Adaptive window drilling condition identification method with event uncertainty gating

Abstract

The application provides an event uncertainty gated adaptive window drilling operation condition identification method, which can acquire current data of a field sensor to determine an observation node of a position characteristic layer, then calculate a trend node of a motion characteristic layer according to a data sequence of a current sliding window, determine an event layer posterior probability distribution and an event layer uncertainty index according to a drilling operation condition identification Bayesian network, adjust the length of the sliding window according to the event layer uncertainty index and redetermine the motion characteristic layer, correct the event layer posterior probability distribution according to the updated motion characteristic layer and determine a state layer posterior probability distribution, and determine a state layer output according to the state layer posterior probability distribution and the event layer uncertainty index, so that the window is expanded to enhance stability when the event layer posterior probability distribution is uncertain, the window is shortened to improve real-time performance when the event layer posterior probability distribution is clear, and the accuracy of drilling operation condition identification and the real-time performance of abnormal event early warning are improved.

Description

Technical Field

[0001] This invention relates to the field of drilling engineering technology, and in particular to an adaptive window drilling condition identification method with event uncertainty gating. Background Technology

[0002] Drilling operations are characterized by multi-parameter coupling, frequent switching of operating conditions, strong uncertainty of the underground environment, and high safety risks. In order to provide early warning of complex downhole conditions and reduce the occurrence of drilling accidents, it is necessary to accurately identify drilling conditions.

[0003] Existing drilling condition identification schemes mainly include expert systems based on threshold rules, classification models based on machine learning, time series models based on hidden Markov models or dynamic Bayesian networks, and pseudo-dynamic Bayesian network models.

[0004] Threshold-based expert systems offer strong interpretability, but rule maintenance is costly and they struggle to express probabilistic dependencies between multiple parameters. Deep learning-based models can improve classification accuracy, but their inference process is opaque, making it difficult to meet the auditability and causal explanation requirements of safety-critical industrial scenarios. Temporal models based on Hidden Markov Models or Dynamic Bayesian Networks can describe certain temporal dependencies, but they face limitations in terms of multi-source sensor conditional dependencies, real-time inference complexity, and edge deployment.

[0005] The pseudo-dynamic Bayesian network scheme, which combines Naive Bayesian networks, state transition probabilities, likelihood probabilities, and attribute weight matrices, typically transforms the temporal relationship between adjacent time points into state transition probabilities. It then trains the condition recognition model by combining likelihood probabilities and attribute weight matrices. This approach can solve the model generalization problem by optimizing attribute weights or category attribute contributions during training. However, it faces the challenge of simultaneously ensuring real-time performance and stability in transitional conditions, the initial stage of abnormal symptoms, and short-term sensor fluctuation scenarios.

[0006] In addition, pseudo-dynamic Bayesian network solutions usually use the sliding window length as a preset parameter, or adjust the window only according to the degree of data fluctuation. When the sliding window is set too short, it is sensitive to noise and prone to misjudgment when drilling pressure fluctuates instantaneously, torque spikes or pump pressure shakes. When the sliding window is too long, it has problems such as delayed response to working condition switching, delayed warning of abnormal events such as stuck drill bit, well leakage, and overload.

[0007] Therefore, a new technological solution is needed that can improve the stability of transition condition identification and the real-time performance of abnormal event early warning while maintaining the interpretability of Bayesian networks. Summary of the Invention

[0008] One objective of this application is to provide an adaptive window drilling condition identification method with event uncertainty gating, which addresses the problems in existing technologies such as the difficulty in balancing real-time performance and stability with fixed sliding windows, the lack of event layer feedback adjustment mechanism in pseudo-dynamic Bayesian models, the susceptibility to jitter in transition condition identification, and insufficient early warning of abnormal events.

[0009] To achieve the above objectives, some embodiments of this application provide an adaptive window drilling condition identification method with event uncertainty gating, the method comprising:

[0010] The current data from multiple field sensors are acquired and discretized to determine the observation nodes of the location feature layer. The observation nodes include a combination of two or more of the following: mechanical drilling rate node, hook load node, drill pressure node, torque node, standpipe pressure node, pump flushing node, and well depth node.

[0011] The corresponding trend states are calculated and discretized based on the data sequences of the current sliding windows corresponding to multiple field sensors to determine the trend nodes of the motion feature layer. The trend states include one or more combinations of the following: relative rate of change, window slope and window fluctuation intensity. The trend nodes include mechanical drilling rate trend nodes, drilling pressure trend nodes, torque trend nodes and riser pressure trend nodes.

[0012] Based on a preset drilling condition identification Bayesian network, the posterior probability distribution of the event layer is determined. The drilling condition identification Bayesian network is a directed acyclic graph including a location feature layer, a motion feature layer, an event layer, and a state layer. The event layer includes normal nodes, stuck pipe warning nodes, leakage loss sign nodes, overload risk nodes, and drill bit wear nodes.

[0013] Based on the posterior probability distribution of the event layer, an event layer uncertainty index is determined, wherein the event layer uncertainty index includes one or more combinations of the following: the maximum posterior probability of the event layer, the event layer entropy value, and the difference between the two largest probabilities of the event layer;

[0014] The length of the sliding window is adjusted according to the event layer uncertainty index, and the motion feature layer is redefined based on the adjusted sliding window.

[0015] Based on the updated motion feature layer, the posterior probability distribution of the event layer is corrected and the posterior probability distribution of the state layer is determined;

[0016] The state layer output is determined based on the posterior probability distribution of the state layer and the uncertainty index of the event layer, wherein the state layer output includes at least the drilling conditions.

