with the current mechanism state, and calculating the incremental adjustment amount of the control execution; matching the incremental adjustment amount with the underlying control command set of the well construction control system to generate the corresponding control instruction sequence; and transmitting the control instruction sequence to the control system through industrial Ethernet, fieldbus or other industrial communication protocols. The system sends the execution feedback of the well construction control system to the corresponding execution unit of the well construction control system in real time; continuously receives the execution feedback of the well construction control system to monitor whether the parameters are adjusted as expected; if it is detected that the execution deviation exceeds the allowable range, the feedback data is passed to the decision engine module to determine whether the decision strategy needs to be regenerated; the execution process is presented on the human-computer interaction interface to support manual monitoring and intervention. If an abnormality is found manually, the correction instruction is input to the control execution module through the natural language interaction interface; after receiving the correction instruction, the control execution module regenerates and sends the control instruction sequence; if the parameter adjustment is completed and the execution result is normal, the final status feedback is passed to the decision engine module.

[0014] In the second aspect, an embodiment of the present invention provides an intelligent well construction safety management and control method, which includes collecting various raw data in the well construction process, preprocessing the raw data, and constructing a standardized time series data sample set; constructing a risk assessment model based on the Transformer encoder-decoder framework, receiving the time series data sample set, and evaluating the potential risks of well construction in real time; inputting the risk assessment results into the decision generation link and the human-computer interaction interface; sending a query request to the decision knowledge base, retrieving relevant data from the decision knowledge base according to the query request, and using the retrieval results for decision generation; based on the risk assessment results and in combination with the query results of the decision knowledge base, constructing a decision strategy generation model, generating high-level decision instructions and parameter adjustment sequences; inputting the high-level decision instructions into the human-computer interaction interface, and inputting the parameter adjustment sequence into the control execution link; the visual interface of the human-computer interaction interface displays the risk assessment results, decision results and control execution status; the natural language interaction interface of the human-computer interaction interface supports manual intervention in decision-making, and feeds back the intervention instructions to the risk assessment, decision generation or control execution link; receiving the parameter adjustment sequence output by the decision generation link, converting it into a control instruction sequence, and sending it to the actual well construction control system for execution through the data communication interface.

[0015] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the intelligent well construction safety management and control system as described in the first aspect of the present invention are implemented.

[0016] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the intelligent well construction safety management and control system as described in the first aspect of the present invention are implemented.

[0017] The beneficial effects of the present invention are as follows: the present invention integrates multi-source heterogeneous data, assigns adaptive weights through the self-attention mechanism, improves data quality and feature importance recognition capabilities, and provides high-quality input data for the risk assessment model. Based on the Transformer encoder-decoder structure, the augmentation strategy and self-supervised comparative learning are introduced to improve the generalization ability and robustness of the risk assessment model, and realize the time series prediction of risk situations and risk level classification. Combining the knowledge base and machine learning algorithms, a variety of candidate decision strategies are generated, and the strategy with the lowest risk is optimized and selected, which realizes the intelligent generation of decision-making and improves the effectiveness and accuracy of decision-making. Through the human-computer interaction interface, it supports real-time manual monitoring and intervention in the decision-making and control execution process, giving full play to the advantages of human-computer collaboration and improving the interpretability and security of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 This is the module connection diagram of the intelligent well construction safety management and control system.

[0020] Figure 2 This is the operation flow chart of the data collection and fusion module of the intelligent well construction safety management and control system.

[0021] Figure 3 Flowchart for training risk assessment model for intelligent well construction safety management and control system.

[0022] Figure 4 This is a risk assessment model framework diagram for the intelligent well construction safety management and control system.

[0023] Figure 5 Diagram of computer equipment for intelligent well safety management and control system. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from the description. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0027] Example 1

[0028] Reference Figure 1~Figure 5 , which is the first embodiment of the present invention, provides an intelligent well construction safety management and control system, including: a data acquisition and fusion module, which is used to collect various raw data in the well construction process, pre-process the data, construct a standardized time series data sample set, and input the time series data sample set into the risk assessment module.

