Oil and gas field yield prediction and fluctuation tracing method based on three-level architecture

Abstract

The invention relates to the field of artificial intelligence, and discloses an oil and gas field yield prediction and fluctuation tracing method based on a three-level architecture. The method comprises the following steps: step S10, collecting multi-source heterogeneous data based on'large tank-branching-single well ', and preprocessing the multi-source heterogeneous data; step S20, constructing a level interconnection type prediction network, and predicting data of a single well, a branch line and a large tank; step S30, constructing a physical feasible constraint, and applying a secondary penalty term to prediction violating the constraint; and S40, constructing a multi-stage alarm classifier, and positioning an abnormal conduction path and a fluctuation root cause classification result. According to the method, the problem that a predicted value is not feasible (such as negative yield and overpressure flow) due to the fact that an existing data-driven model neglects the physical law of an oil reservoir-pipe network system is solved, and the problems of error accumulation and fault traceability blind areas caused by a traditional single-point prediction framework in a three-level linkage system are solved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence, and in particular relates to an oil and gas field production prediction and fluctuation tracing method based on a three-level architecture. Background Art

[0002] In oilfield production systems, production fluctuations can quickly identify equipment failures, formation problems, and human errors, reducing downtime losses. A three-tiered approach to oil and gas field production forecasting and fluctuation tracing is crucial for ensuring stable production and increased efficiency. Existing methods primarily rely on two types of technologies: First, numerical simulations based on physical mechanisms, adjusting parameters to match production data; and second, independent analysis of single-point data using traditional time series models or anomaly detection algorithms such as isolation forests. However, existing technologies have significant defects: (1) Traditional data-driven models do not embed the physical laws of the reservoir-pipeline system (such as single-well production thresholds and branch line pressure difference limits), resulting in prediction results that often show infeasible solutions such as negative production and overpressure flow, which violate engineering reality; (2) The single-point prediction framework ignores the spatiotemporal coupling characteristics of the three-level structure of "large tank-branch line-single well", and the single-well prediction error will be amplified through the pipeline network level. For example, when the branch line flow deviation accumulates to the large tank level, the error will increase by 3-5 times; (3) Conventional anomaly detection methods lack multi-level topological modeling capabilities and cannot distinguish the transmission paths of wellhead failures, pipeline leakage and tank body anomalies. For example, pipeline pressure fluctuations can easily be misjudged as a decline in single-well pump efficiency.

[0003] Therefore, there is an urgent need to build an integrated intelligent prediction and tracing method that integrates physical constraints and hierarchical spatiotemporal characteristics, realize dynamic correlation modeling of three-level systems, and design multi-level wave propagation maps to accurately locate the source of anomalies. Summary of the Invention

[0004] The purpose of this invention is to provide an oil and gas field production prediction and fluctuation tracing method based on a three-level architecture, which realizes the coordinated prediction of production and accurate positioning of abnormal sources in oil field production systems by integrating physical constraints and multi-level spatiotemporal coupling characteristics.

[0005] Step 1: The oil and gas field production forecasting and fluctuation tracing method based on the three-level architecture includes the following steps: Step S10, collect multi-source heterogeneous data based on the three-level architecture of "large tank-branch line-single well", and perform data cleaning, missing value filling and normalization on the multi-source heterogeneous data, wherein the multi-source heterogeneous data includes single well level data, branch line level data, large tank level data and historical fault records; single well level data includes single well data acquisition time, single well number, historical liquid production, historical gas production, historical oil production, historical water production, wellhead pressure, bottom hole flow pressure, wellhead temperature, wellbore radius, water content, gas-liquid ratio, electric pump current, electric pump voltage, electric pump frequency, pumping unit strokes, pumping unit load, reservoir permeability, effective thickness of oil layer, and pressure difference generated , crude oil underground viscosity, crude oil volume coefficient, oil leakage radius; branch line level data include branch line data collection time, branch line number, historical branch line flow, pipeline flow, pipeline pressure, pressure difference fluctuation, temperature gradient, maximum allowable fluid flow rate, fluid density, fluid viscosity, pipeline radius, safety factor; large tank level data include large tank data collection time, historical tank storage level, pipe pressure, oil temperature, oil density, well tank flow, tank flow, sedimentation rate, ambient temperature; historical fault records include fault time, fault type, fault level attribution, single well liquid production 6 hours before the fault, branch line flow 6 hours before the fault, large tank level 6 hours before the fault; Step S20: constructing a hierarchical interconnected prediction network through spatiotemporal attention mechanism and feature alignment to achieve single well prediction, line flow prediction, and large tank prediction, and transferring and correcting errors step by step; Step S30: Based on the reservoir material balance and oil production engineering specifications, a loss function for physical constraints is constructed, and a quadratic penalty term is applied to the prediction results that violate the constraints, so that the prediction results are within a reasonable range; Step S40, construct a multi-level alarm classifier, use the exponentially weighted moving average method to calculate the control limits of the prediction residual sequence, and establish a dynamic threshold fluctuation detector. After the fluctuation exceeds the control limit within three consecutive time points, use the decision tree to output the alarm type and alarm level attribution, so as to effectively identify the type of abnormal event and the location where the abnormal event occurs.