[0017] Furthermore, the relative rate of change in the trend state is calculated based on the data sequence of the current sliding window corresponding to the multiple field sensors, expressed by the formula:

[0018] ,

[0019] Among them, field sensors j At any moment t The corresponding data sequence of the current sliding window is , , For field sensors j At any moment t The relative rate of change To prevent division by zero of positive numbers, For a moment t The corresponding length of the current sliding window;

[0020] The window slope in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, expressed by the formula:

[0021] ,

[0022] in, For field sensors j At any moment t The window slope, This represents the average value of the sampling point indices within the current sliding window. For field sensors j The average of all sampled values ​​within the current sliding window;

[0023] The intensity of window fluctuations in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, expressed by the formula:

[0024] ,

[0025] in, For field sensors j At any moment t The intensity of window fluctuations.

[0026] Furthermore, based on the preset drilling conditions, a Bayesian network is identified to determine the posterior probability distribution of the event layer, expressed by the formula:

[0027] ,

[0028] in, This represents the posterior probability distribution of the event layer. Evidence for the location feature layer, Evidence for the motion feature layer, For event layer states, e This represents the current state of the event layer. This is an index variable used to traverse all states of the event layer. Evidence at the motion feature layer When it is known, the event layer is a state. e The conditional probability, For evidence at the location feature layer Knowing the motion feature layer is evidence The conditional probability, Evidence for the location feature layer The prior probability, This is the event layer state space;

[0029] The normalized posterior probability distribution of the event layer is expressed by the formula:

[0030] ,

[0031] in, It is proportional to the sign.

[0032] Furthermore, the posterior probability distribution of the state layer is determined, expressed by the formula:

[0033] ,

[0034] in, This represents the posterior probability distribution of the state layer. For the event layer state space, This refers to the state-level operating condition.

[0035] Furthermore, the maximum posterior probability of the event layer is expressed by the formula:

[0036] ,

[0037] in, This represents the maximum posterior probability of the event layer. Evidence for the location feature layer, Evidence for the motion feature layer, For event layer states, For the event layer state space, This represents the posterior probability distribution of the event layer.

[0038] The event layer entropy value is expressed by the formula:

[0039] ,

[0040] in, This represents the event layer entropy value.

[0041] The difference between the two largest probabilities of the event layer is expressed by the formula:

[0042] ,

[0043] in, The difference between the two largest probabilities at the event level. and These are the maximum and second largest values ​​in the posterior probability of the event layer, respectively.

[0044] Furthermore, based on the posterior probability distribution of the event layer, the uncertainty index of the event layer is determined, expressed by the formula:

[0045] ,

[0046] in, As an indicator of event-level uncertainty, , , These are the weighting coefficients. ,and .

[0047] Furthermore, the length of the sliding window is adjusted according to the event layer uncertainty index, expressed by the formula:

[0048] ,

[0049] in, For a moment t The corresponding length of the current sliding window, For a moment t The length of the sliding window at the next moment, As an indicator of event-level uncertainty, The threshold for the uncertainty of the target. For adjustment coefficients, As the baseline window length, and These are the lower and upper limits of the window length, respectively. This is a limiting function;

[0050] Or it can be expressed as a formula:

[0051] ,

[0052] in, The adjustment step size for the window length each time. For low uncertainty threshold, For high uncertainty threshold, and These are the lower and upper limits of the window length, respectively.

[0053] Furthermore, the state layer output is determined based on the posterior probability distribution of the state layer and the uncertainty index of the event layer, expressed by the formula:

[0054] ,

[0055] in, This represents the maximum posterior probability of the state layer. , For the state layer state space, This is a state-level operating condition. As an indicator of event-level uncertainty, This is the confidence threshold for the state layer. This represents the uncertainty threshold for the event layer. Used to select posterior probabilities among all operating conditions s in the state layer. Maximum operating conditions.

[0056] Some embodiments of this application also provide an adaptive window drilling condition identification system with event uncertainty gating, characterized in that the system includes:

[0057] The location feature generation module is used to acquire and discretize the current data of multiple field sensors to determine the observation nodes of the location feature layer. The observation nodes include two or more of the following combinations: mechanical drilling rate node, hook load node, drilling pressure node, torque node, standpipe pressure node, pump flushing node, and well depth node.

[0058] The motion feature generation module is used to calculate and discretize the corresponding trend states based on the data sequences of the current sliding windows corresponding to multiple field sensors, and determine the trend nodes of the motion feature layer. The trend states include one or more combinations of the following: relative rate of change, window slope and window fluctuation intensity. The trend nodes include mechanical drilling rate trend nodes, drilling pressure trend nodes, torque trend nodes and riser pressure trend nodes.

[0059] The Bayesian network inference module is used to identify a Bayesian network based on preset drilling conditions and determine the posterior probability distribution of the event layer. The drilling condition identification Bayesian network is a directed acyclic graph including a position feature layer, a motion feature layer, an event layer, and a state layer. The event layer includes normal nodes, stuck pipe warning nodes, well leakage sign nodes, overload risk nodes, and drill bit wear nodes.

[0060] The event layer uncertainty calculation module is used to determine the event layer uncertainty index based on the event layer posterior probability distribution. The event layer uncertainty index includes one or more combinations of the following: the maximum posterior probability of the event layer, the event layer entropy value, and the difference between the two largest probabilities of the event layer.

[0061] A sliding window adjustment module is used to adjust the length of the sliding window according to the event layer uncertainty index, and to redetermine the motion feature layer according to the adjusted sliding window;

[0062] The event layer probability update module is used to correct the event layer posterior probability distribution and determine the state layer posterior probability distribution based on the updated motion feature layer.