[0029] Preferably, the operation process of the data acquisition and fusion module is as follows: collect various raw data in the well construction process, including well construction data, geological data, well condition data, etc., to construct an original well construction process data set; select feature data related to well construction risk assessment from the original well construction process data set; use the self-attention mechanism to fuse the selected feature data, and assign adaptive weights to each feature according to its importance to risk assessment; perform missing value processing on the fused feature data, detect features with missing values, and use the historical data interpolation method of similar features to fill in the missing values; perform feature encoding on the processed feature data set, perform One-Hot encoding on discrete features, and perform One-Hot encoding on continuous features. The continuous features are coded in a normalized manner; according to the time attributes of the features, the coded feature data are divided into three categories: time series features, discrete features, and continuous features; the time series features are segmented by sliding windows, and the window size (time step) is set to B to obtain a time series feature sequence; the segmented time series feature sequence is concatenated with the discrete features and continuous features in the corresponding window to construct multivariate feature data; the concatenated multivariate feature data is formatted into a time series data sample set, including information such as the number of samples, time step B, time series feature dimension, discrete feature dimension, and continuous feature dimension; the formatted time series data sample set is input into the risk assessment module as the input data for the risk assessment model.

[0030] The risk assessment module builds a risk assessment model based on the encoder-decoder framework of Transformer, receives the time series data sample set output by the data acquisition and fusion module, evaluates the potential risks of well construction in real time, and inputs the risk situation sequence into the decision engine module and the human-computer interaction module.

[0031] Preferably, the operation process of the risk assessment module is as follows: receiving the time series data sample set output by the data acquisition and fusion module, and dividing the time series data sample set into a training set and a test set according to a preset ratio; augmenting the training set data including time series perturbation augmentation, feature mask augmentation and semantic preservation augmentation to obtain an augmented data set; designing a Transformer-based encoder-decoder model framework as a risk assessment model; adopting a self-supervised comparative learning strategy of a masked language model (MLM) to train the risk assessment model, and simultaneously optimize the parameters of the encoder and decoder; during the training process, periodically evaluating the current model performance on the test set, and if the performance meets the predefined conditions If the risk assessment model meets the preset evaluation criteria, the training is stopped in advance; otherwise, the training is continued until the maximum number of training rounds is reached; the generalization performance of the risk assessment model after training is comprehensively evaluated on the test set, and if it meets the preset evaluation indicators, it is deployed; the deployed risk assessment model is used to conduct risk assessment on the real-time well construction data stream to obtain the risk situation sequence and risk level; the risk situation sequence is input into the human-computer interaction module, feedback instructions are received, and it is determined whether the risk assessment model needs to be retrained; if necessary, new well construction process data is collected, and retraining and evaluation are performed; if not, the risk situation sequence and risk level are input into the decision engine module; feedback data of the actual well construction process is recorded for continuous optimization of the risk assessment model.

[0032] Specifically, the framework of the risk assessment model includes an encoder part and a decoder part. The encoder part includes a time series encoder, a risk relationship modeler and an attention fusion module, and the decoder part includes a first decoding head and a second decoding head. The time series encoder is based on the Transformer encoder structure, and the input is the original training set and the augmented data set. It encodes the two input sequences, learns the adversarial time series representation, and outputs the time series feature representation; the risk relationship modeler is based on the relational graph neural network (RelGNN) structure, constructs a heterogeneous relationship graph of time series factors and risk factors, learns the spatiotemporal risk evolution pattern, and outputs the spatiotemporal risk pattern representation; the attention fusion module, based on the multi-head attention mechanism, splices the time series representation vector and the spatiotemporal risk pattern representation and inputs them, calculates the attention weight, fuses the local time series and global risk representation, and outputs the fused risk feature representation; the first decoding head is a risk situation generator, based on the Transformer decoder structure, the input is the fused risk feature representation, the autoregression generates a continuous risk situation sequence, and the output is the predicted risk situation sequence; the second decoding head is a risk classifier, based on the fully connected neural network structure, the input is the fused risk representation tensor, and the output is the discrete risk state category prediction probability.