[0006] Step 2: According to the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in step 1, in step S20, the single well production is predicted using the single well prediction module. The single well prediction results are aligned with the line features through the spatiotemporal attention mechanism to predict the line flow rate. The line prediction results are fused with the large tank features through the spatiotemporal attention mechanism to predict the large tank data. The single well production prediction is defined as: in represents the predicted production of a single well, represents the single well prediction model, Represents single-well level data; The prediction sequence output by the single well module is aligned with the line features through the spatiotemporal attention mechanism and is defined as: in represents the spatiotemporal attention fusion function from single well to branch line, Indicates line level data, represents the binding of single well data to the spatial encoder, Defined as ,in is element-wise multiplication, where is the spatial position code, is the dot product operation, Indicates the binding of line data and time encoder, Defined as ， in represents the time position code, For transpose operation, by transposing and Perform dot product operation to obtain the correlation score between the single well predicted flow rate and the line level data. Represents the scaling factor, and the scaled correlation score is The attention weight is obtained by operation and the attention weight is combined with Perform dot product operations to scale the original line-level data according to label strength The impact of the original line level data Keep the branch's own information; The line flow prediction is defined as: in Indicates the predicted flow rate of each line. represents the line splitting prediction model; The prediction sequence output by the splitting module is aligned with the large tank features through the spatiotemporal attention mechanism and is defined as: in Represents the spatiotemporal attention fusion function of the branch line to the large tank, Represents the large tank level data, Indicates the binding of line data and spatial encoder, Defined as ,in is the spatial position code, Indicates the binding of large tank data and time encoder, Defined as ,in Represents the time position encoding, by transposing and Perform dot product operation to obtain the correlation score between the predicted flow rate of the split line and the large tank level data. Represents the scaling factor, and the scaled correlation score is The attention weight is obtained by operation and the attention weight is combined with Perform dot product operations to scale the original large tank level data according to label strength The impact of the original large tank level data Keep the big jar's own information; The large tank data prediction is defined as: in Represents the forecast data of the large tank, Represents the large tank prediction model.

[0007] Step 3: According to the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in step 1, in step S30, the total loss function is defined as: in represents the total loss, represents the mean square error loss, Represents the secondary penalty loss for a single well, which imposes a secondary penalty on the part of the predicted production of a single well that exceeds the upper and lower limits. It represents the secondary penalty loss of the split line, which imposes a secondary penalty on the part of the split line predicted flow rate that exceeds the threshold. The secondary penalty loss of a single well is defined as: in represents the loss weight, represents the activation function, represents the predicted production of a single well, It represents the theoretical maximum production of a single well calculated using Darcy's formula. It represents the minimum single-well production calculated using the minimum economic production. The maximum single-well production is defined as: in represents the theoretical maximum production of a single well, represents the effective permeability of the reservoir, represents the effective thickness of the oil layer, Indicates the production pressure difference, Indicates the underground viscosity of crude oil, represents the crude oil volume coefficient, represents the oil leakage radius, represents the wellbore radius, represents the driving energy term of the radial flow of the fluid, It represents the radial flow resistance when the fluid converges from the drainage boundary to the wellbore. It represents the total resistance term for radial flow of fluid; The minimum single well production is defined as: in, Indicates the minimum production of a single well, represents the annual fixed cost, represents the net price of crude oil, represents the unit variable cost; The split-line quadratic penalty loss is defined as: in represents the secondary penalty loss of the split line, represents the loss weight, represents the activation function, represents the predicted flow of the branch line, Indicates the maximum flow rate allowed in the branch pipeline, which is defined as: in represents the pipe radius, Indicates the maximum allowable flow rate of the fluid. represents the fluid density, represents the safety factor, Indicates the cross-sectional area of the branch pipe.