[0063] The state layer output module is used to determine the state layer output based on the state layer posterior probability distribution and the event layer uncertainty index, wherein the state layer output includes at least the drilling conditions.

[0064] Some embodiments of this application also provide a computer-readable medium having computer-readable instructions stored thereon, which can be executed by a processor to implement the aforementioned adaptive window drilling condition identification method with event uncertainty gating.

[0065] Some embodiments of this application also provide an electronic device, which includes a memory for storing computer program instructions and a processor for executing the computer program instructions, wherein when the computer program instructions are executed by the processor, the electronic device performs the aforementioned adaptive window drilling condition identification method with event uncertainty gating.

[0066] Compared with existing technologies, the solution provided in this application can obtain the current data of multiple field sensors to determine the observation nodes of the position feature layer, and then calculate the corresponding trend state based on the data sequence of the current sliding window corresponding to multiple field sensors to determine the trend nodes of the motion feature layer. Based on the preset drilling conditions, the Bayesian network is used to determine the posterior probability distribution of the event layer, and the event layer uncertainty index is determined based on the event layer posterior probability distribution. Based on the event layer uncertainty index, the length of the sliding window is adjusted and the motion feature layer is re-determined based on the adjusted sliding window. Based on the updated motion feature layer, the event layer posterior probability distribution is corrected and the state layer posterior probability distribution is determined. Based on the state layer posterior probability distribution and the event layer uncertainty index, the state layer output is determined. This enables the expansion of the window or delay of the output to enhance stability when the event layer posterior probability distribution is uncertain, and the shortening of the window to improve real-time performance when the event layer posterior probability distribution is clear. It can improve the accuracy of drilling condition identification and the real-time performance of abnormal event warning while maintaining the interpretability of the Bayesian network, and reduce the risk of misjudgment in transitional conditions and the early stage of abnormal events. Attached Figure Description

[0067] Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0068] Figure 1A flowchart of an adaptive window drilling condition identification method with event uncertainty gating provided for some embodiments of this application.

[0069] Figure 2 A flowchart for extracting trend features from a sliding window is provided for some embodiments of this application.

[0070] Figure 3 This is a schematic diagram of a four-layer directed acyclic graph topology of a Bayesian network for drilling condition identification, provided for some embodiments of this application.

[0071] Figure 4 A closed-loop flowchart for adjusting a sliding window based on event-layer uncertainty feedback is provided for some embodiments of this application.

[0072] Figure 5 A state layer output gating flowchart is provided for some embodiments of this application. Detailed Implementation

[0073] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0074] Here, the adaptive window drilling condition identification method with event uncertainty gating in this application embodiment is suitable for scenarios where drilling conditions need to be identified in real time during drilling operations.

[0075] In this scenario, drilling operations involve frequent switching of drilling conditions and high safety risks. To ensure the safety and controllability of the drilling process, it is crucial to accurately identify the current drilling conditions and provide timely warnings for abnormal events. However, existing drilling condition identification schemes suffer from several problems, such as susceptibility to misjudgment and delayed warnings of abnormal events, making it difficult to guarantee the stability and real-time performance of drilling condition identification.

[0076] The adaptive window drilling condition identification method with event uncertainty gating provided in this application can acquire current data from multiple field sensors to determine the observation nodes of the position feature layer, calculate the corresponding trend state based on the data sequence of the current sliding window corresponding to multiple field sensors to determine the trend nodes of the motion feature layer, determine the posterior probability distribution of the event layer based on a preset drilling condition identification Bayesian network, determine the event layer uncertainty index based on the event layer posterior probability distribution, adjust the length of the sliding window based on the event layer uncertainty index, redetermine the motion feature layer based on the adjusted sliding window, correct the event layer posterior probability distribution based on the updated motion feature layer, determine the state layer posterior probability distribution, and determine the state layer output based on the state layer posterior probability distribution and the event layer uncertainty index. This allows for expanding the window or delaying the output to enhance stability when the event layer posterior probability distribution is uncertain, and shortening the window to improve real-time performance when the event layer posterior probability distribution is clear. It can improve the accuracy of drilling condition identification and the real-time performance of abnormal event warnings while maintaining the interpretability of the Bayesian network, and reduce the risk of misjudgment in transitional conditions and the initial stage of abnormal events.

[0077] Figure 1 The flowcharts of an adaptive window drilling condition identification method with event uncertainty gating, executed by an electronic device in some embodiments of this application, are shown. The electronic device is the execution subject of this method. Figure 1 As shown, the method may include the following steps:

[0078] Step S101: Obtain the current data from multiple field sensors and discretize them to determine the observation nodes of the location feature layer.

[0079] Electronic devices may include, but are not limited to, laptop computers, desktop computers, tablet computers, mobile phones, wearable devices, head-mounted displays, servers, mobile email devices, portable game consoles, portable music players, e-reader devices, televisions in which one or more processors are embedded or coupled, or other electronic devices capable of accessing networks.

[0080] Here, the field sensors include a variety of sensors used to collect different types of data during the drilling process. The collected data may include, but is not limited to: Rate of Penetration (ROP), Hookload, Weight On Bit (WOB), Torque, Stand Pipe Pressure (SPP), Strokes Per Minute (SPM), and Depth. These data exhibit different combinations and variation patterns under different drilling conditions.

[0081] Accordingly, the observation nodes of the location feature layer may include two or more combinations of the following: mechanical drilling rate node, hook load node, drilling pressure node, torque node, riser pressure node, pump flushing node, and well depth node, etc.