[0033] Specifically, training the risk assessment model includes the following steps: obtaining the original training set and the augmented data set; sampling a batch of data X and X' from the original training data and the augmented data set respectively; randomly masking parts of X and X' to obtain masked inputs X_masked and X'_masked; inputting X_masked and X'_masked into the encoder respectively, and the encoder outputs the corresponding time series feature representation and ; Calculate the contrast loss of the encoder output ;Will and Input the first decoding head and the second decoding head of the decoder respectively. The first decoding head (risk situation generator) outputs the predicted sequences Y_pred and Y'_pred. Calculate the generation loss of Y_pred and the original label Y ; The second decoding head (risk classifier) ​​outputs the classification probabilities P and P', and calculates the classification loss between P and the original label C ; The contrast loss , generate loss , classification loss Perform weighted summation to get the total loss ; Based on total loss , calculate the gradient of the training parameters, and use the gradient descent algorithm to back-propagate and optimize the training parameters.

[0034] It should be noted that the contrast loss , generate loss , classification loss and total loss The specific formula is as follows: ; L\_ gen=-\frac {\sum{log(y\_ pred[t])(y\_ true[t])}} {B} ; ; ;

[0035] in, and Represent the output representation of the encoder for the original input X and the augmented input X', respectively. for The transpose of is the L2 norm, is the original risk situation sequence label, is the predicted sequence probability distribution output by the decoder, The index iterates over each time step of the sequence, represents the sequence time step (window size), One-Hot label for the original risk level. is the classification probability distribution output by the decoder, and is an adjustable weight coefficient (used to balance the relative importance of the three losses), is a scalar, express and is a positive sample pair (original sample and augmented sample), represents a negative sample pair.

[0036] The decision knowledge base module is used to build a decision knowledge base, receive query requests from the decision engine module, retrieve relevant data (a subset of well construction operation parameter data, the operation process template of the current stage, historical cases similar to the query risk, and safety constraints of the current formation environment) from the knowledge base according to the query request, and input the retrieval results into the decision engine module.

[0037] Specifically, parameter data recorded in historical well construction operations are collected to construct a well construction operation parameter data set, which covers the historical values ​​of various well construction parameters, such as drilling speed, torque, pump pressure, etc.; operation process templates are extracted from well construction operation specification documents to build an operation process template library, which describes the standard operating procedures and parameter ranges for each stage of well construction; historical well construction risk events and their treatment plans are summarized and analyzed to build a risk case library, which provides reference and reference for new risk scenarios; a safety constraint condition library is established, including safety constraint conditions under various formation environments, such as wellbore stability, pressure window, etc.; new decision cases output by the decision engine are received and added to the risk case library to continuously enrich the knowledge base.

[0038] The decision engine module is used to build a decision strategy generation model based on the risk situation sequence and risk level output by the risk assessment module and the decision knowledge base, generate high-level decision instructions and parameter adjustment sequences, input the high-level decision instructions into the human-computer interaction module, and input the parameter adjustment sequence into the control execution module.

[0039] Preferably, the operation flow of the decision engine module is as follows: receiving the risk situation sequence and risk level output by the risk assessment module; sending a query request to the decision knowledge base module, the query conditions including the risk situation sequence, risk level and the current well construction stage, receiving the query results returned by the decision knowledge base module, including the well construction operation parameter data subset, the operation process template of the current stage, the historical cases similar to the query risk, and the safety constraints of the current formation environment; constructing a decision strategy generation model, taking the parameter data subset, the operation process template, the historical risk cases and the safety constraints as input, and generating a series of candidate decision strategy sets based on the machine learning algorithm; for each decision strategy in the candidate strategy set, evaluating its prediction The risk level in the future period is calculated, and the decision strategy with the lowest risk is selected as the output; the decision strategy is decomposed into two parts: the well construction parameter adjustment plan and the high-level decision instructions (parameter adjustment / correction / pause, etc.), and the high-level decision instructions are input into the human-computer interaction module for manual review; if the manual feedback decision is unreasonable, the modification instruction is input through the natural language interaction interface, and the modification instruction is fed back to the decision engine module to regenerate the decision strategy; if the manual feedback decision is reasonable, the well construction parameter adjustment plan is combined with the real-time well construction parameters, and the model predictive control algorithm is used to calculate the optimized parameter adjustment sequence, and the parameter adjustment sequence is input into the control execution module, and the decision case is input into the decision knowledge base module for knowledge base update.

[0040] The control execution module is used to receive the parameter adjustment sequence output by the decision engine module, convert it into a corresponding control instruction sequence, and send it to the actual well construction control system for execution through the data communication interface.