[0008] Step 4: According to the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in step 1, in step S40, the residual sequence is defined as: in represents the residual sequence, represents the time step, represents the predicted value, represents the true value; The control limits are defined as: in express The limit value of the time, Indicates the current moment, represents the attenuation factor, express The residual of time, express The smoothed value of the residual sequence at time , is given by definition, Indicates the previous moment. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly explain the embodiments of the present invention and the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. It is obvious that 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 paying any creative work.

[0010] Figure 1 This is a structural diagram of the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in the present invention.

[0011] Figure 2 This is a flow chart of the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in the present invention. DETAILED DESCRIPTION

[0012] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. However, it is obvious that the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0013] The purpose of this invention is to provide an oil and gas field production prediction and fluctuation tracing method based on a three-level architecture. By integrating physical constraints and multi-level spatiotemporal coupling characteristics, it can achieve collaborative production prediction and accurate positioning of abnormal sources in oil field production systems.

[0014] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] Step 1: Oil and gas field production prediction and fluctuation tracing method based on three-level architecture, such as Figure 1 FIG. 1 shows a structural diagram of the oil and gas field production prediction and fluctuation tracing method based on a three-level architecture, which includes the following steps: Step S10, collect multi-source heterogeneous data based on the three-level architecture of "large tank-branch line-single well", and perform data cleaning, missing value filling and normalization on the multi-source heterogeneous data, wherein the multi-source heterogeneous data includes single well level data, branch line level data, large tank level data and historical fault records; single well level data includes single well data acquisition time, single well number, historical liquid production, historical gas production, historical oil production, historical water production, wellhead pressure, bottom hole flow pressure, wellhead temperature, wellbore radius, water content, gas-liquid ratio, electric pump current, electric pump voltage, electric pump frequency, pumping unit strokes, pumping unit load, reservoir permeability, effective thickness of oil layer, and pressure difference generated , crude oil underground viscosity, crude oil volume coefficient, oil leakage radius; branch line level data include branch line data collection time, branch line number, historical branch line flow, pipeline flow, pipeline pressure, pressure difference fluctuation, temperature gradient, maximum allowable fluid flow rate, fluid density, fluid viscosity, pipeline radius, safety factor; large tank level data include large tank data collection time, historical tank storage level, pipe pressure, oil temperature, oil density, well tank flow, tank flow, sedimentation rate, ambient temperature; historical fault records include fault time, fault type, fault level attribution, single well liquid production 6 hours before the fault, branch line flow 6 hours before the fault, large tank level 6 hours before the fault; Step S20: constructing a hierarchical interconnected prediction network through spatiotemporal attention mechanism and feature alignment to achieve single well prediction, line flow prediction, and large tank prediction, and transferring and correcting errors step by step; Step S30: Based on the reservoir material balance and oil production engineering specifications, a loss function for physical constraints is constructed, and a quadratic penalty term is applied to the prediction results that violate the constraints, so that the prediction results are within a reasonable range; Step S40, construct a multi-level alarm classifier, use the exponentially weighted moving average method to calculate the control limits of the prediction residual sequence, and establish a dynamic threshold fluctuation detector. After the fluctuation exceeds the control limit within three consecutive time points, use the decision tree to output the alarm type and alarm level attribution, so as to effectively identify the type of abnormal event and the location where the abnormal event occurs.