[0082] This application embodiment uses a drilling condition identification Bayesian network to identify drilling conditions. The drilling condition identification Bayesian network in this application embodiment includes a position feature layer, a motion feature layer, an event layer, and a state layer. The position feature layer is used to represent the discrete sensor state at the current sampling time. The motion feature layer is used to represent the trend state within the sliding window. The event layer is used to represent abnormal events such as stuck drill warning, lost circulation, overload risk, and drill bit wear. The state layer is used to represent drilling conditions such as rotary drilling, sliding drilling, single connection, circulation, stationary, tripping, and running.

[0083] In some embodiments of this application, the current data collected by the field sensors can be discretized to determine the discretized value for each observation node. The discretized values ​​can be, for example, {low, normal, high, abnormal}.

[0084] Step S102: Calculate the corresponding trend state based on the data sequence of the current sliding window corresponding to multiple field sensors and discretize it to determine the trend node of the motion feature layer.

[0085] Here, each field sensor corresponds to a sliding window for data acquisition, and the data sequence in each sliding window includes the data acquired at the current moment and data from multiple historical moments acquired previously.

[0086] In some embodiments of this application, the trend state may include one or more combinations of the following: relative rate of change, window slope, and window fluctuation intensity, etc.

[0087] In some embodiments of this application, the relative rate of change in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, which can be expressed by the following formula:

[0088] ,

[0089] Among them, field sensors j At any moment t The corresponding data sequence of the current sliding window is , , For field sensors j At any moment t The relative rate of change To prevent division by zero of positive numbers, For a moment t The corresponding length of the current sliding window.

[0090] In some embodiments of this application, the window slope in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, which can be expressed by the following formula:

[0091] ,

[0092] in, For field sensors j At any moment t The window slope, This represents the average value of the sampling point indices within the current sliding window. , For field sensors j The average of all sampled values ​​within the current sliding window. .

[0093] In some embodiments of this application, the window fluctuation intensity in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, which can be expressed by the following formula:

[0094] ,

[0095] in, For field sensors j At any moment t The intensity of window fluctuations.

[0096] Here, trend nodes may include, but are not limited to: mechanical drilling rate trend nodes, drilling pressure trend nodes, torque trend nodes, and riser pressure trend nodes. Each trend node corresponds to the trend state of the sliding window of the field sensor; for example, the mechanical drilling rate trend node corresponds to the trend state of the sliding window of the speed sensor.

[0097] In some embodiments of this application, the calculated trend state can be discretized to determine the discretized value of each trend node. The discretized values ​​can be, for example, {rapid decline, slow decline, stable, slow rise, rapid rise, increased volatility}.

[0098] Taking drill pressure (WOB) as an example, let the current window length be... If the sampling frequency is 1 Hz and the window includes 60 sampling points, the formula for calculating the relative rate of change in the trend state is as follows:

[0099] .

[0100] Examples of rules for discretizing relative rates of change are shown in Table 1 below.

[0101] Table 1

[0102] in, A value of 0.10 to 0.20 is acceptable. A value of 0.03 to 0.08 is acceptable. It can be determined based on the fluctuation distribution under historical normal operating conditions.

[0103] Figure 2 The flowchart of sliding window trend feature extraction is shown in some embodiments of this application.

[0104] Step S103: Identify the Bayesian network based on the preset drilling conditions and determine the posterior probability distribution of the event layer.

[0105] Here, the Bayesian network for drilling condition identification is a directed acyclic graph topology consisting of a location feature layer, a motion feature layer, an event layer, and a state layer. Each layer includes more than one node, and there are directed edges between layers.

[0106] The four-layer directed acyclic graph topology of the Bayesian network for drilling condition identification is denoted as:

[0107]

[0108] Node set Represented as:

[0109]

[0110] in, Represents the set of nodes in the location feature layer. Represents the set of nodes in the motion feature layer. Represents the set of event layer nodes. This represents the set of state layer nodes.

[0111] Directed edge set Represented as:

[0112]

[0113] in, This represents the dependency relationship from the position feature layer to the motion feature layer. This represents the dependency relationship from the motion feature layer to the event layer. This indicates the dependency relationship from the event layer to the state layer. This represents the auxiliary dependency relationship from the location feature layer to the state layer.

[0114] An example of a four-layer node definition for a Bayesian network for drilling condition identification is shown in Table 2 below.

[0115] Table 2

[0116] Figure 3The following illustrates a four-layer directed acyclic graph topology of a Bayesian network for drilling condition identification in some embodiments of this application.

[0117] The directed edge between the location feature layer and the motion feature layer describes the conditional constraints of the current parameter level on the interpretation of the trend. For example, the same upward trend in drill pressure has different risk implications under low and high drill pressure conditions. The directed edge between the motion feature layer and the event layer describes the probability of triggering abnormal events by the trend combination. The directed edge between the event layer and the location feature layer, pointing to the state layer, describes the combined impact of abnormal events and the current parameter combination on the final working condition judgment.

[0118] Here, the event layer is used to calculate the probability of each node in the event layer based on the combination of various trend nodes in the motion feature layer. The nodes in the event layer may include, but are not limited to: normal nodes, stuck drill warning nodes, leakage warning nodes, overload risk nodes, and drill bit wear nodes. The event layer is an intermediate hidden variable layer used to receive abnormal events.

[0119] Here, the posterior probability of nodes in the event layer is determined by the pre-defined conditional probabilities of each layer. For example, the conditional probability representation from the motion feature layer to the event layer is shown in Table 3 below.

[0120] Table 3

[0121] In practice, conditional probabilities can be determined by expert priors, historical labeled samples, or Bayesian estimation.