[0041] Preferably, the operation flow of the control execution module is as follows: receiving the optimized parameter adjustment sequence from the decision engine module; parsing the parameter adjustment sequence, identifying the adjustment object (such as well construction fluid performance parameters, drill pipe speed, etc.) and its adjustment target value; obtaining the current state data of the real-time well construction control system, including the state feedback of each actuator; comparing the parameter adjustment target value with the current mechanism state, and calculating the incremental adjustment amount of the control execution; matching the incremental adjustment amount with the underlying control command set of the well construction control system, and generating the corresponding control instruction sequence; sending the control instruction sequence to the well construction control system in real time through industrial Ethernet, fieldbus or other industrial communication protocols. The corresponding execution unit of the well construction control system continuously receives the execution feedback of the well construction control system to monitor whether the parameters are adjusted as expected; if it is detected that the execution deviation exceeds the allowable range, the feedback data is passed to the decision engine module to determine whether the decision strategy needs to be regenerated; at the same time, the execution process is presented on the human-computer interaction interface to support manual monitoring and intervention. If an abnormality is found manually, the correction instruction is input through the natural language interaction interface. After receiving the correction instruction, the control execution module regenerates and issues a new control instruction sequence to cover the previous control action; once the parameter adjustment is completed and the execution result is normal, the final status feedback is passed to the decision engine module to provide feedback for subsequent decisions.

[0042] The human-computer interaction module includes a visualization interface and a natural language interaction interface. The visualization interface is used to display risk assessment results, decision results and control execution status. The natural language interaction interface supports manual intervention in decision-making and feeds back manual intervention instructions to the risk assessment module, decision engine module or optimization control module.

[0043] It should be noted that manual intervention includes feedback instructions, modification instructions, and correction instructions.

[0044] Furthermore, this embodiment also provides an intelligent well construction safety management and control method, including the present invention fusing multi-source heterogeneous data, assigning adaptive weights through the self-attention mechanism, improving data quality and feature importance recognition capabilities, and providing high-quality input data for the risk assessment model. Based on the Transformer encoder-decoder structure, the augmentation strategy and self-supervised comparative learning are introduced to improve the generalization ability and robustness of the risk assessment model, and realize the time series prediction of risk situation and risk level classification. Combined with the knowledge base and machine learning algorithm, a variety of candidate decision strategies are generated, and the strategy with the lowest risk is optimized and selected, which realizes the intelligence of decision generation and improves the effectiveness and accuracy of decision making. Through the human-computer interaction interface, it supports real-time manual monitoring and intervention in the decision-making and control execution process, giving full play to the advantages of human-computer collaboration and improving the interpretability and safety of the system.

[0045] This embodiment also provides a computer device, which is suitable for the case of an intelligent well construction safety management and control system, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to realize the intelligent well construction safety management and control system proposed in the above embodiment.

[0046] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0047] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, an intelligent well construction safety management and control system as proposed in the above embodiment is implemented.

[0048] In summary, the present invention integrates multi-source heterogeneous data, assigns adaptive weights through the self-attention mechanism, improves data quality and feature importance recognition capabilities, and provides high-quality input data for the risk assessment model. Based on the Transformer encoder-decoder structure, the augmentation strategy and self-supervised comparative learning are introduced to improve the generalization ability and robustness of the risk assessment model, and realize the time series prediction of risk situation and risk level classification. Combining the knowledge base and machine learning algorithm, a variety of candidate decision strategies are generated, and the strategy with the lowest risk is optimized to achieve intelligent decision generation and improve the effectiveness and accuracy of the decision. Through the human-computer interaction interface, it supports real-time manual monitoring and intervention in the decision-making and control execution process, giving full play to the advantages of human-computer collaboration and improving the interpretability and security of the system.

[0049] Example 2 Reference Figure 1~Figure 5 , which is the second embodiment of the present invention, and this embodiment provides an intelligent well construction safety management and control system. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0050] From the actual operation database of 160 wells in the Fuling shale gas field in the Sichuan Basin, 80,000 original records were extracted through the well site monitoring data acquisition system, including well construction parameter data, inclination measurement data, well construction fluid performance data, etc., covering normal well construction, abnormal events, various risk scenarios, etc. These original data are directly recorded by the data acquisition system and have high authenticity and integrity.