[0122] In some embodiments of this application, the conditional probability table can be initialized by expert prior knowledge and updated using historical samples.

[0123] Let the parameters of the conditional probability table be... Historically labeled samples are Then, Bayesian update can be used, expressed by the following formula:

[0124] .

[0125] For discrete conditional probability tables, Dirichlet priors can be used, expressed by the following formula:

[0126] , in, Indicates that the parent node combination is i The current node value is k Sample count, For expert priori counting.

[0127] In the embodiments of this application, only the local conditional probability table related to the current well type, formation interval or drill string combination is updated, without retraining the global weight matrix, thereby reducing field deployment costs and avoiding overfitting.

[0128] In some embodiments of this application, a Bayesian network is identified based on preset drilling conditions to determine the posterior probability distribution of the event layer, which can be expressed by the following formula:

[0129] ,

[0130] in, This represents the posterior probability distribution of the event layer. Evidence for the location feature layer, Evidence for the motion feature layer, For event layer states, e This represents the current state of the event layer. This is an index variable used to traverse all states of the event layer. Evidence at the motion feature layer When it is known, the event layer is a state. e The conditional probability, For evidence at the location feature layer Knowing the motion feature layer is evidence The conditional probability, Evidence for the location feature layer The prior probability, This is the event layer state space.

[0131] Here, For the discretized values ​​of nodes in the location feature layer, The values ​​are discretized for the nodes in the motion feature layer. This represents the value of a node in the event layer. Event layer state space. ={Normal, Stuck drill warning, Well leakage signs, Overload risk, Drill bit wear}.

[0132] In actual calculations, due to and As observational evidence, the discretization results of the current position feature layer and the trend results of the motion feature layer have been determined by the field data and used as input for Bayesian network inference. This can normalize the posterior probability distribution of the event layer. The purpose of normalization is to convert the inference scores corresponding to each candidate event state into a probability distribution so that the sum of the probabilities of all event states is 1.

[0133] The normalized posterior probability distribution of the event layer can be expressed by the following formula:

[0134] ,

[0135] in, It is proportional to the sign.

[0136] Step S104: Determine the event layer uncertainty index based on the posterior probability distribution of the event layer.

[0137] Here, the event-level uncertainty index can include one or more combinations of the following: the event-level maximum posterior probability, the event-level entropy value, and the difference between the two largest probabilities of the event level.

[0138] In some embodiments of this application, the maximum posterior probability of the event layer can be expressed by the following formula:

[0139] ,

[0140] in, This represents the maximum posterior probability of the event layer. Evidence for the location feature layer, Evidence for the motion feature layer, For event layer states, For the event layer state space, This represents the posterior probability distribution of the event layer.

[0141] In some embodiments of this application, the event layer entropy value can be expressed by the following formula:

[0142] ,

[0143] in, This is the entropy value of the event layer.

[0144] In some embodiments of this application, the difference between the two largest probabilities of the event layer can be expressed by the following formula:

[0145] ,

[0146] in, The difference between the two largest probabilities at the event level. and These are the maximum and second largest values ​​in the posterior probability of the event layer, respectively.

[0147] In some embodiments of this application, the event-layer uncertainty index is determined based on the posterior probability distribution of the event layer, and can be expressed by the following formula:

[0148] ,

[0149] in, As an indicator of event-level uncertainty, , , These are the weighting coefficients. ,and .

[0150] Step S105: Adjust the length of the sliding window according to the event layer uncertainty index, and redetermine the motion feature layer according to the adjusted sliding window.

[0151] In some embodiments of this application, adjusting the length of the sliding window based on an event-layer uncertainty index may include the following:

[0152] When the event-layer uncertainty index is below the first threshold, the sliding window length is shortened; when the event-layer uncertainty index is above the second threshold, the sliding window length is increased; when the event-layer uncertainty index is between the first and second thresholds, the sliding window length remains unchanged. Here, the first and second thresholds are determined based on the actual requirements of the event-layer uncertainty index, such as engineering experience.

[0153] In some embodiments of this application, the length of the sliding window is adjusted according to the event-layer uncertainty index, which can be expressed by the following formula:

[0154] ,

[0155] in, For a moment t The corresponding length of the current sliding window, For a moment t The length of the sliding window at the next moment, As an indicator of event-level uncertainty, The threshold for the uncertainty of the target. For adjustment coefficients, As the baseline window length, and These are the lower and upper limits of the window length, respectively. This is the amplitude limiting function.

[0156] In some embodiments of this application, adjusting the length of the sliding window based on the event-layer uncertainty index can also be expressed by the following formula:

[0157] ,

[0158] in, The adjustment step size for the window length each time. For low uncertainty threshold, For high uncertainty threshold, and These are the lower and upper limits of the window length, respectively.

[0159] For example, 15 seconds or the corresponding number of sampling points; below Values ​​higher than 1 indicate clear event criteria and allow for window shrinking; values ​​higher than 1 indicate a more specific event. The value indicates that the event judgment is unclear, and the window can be enlarged; the length of the window should be rounded to an integer number of sampling points according to the sampling period.

[0160] Here, when the event-level uncertainty index is low, it indicates that the abnormal event is clearly identified, and the system shortens the window to improve the response speed. When the event-level uncertainty index is high, it indicates that the abnormal event is not identified, and the system expands the window to enhance the stability of trend estimation, or triggers delayed output.

[0161] The process for redefining the motion feature layer can be found in step S102, and will not be repeated here.

[0162] Figure 4 The following embodiments of this application illustrate a closed-loop process for adjusting the sliding window based on event-layer uncertainty feedback.

[0163] The following is an illustration using a specific example.

[0164] Event layer posterior probability distribution The probability of the largest event in this distribution is 0.30, and the difference between the two largest probabilities is 0.05. The calculated event layer entropy value is high, indicating that the event judgment is unclear and may be in a transitional state or a stage affected by sensor noise. In this case, the sliding window length should be increased to enhance the stability of trend estimation.

[0165] Table 4 below shows an example of a rule for adjusting the event-level uncertainty index and sliding window.

[0166] Table 4

[0167] The following is Second, Second, seconds, initial window The adaptive changes of the sliding window over consecutive time intervals are illustrated in Table 5 below, using seconds as an example.

[0168] Table 5

[0169] As shown in Table 5 above, the sliding window length is not a fixed parameter, nor is it adjusted solely based on fluctuations in the original signal. Instead, it is controlled by feedback from the event layer probability inference results. This mechanism makes the model more stable during transitional conditions and faster when the event is clear.

[0170] In some embodiments of this application, lightweight event transition constraints can be introduced at the event layer. Event transition refers to a change in event category between adjacent time points, such as a normal event transitioning to a stuck drill warning, or a stuck drill warning persisting. Lightweight event transition constraints utilize the probability distribution of event categories from the previous time point to provide a continuous prior for event inference at the current time point, without constructing a complete dynamic Bayesian network. This constraint can be used when the uncertainty at the event layer is at a medium to high level and when it is necessary to suppress event jumps caused by short-term noise, thereby enhancing the continuity of event judgment.

[0171] The event transition probability can be expressed by the following formula:

[0172] ,

[0173] in, This represents the event transition probability.

[0174] This event transition constraint is used to correct the posterior probability of the event layer, and can be expressed by the following formula:

[0175] ,

[0176] in, This is the corrected event layer probability distribution. For event transition probability, This represents the posterior probability distribution of the event layer.

[0177] An example of an event layer transition probability table is shown in Table 6 below.

[0178] Table 6

[0179] The event transition probability is not used to replace the full dynamic Bayesian network, but rather as a lightweight continuity constraint for the event layer. This constraint can work together with the event layer uncertainty index to adjust the sliding window and gate the state layer output.

[0180] Step S106: Based on the updated motion feature layer, correct the posterior probability distribution of the event layer and determine the posterior probability distribution of the state layer.

[0181] Here, the process of correcting the posterior probability distribution of the event layer based on the updated motion feature layer can be found in step S103, and will not be repeated here.

[0182] In some embodiments of this application, the determination of the posterior probability distribution of the state layer can be expressed by the following formula:

[0183] ,

[0184] in, This represents the posterior probability distribution of the state layer. For the event layer state space, This is a state-level operating condition. Given that the current event state is e and the location feature layer evidence is... The conditional probability of state s at time level. Evidence at the motion feature layer Given the conditional probability that the event layer is state e, For evidence at the location feature layer Given that the motion feature layer is The conditional probability, Evidence for the location feature layer The prior probability.

[0185] Used to characterize the dependency between the current parameter level and the window trend state. Used to characterize the probability of the current combination of sensor parameters appearing in historical samples or expert priors.

[0186] Step S107: Determine the state layer output based on the posterior probability distribution of the state layer and the uncertainty index of the event layer.

[0187] Here, the state layer output includes at least drilling conditions, and may also include abnormal event probabilities, confidence levels, and rejection results.

[0188] In some embodiments of this application, drilling conditions may include, but are not limited to: rotary drilling, sliding drilling, single-joint drilling, circulation, stationary drilling, tripping, and running-in drilling.

[0189] In some embodiments of this application, determining the state layer output based on the state layer posterior probability distribution and the event layer uncertainty index may include the following:

[0190] When the maximum a posteriori probability of the state layer is greater than the state layer confidence threshold and the event layer uncertainty index is less than the event layer uncertainty threshold, output the drilling condition corresponding to the maximum a posteriori probability; when the maximum a posteriori probability of the state layer is greater than the state layer confidence threshold but the event layer uncertainty index is greater than the event layer uncertainty threshold, output the transition condition; when the maximum a posteriori probability of the state layer is less than the state layer confidence threshold, output an undefined state or a manual confirmation prompt.

[0191] In some embodiments of this application, the state layer output is determined based on the posterior probability distribution of the state layer and the uncertainty index of the event layer, which can be expressed by the following formula:

[0192] ,

[0193] in, This represents the maximum posterior probability of the state layer. , For the state layer state space, This is a state-level operating condition. As an indicator of event-level uncertainty, This is the confidence threshold for the state layer. This represents the uncertainty threshold for the event layer. Used to select posterior probabilities among all operating conditions s in the state layer. Maximum operating conditions.

[0194] An example of a state layer output gating rule is shown in Table 7 below.

[0195] Table 7

[0196] Figure 5 The state layer output gating process is shown in some embodiments of this application.

[0197] In some embodiments of this application, event interpretation paths can also be established for multiple typical drilling conditions. An example of a combination of drilling conditions, trend nodes of motion feature layers, and event paths is shown in Table 8 below.

[0198] Table 8

[0199] As shown in Table 8, the embodiments of this application can not only output the final operating condition, but also the probability of abnormal events and the explanation path. When the uncertainty of the event layer increases, the embodiments of this application do not directly force classification, but reduce misjudgment through adaptive sliding windows and gating rules.