[0051] Furthermore, the data acquisition and fusion module mines 32 types of feature data highly related to well construction risks from the original data, such as well construction fluid density, pump pressure, torque value, well inclination, etc. The self-attention mechanism is used to give each feature an adaptive weight, and interpolation is used to complete the missing data. Finally, a normalized time series data set containing 65,000 samples with a time step of 5 minutes was constructed. The data set is divided into a training set (52,000 samples) and a test set (13,000 samples) in a ratio of 8:2.

[0052] Furthermore, three types of data augmentation (time perturbation augmentation, feature mask augmentation, and semantic preservation augmentation) were performed on the training set to expand the training set to 156,000 augmented samples. The risk assessment model was trained for 6 rounds of iterations on this scale dataset, with a total training cycle of about 76 hours. During the training process, the contrast loss dropped from the initial 2.7 to 0.68, the generation loss dropped from 6.1 to 1.9, and the classification loss dropped from 2.5 to 0.32. On the test set, the model evaluation indicators showed that the absolute mean error of the risk situation score was 0.072, and the F1 value of the risk classification reached 95.8%, with good generalization performance.

[0053] Furthermore, after deploying the risk assessment model, 8 real well construction risk cases were randomly selected from the test set, including abnormal fracturing pressure, well drift, and well construction fluid contamination. The risk assessment model can issue effective warnings for these risk events 1.8 hours in advance on average, and the accuracy of risk level assessment is 90%. Among them, in the case of abnormal fracturing pressure in a certain well, the model warned 4 hours in advance that the well section would have a level III fracturing risk, and gave a confidence score of 91%. The on-duty well construction consultant took emergency measures such as reducing drilling speed and increasing drilling density in a timely manner based on the warning, successfully avoiding the occurrence of fracturing accidents.

[0054] Furthermore, when another well encountered the risk of well drift during the well construction operation, the decision engine module received the risk level judgment (level IV) and risk situation prediction data from the risk assessment module, and retrieved the parameter statistics of the current well section, standard operating procedure template, 3 similar cases and safety window constraints from the knowledge base. Two candidate strategies were generated through the machine learning algorithm, with corresponding predicted risk scores of 0.41 and 0.58 respectively. Finally, the strategy with a lower score was selected, and it was recommended to adjust the key parameters. The decision plan was implemented after approval by the on-site chief well construction engineer.

[0055] Furthermore, the control execution module converts the decision plan into a series of control instructions, and sends them to the drilling rig, mud tank and other equipment in real time through the well construction data network. The monitoring data of the control execution process is displayed in real time on the human-computer interaction interface. After about 25 minutes, all parameters were adjusted as expected. According to the risk assessment model, the risk score at this time has dropped from 0.83 to 0.32, which is within the controllable range. The decision experience and full data records are added to the knowledge base, providing new samples for subsequent model training.

[0056] Preferably, the comparison indicators of the present invention and the traditional method are shown in Table 1.

[0057] Table 1 Comparison indexes of the present invention and the traditional method index The present invention Traditional methods Data fusion capability Using self-attention mechanism to fuse multi-source heterogeneous data Manual rule splicing Risk assessment accuracy 90% 65% Risk warning lead time (average value) 4 0.5 Decision-making optimization efficiency 61% reduction 35% reduction Human-machine collaboration capabilities Support natural language manual intervention and closed-loop optimization Human-machine collaboration and lack of optimization mechanism Generalization The risk situation score error is 0.072, and the F1 value is 95.8%. Poor performance on the test set and difficult to generalize Specifically, it can be seen from Table 1 that the present invention is significantly superior to traditional manual decision-making and rule model methods in terms of risk assessment accuracy, risk warning lead time, decision optimization efficiency, etc., and has good application prospects.