[0200] In some embodiments of this application, a drilling condition intelligent identification system is also provided, including the following modules:

[0201] 1. Data acquisition module, used to acquire multi-channel sensor data at the drilling site;

[0202] 2. Data preprocessing module, used to perform missing value handling, outlier handling, filtering, and normalization;

[0203] 3. Location feature generation module, used to discretize the current sampled values ​​and generate location feature layer observation nodes;

[0204] 4. Sliding window caching module, used to maintain the dynamic windows of each sensor channel;

[0205] 5. Motion feature extraction module, used to calculate the trend state based on the current window and generate motion feature layer nodes;

[0206] 6. A four-layer Bayesian network inference module, used to perform probabilistic inference at the event layer and state layer;

[0207] 7. Event layer uncertainty calculation module, used to calculate the event layer entropy value, maximum probability, and the difference between the two largest probabilities;

[0208] 8. Adaptive window adjustment module, used to adjust the window length for the next moment based on the uncertainty of the event layer;

[0209] 9. Status output gating module, used to output working conditions, transitional working conditions, undefined states or manual confirmation prompts based on the confidence level of the status layer and the uncertainty of the event layer;

[0210] 10. Online update module, used to update the conditional probability table and event transition probability table based on expert-annotated samples or historical samples.

[0211] The relationship between the drilling condition intelligent identification system module and its inputs and outputs is shown in Table 9 below.

[0212] Table 9

[0213] In addition, the solution of this application embodiment has at least the following beneficial effects:

[0214] 1. Forming a closed-loop adjustment mechanism: This application does not simply input the sliding window features into the Bayesian network, but uses the posterior uncertainty of the event layer to adjust the length of the sliding window in reverse, so that feature extraction and probabilistic inference form a closed loop.

[0215] 2. Balancing real-time performance and stability: When the event judgment is clear, shorten the window to improve the response speed of the working condition switching; when the event judgment is uncertain, expand the window to reduce misjudgments caused by noise and short-term fluctuations.

[0216] 3. Enhance the ability to identify transitional working conditions: Uncertainty at the event layer can reflect the intermediate state during the working condition switching process. Combined with state layer gating, it can output transitional working conditions or manual confirmation prompts to avoid forced misclassification.

[0217] 4. Enhance the interpretability of anomaly warnings: The event layer, as an intermediate hidden variable, can output the probability of abnormal events such as stuck drill bit, well leakage, overload, and drill bit wear, so that the results have an interpretive path from sensor trends to events and then to operating conditions.

[0218] 5. Unlike the pseudo-dynamic Bayesian scheme with attribute weight optimization: The focus of this application is not on optimizing the attribute weight matrix through negative conditional log-likelihood iteration, but on the window adaptive adjustment and state output gating driven by event layer uncertainty feedback.

[0219] 6. Suitable for edge deployment: This application can be implemented based on conditional probability tables and finite state reasoning, with computational complexity lower than large-scale deep learning models and complete dynamic Bayesian networks, making it suitable for real-time operation of edge computing equipment at drilling sites.

[0220] In summary, the solution provided in this application can acquire current data from multiple field sensors to determine the observation nodes of the position feature layer, then calculate the corresponding trend state based on the data sequence of the current sliding window corresponding to multiple field sensors to determine the trend nodes of the motion feature layer, determine the posterior probability distribution of the event layer based on the preset drilling condition identification Bayesian network, determine the event layer uncertainty index based on the event layer posterior probability distribution, adjust the length of the sliding window based on the event layer uncertainty index and redetermine the motion feature layer based on the adjusted sliding window, correct the event layer posterior probability distribution based on the updated motion feature layer and determine the state layer posterior probability distribution, and determine the state layer output based on the state layer posterior probability distribution and the event layer uncertainty index. This allows for expanding the window or delaying the output to enhance stability when the event layer posterior probability distribution is uncertain, and shortening the window to improve real-time performance when the event layer posterior probability distribution is clear. It can improve the accuracy of drilling condition identification and the real-time performance of abnormal event warnings while maintaining the interpretability of the Bayesian network, and reduce the risk of misjudgment in transitional conditions and the initial stage of abnormal events.

[0221] It should be noted that this application can be implemented in software and / or a combination of software and hardware, for example, using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In one embodiment, the software program of this application can be executed by a processor to implement the steps or functions described above. Similarly, the software program of this application (including related data structures) can be stored in a computer-readable recording medium, such as RAM memory, a magnetic or optical drive, a floppy disk, or similar devices. Furthermore, some steps or functions of this application can be implemented in hardware, for example, as circuitry that cooperates with a processor to perform the various steps or functions.

[0222] In a typical configuration of this application, both the terminal and the network device include one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0223] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0224] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include non-transitory computer-readable media, such as modulated data signals and carrier waves.

[0225] Furthermore, a portion of this application can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to this application through the operation of the computer. The program instructions invoking the methods of this application may be stored in a fixed or removable recording medium, and / or transmitted via a data stream in a broadcast or other signal carrying medium, and / or stored in the working memory of a computer device operating according to the program instructions. Here, one embodiment of this application includes a device comprising a memory for storing computer program instructions and a processor for executing the program instructions, wherein, when the computer program instructions are executed by the processor, the device is triggered to run methods and / or technical solutions based on the foregoing embodiments of this application.

[0226] It will be apparent to those skilled in the art that this application is not limited to the details of the exemplary embodiments described above, and that this application can be implemented in other specific forms without departing from the spirit or essential characteristics of this application. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of this application is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within this application. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in the apparatus claims may also be implemented by a single unit or device in software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.

Claims

1. An adaptive window drilling condition identification method with event uncertainty gating, characterized in that, The method includes: The current data from multiple field sensors are acquired and discretized to determine the observation nodes of the location feature layer. The observation nodes include a combination of two or more of the following: mechanical drilling rate node, hook load node, drill pressure node, torque node, standpipe pressure node, pump flushing node, and well depth node. The corresponding trend states are calculated and discretized based on the data sequences of the current sliding windows corresponding to multiple field sensors to determine the trend nodes of the motion feature layer. The trend states include one or more combinations of the following: relative rate of change, window slope and window fluctuation intensity. The trend nodes include mechanical drilling rate trend nodes, drilling pressure trend nodes, torque trend nodes and riser pressure trend nodes. Based on a preset drilling condition identification Bayesian network, the posterior probability distribution of the event layer is determined. The drilling condition identification Bayesian network is a directed acyclic graph including a location feature layer, a motion feature layer, an event layer, and a state layer. The event layer includes normal nodes, stuck pipe warning nodes, leakage loss sign nodes, overload risk nodes, and drill bit wear nodes. Based on the posterior probability distribution of the event layer, an event layer uncertainty index is determined, wherein the event layer uncertainty index includes one or more combinations of the following: the maximum posterior probability of the event layer, the event layer entropy value, and the difference between the two largest probabilities of the event layer; The length of the sliding window is adjusted according to the event layer uncertainty index, and the motion feature layer is redefined based on the adjusted sliding window. Based on the updated motion feature layer, the posterior probability distribution of the event layer is corrected and the posterior probability distribution of the state layer is determined; The state layer output is determined based on the posterior probability distribution of the state layer and the uncertainty index of the event layer, wherein the state layer output includes at least the drilling conditions.

2. The method according to claim 1, characterized in that, The relative rate of change in the trend state is calculated based on the data sequence of the current sliding window corresponding to the multiple field sensors, expressed by the formula: ， Among them, field sensors j At any moment t The corresponding data sequence of the current sliding window is , , For field sensors j At any moment t The relative rate of change To prevent division by zero of positive numbers, For a moment t The corresponding length of the current sliding window; The window slope in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, expressed by the formula: ， in, For field sensors j At any moment t The window slope, This represents the average value of the sampling point indices within the current sliding window. For field sensors j The average of all sampled values ​​within the current sliding window; The intensity of window fluctuations in the trend state is calculated based on the data sequence of the current sliding window corresponding to multiple field sensors, expressed by the formula: ， in, For field sensors j At any moment t The intensity of window fluctuations.

3. The method according to claim 1, characterized in that, Based on the preset drilling conditions, a Bayesian network is identified to determine the posterior probability distribution of the event layer, expressed by the formula: ， in, This represents the posterior probability distribution of the event layer. Evidence for the location feature layer, Evidence for the motion feature layer, For event layer states, e This represents the current state of the event layer. This is an index variable used to traverse all states of the event layer. For evidence at the motion feature layer When it is known, the event layer is a state. e The conditional probability, For evidence at the location feature layer Knowing the motion feature layer is evidence The conditional probability, Evidence for the location feature layer The prior probability, This is the event layer state space; The normalized posterior probability distribution of the event layer is expressed by the formula: ， in, It is proportional to the sign.

4. The method according to claim 3, characterized in that, The posterior probability distribution of the state layer can be determined by the following formula: ， in, This represents the posterior probability distribution of the state layer. For the event layer state space, This refers to the state-level operating condition.

5. The method according to claim 1, characterized in that, The maximum posterior probability of the event layer is expressed by the formula: ， in, This represents the maximum posterior probability of the event layer. Evidence for the location feature layer, Evidence for the motion feature layer, For event layer states, For the event layer state space, This represents the posterior probability distribution of the event layer. The event layer entropy value is expressed by the formula: ， in, This represents the entropy value of the event layer. The difference between the two largest probabilities of the event layer is expressed by the formula: ， in, The difference between the two largest probabilities at the event level. and These are the maximum and second largest values ​​in the posterior probability of the event layer, respectively.

6. The method according to claim 5, characterized in that, Based on the posterior probability distribution of the event layer, the uncertainty index of the event layer is determined, expressed by the formula: ， in, This is an indicator of uncertainty at the event level. , , These are the weighting coefficients. ,and .

7. The method according to claim 6, characterized in that, The length of the sliding window is adjusted based on the event layer uncertainty index, expressed by the formula: ， in, For a moment t The corresponding length of the current sliding window, For a moment t The length of the sliding window at the next moment, This is an indicator of uncertainty at the event level. The threshold for the uncertainty of the target. For adjustment coefficients, As the baseline window length, and These are the lower and upper limits of the window length, respectively. This is a limiting function; Or it can be expressed as a formula: ， in, The adjustment step size for the window length each time. For low uncertainty threshold, For high uncertainty threshold, and These are the lower and upper limits of the window length, respectively.

8. The method according to claim 5, characterized in that, The state layer output is determined based on the posterior probability distribution of the state layer and the uncertainty index of the event layer, expressed by the formula: ， in, This represents the maximum posterior probability of the state layer. , For the state layer state space, This is a state-level operating condition. This is an indicator of uncertainty at the event level. This is the confidence threshold for the state layer. This represents the uncertainty threshold for the event layer. Used to select posterior probabilities among all operating conditions s in the state layer. Maximum operating conditions.

9. A computer-readable medium having stored thereon computer-readable instructions that can be executed by a processor to implement the method as described in any one of claims 1 to 8.

10. An electronic device comprising a memory for storing computer program instructions and a processor for executing the computer program instructions, wherein, When the computer program instructions are executed by the processor, the electronic device performs the method as described in any one of claims 1 to 8.

Images