[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent well construction safety management and control system, characterized in that: include: The data acquisition and fusion module is used to collect various raw data in the well construction process, pre-process the raw data, construct a standardized time series data sample set, and input the time series data sample set into the risk assessment module; The risk assessment module builds a risk assessment model based on the Transformer encoder-decoder framework, receives the time series data sample set output by the data acquisition and fusion module, evaluates the potential risk of well construction in real time, and inputs the evaluation results into the decision engine module and the human-computer interaction module. The evaluation results include risk situation sequence and risk level; The decision engine module is used to send a query request to the decision knowledge base module, build a decision strategy generation model based on the risk situation sequence and risk level output by the risk assessment module and the query results, generate high-level decision instructions and parameter adjustment sequences, input the high-level decision instructions to the human-computer interaction module, and input the parameter adjustment sequence to the control execution module; The control execution module is used to receive the parameter adjustment sequence output by the decision engine module, convert the parameter adjustment sequence into a corresponding control instruction sequence, and send the control instruction sequence to the actual well construction control system for execution through the data communication interface; Human-computer interaction module, including a visualization interface and a natural language interaction interface. The visualization interface is used to display risk assessment results, decision results and control execution status. The natural language interaction interface supports manual intervention in decision-making and feeds back manual intervention instructions to the risk assessment module, decision engine module or optimization control module; The decision knowledge base module is used to build a decision knowledge base, receive query requests from the decision engine module, retrieve relevant data from the knowledge base according to the query requests, and input the retrieval results into the decision engine module.

2. The intelligent well construction safety management and control system according to claim 1, characterized in that: The operation process of the data acquisition and fusion module is as follows: Collect various raw data during the well construction process and construct the original well construction process data set; Select characteristic data related to well construction risk assessment from the original well construction process data set; A self-attention mechanism is used to fuse the selected feature data and to assign adaptive weights to each feature based on its importance to risk assessment; Perform missing value processing on the fused feature data, detect features with missing values, and use historical data interpolation of similar features to fill in the missing values; Perform feature encoding on the processed feature data set, perform One-Hot encoding on discrete features, and perform normalized encoding on continuous features; According to the time attribute of the feature, the encoded feature data is divided into three categories: time series feature, discrete feature and continuous feature; Perform sliding window segmentation on the time series features, set the window size, and obtain the time series feature sequence; The segmented time series feature sequence is concatenated with the discrete features and continuous features in the corresponding window to construct multivariate feature data; The concatenated multivariate feature data is formatted into a time series data sample set, and the formatted time series data sample set is input into the risk assessment module.

3. The intelligent well construction safety management and control system according to claim 1, characterized in that: The operation process of the risk assessment module is as follows: Receive the time series data sample set output by the data acquisition and fusion module, and divide the time series data sample set into a training set and a test set according to a preset ratio; Augmenting the training set data to obtain an augmented data set, wherein the augmentation includes temporal perturbation augmentation, feature mask augmentation, and semantic preservation augmentation; Design a Transformer-based encoder-decoder model framework as a risk assessment model; A self-supervised contrastive learning strategy of a masked language model is used to train the risk assessment model and optimize the parameters of the encoder and decoder at the same time; During the training process, the current model performance is periodically evaluated on the test set. If the performance meets the predetermined conditions, the training is stopped early. Otherwise, the training continues until the maximum number of training rounds is reached. The generalization performance of the trained risk assessment model is fully evaluated on the test set. If the preset evaluation indicators are met, the model is deployed. Using the deployed risk assessment model, risk assessment is performed on the real-time well construction data stream to obtain the risk situation sequence and risk level; Input the risk situation sequence into the human-computer interaction module, receive feedback instructions, and determine whether the risk assessment model needs to be retrained; If necessary, new well construction process data is collected, retrained and evaluated. If not necessary, the risk situation sequence and risk level are input into the decision engine module.

4. The intelligent well construction safety management and control system according to claim 3, characterized in that: The framework of the risk assessment model includes an encoder part and a decoder part, the encoder part includes a temporal encoder, a risk relationship modeler and an attention fusion module, and the decoder part includes a first decoding head and a second decoding head.

5. The intelligent well construction safety management and control system according to claim 3, characterized in that: The training of the risk assessment model comprises the following steps: Get the original training set and augmented dataset; Sample a batch of data X and X' from the original training data and the augmented dataset respectively; Randomly mask parts of X and X' to get masked inputs X_masked and X'_masked; X_masked and X'_masked are input into the encoder respectively, and the encoder outputs the corresponding time series feature representation and ; Compute the contrastive loss of the encoder output ; Will and Input the first decoding head and the second decoding head of the decoder respectively, The first decoding head outputs the predicted sequences Y_pred and Y'_pred, and calculates the generation loss of Y_pred and the original label Y ; The second decoding head outputs the classification probabilities P and P' and calculates the classification loss between P and the original label C ; The contrast loss , generate loss , classification loss Perform weighted summation to get the total loss ; Based on total loss , calculate the gradient of the training parameters, and use the gradient descent algorithm to back-propagate and optimize the training parameters; The contrast loss The specific formula is as follows: in, and Represent the output representation of the encoder for the original input X and the augmented input X', respectively. for The transpose of is the L2 norm, is a scalar.

6. The intelligent well construction safety management and control system according to claim 1, characterized in that: The operation process of the decision engine module is as follows: Receive the risk situation sequence and risk level output by the risk assessment module; Sending a query request to the decision knowledge base module, the query conditions include risk situation sequence, risk level and current well construction stage, and receiving the query results returned by the decision knowledge base module, the query results include parameter data subsets, operation process templates, historical risk cases and safety constraints; Build a decision strategy generation model, take the query results as input, and generate a series of candidate decision strategy sets based on machine learning algorithms; For each decision strategy in the candidate strategy set, evaluate its expected risk level and select the decision strategy with the lowest risk as output; Decompose the decision strategy into two parts: well construction parameter adjustment plan and high-level decision instructions, and input the high-level decision instructions into the human-computer interaction module for manual review; If the manual feedback decision is unreasonable, the modification instructions are input through the natural language interactive interface, and the modification instructions are fed back to the decision engine module to regenerate the decision strategy; If the manual feedback decision is reasonable, The well construction parameter adjustment plan is combined with the real-time well construction parameters, and the model predictive control algorithm is used to calculate the optimized parameter adjustment sequence and input it into the control execution module.

7. The intelligent well construction safety management and control system according to claim 1, characterized in that: The operation flow of the control execution module is as follows: Receiving the optimization parameter adjustment sequence output by the decision engine module; Parse the parameter adjustment sequence, identify the adjustment object and its adjustment target value; Obtain the current status data of the real-time well construction control system, including the status feedback of each actuator; Compare the parameter adjustment target value with the current mechanism state and calculate the incremental adjustment amount of control execution; Match the incremental adjustment amount with the underlying control command set of the well construction control system to generate a corresponding control instruction sequence; Through industrial Ethernet, fieldbus or other industrial communication protocols, the control instruction sequence is sent to the corresponding execution unit of the well construction control system in real time; Continuously receive execution feedback from the well construction control system to monitor whether parameters are adjusted as expected; If it is detected that the execution deviation exceeds the allowable range, the feedback data will be passed to the decision engine module to determine whether the decision strategy needs to be regenerated; The execution process is presented on the human-computer interaction interface, which supports manual monitoring and intervention. If an abnormality is found manually, the correction instructions are input into the control execution module through the natural language interaction interface; The control execution module regenerates and issues a control instruction sequence after receiving the correction instruction; If the parameter adjustment is completed and the execution result is normal, the final status feedback is passed to the decision engine module.

8. An intelligent well construction safety control method, based on the intelligent well construction safety control system according to any one of claims 1 to 7, characterized in that: Also includes, Collect various raw data during the well construction process, pre-process the raw data, and construct a standardized time series data sample set; A risk assessment model is built based on the Transformer encoder-decoder framework, which receives time series data sample sets and evaluates the potential risks of well construction in real time. Input risk assessment results into decision-making process and human-computer interaction interface; Send a query request to the decision knowledge base, retrieve relevant data from the decision knowledge base according to the query request, and use the retrieval results for decision generation; Based on the risk assessment results and the query results of the decision knowledge base, a decision strategy generation model is constructed to generate high-level decision instructions and parameter adjustment sequences; Input high-level decision instructions into the human-computer interaction interface and input parameter adjustment sequences into the control execution link; The visual interface of the human-computer interaction interface displays risk assessment results, decision results and control execution status; The natural language interface of the human-computer interaction interface supports manual intervention decisions and feeds intervention instructions back to the risk assessment, decision generation or control execution links; Receive the parameter adjustment sequence output by the decision-making generation link, convert it into a control instruction sequence, and send it to the actual well construction control system for execution through the data communication interface.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the intelligent well construction safety management and control system described in any one of claims 1 to 7 are implemented.

10. 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 intelligent well construction safety management and control system described in any one of claims 1 to 7 are implemented.

Images