[0016] Step 2: According to the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in step 1, in step S20, the single well production is predicted using the single well prediction module. The single well prediction results are aligned with the line features through the spatiotemporal attention mechanism to predict the line flow rate. The line prediction results are fused with the large tank features through the spatiotemporal attention mechanism to predict the large tank data. The single well production prediction is defined as: in represents the predicted production of a single well, represents the single well prediction model, Represents single-well level data; The prediction sequence output by the single well module is aligned with the line features through the spatiotemporal attention mechanism and is defined as: in represents the spatiotemporal attention fusion function from single well to branch line, Indicates line level data, represents the binding of single well data to the spatial encoder, Defined as ,in is element-wise multiplication, where is the spatial position code, is the dot product operation, Indicates the binding of line data and time encoder, Defined as ， in represents the time position code, For transpose operation, by transposing and Perform dot product operation to obtain the correlation score between the single well predicted flow rate and the line level data. Represents the scaling factor, and the scaled correlation score is The attention weight is obtained by operation and the attention weight is combined with Perform dot product operations to scale the original line-level data according to label strength The impact of the original line level data Keep the branch's own information; The line flow prediction is defined as: in Indicates the predicted flow rate of each line. represents the line splitting prediction model; The prediction sequence output by the splitting module is aligned with the large tank features through the spatiotemporal attention mechanism and is defined as: in Represents the spatiotemporal attention fusion function of the branch line to the large tank, Represents the large tank level data, Indicates the binding of line data and spatial encoder, Defined as ,in is the spatial position code, Indicates the binding of large tank data and time encoder, Defined as ,in Represents the time position encoding, by transposing and Perform dot product operation to obtain the correlation score between the predicted flow rate of the split line and the large tank level data. Represents the scaling factor, and the scaled correlation score is The attention weight is obtained by operation and the attention weight is combined with Perform dot product operations to scale the original large tank level data according to label strength The impact of the original large tank level data Keep the big jar's own information; The large tank data prediction is defined as: in Represents the forecast data of the large tank, Represents the large tank prediction model.

[0017] Step 3: According to the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in step 1, in step S30, the total loss function is defined as: in represents the total loss, represents the mean square error loss, Represents the secondary penalty loss for a single well, which imposes a secondary penalty on the part of the predicted production of a single well that exceeds the upper and lower limits. It represents the secondary penalty loss of the split line, which imposes a secondary penalty on the part of the split line predicted flow rate that exceeds the threshold. The secondary penalty loss of a single well is defined as: in represents the loss weight, represents the activation function, represents the predicted production of a single well, It represents the theoretical maximum production of a single well calculated using Darcy's formula. It represents the minimum single-well production calculated using the minimum economic production. The maximum single-well production is defined as: in represents the theoretical maximum production of a single well, represents the effective permeability of the reservoir, represents the effective thickness of the oil layer, Indicates the production pressure difference, Indicates the underground viscosity of crude oil, represents the crude oil volume coefficient, represents the oil leakage radius, represents the wellbore radius, represents the driving energy term of the radial flow of the fluid, It represents the radial flow resistance when the fluid converges from the drainage boundary to the wellbore. It represents the total resistance term for radial flow of fluid; The minimum single well production is defined as: in, Indicates the minimum production of a single well, represents the annual fixed cost, represents the net price of crude oil, represents the unit variable cost; The split-line quadratic penalty loss is defined as: in represents the secondary penalty loss of the split line, represents the loss weight, represents the activation function, represents the predicted flow of the branch line, Indicates the maximum flow rate allowed in the branch pipeline, which is defined as: in represents the pipe radius, Indicates the maximum allowable flow rate of the fluid. represents the fluid density, represents the safety factor, Indicates the cross-sectional area of the branch pipe.

[0018] Step 4: According to the oil and gas field production prediction and fluctuation tracing method based on the three-level architecture described in step 1, in step S40, the residual sequence is defined as: in represents the residual sequence, represents the time step, represents the predicted value, represents the true value; The control limits are defined as: in express The limit value of the time, Indicates the current moment, represents the attenuation factor, express The residual of time, express The smoothed value of the residual sequence at time , is given by definition, Indicates the previous moment.

[0019] In summary, the oil and gas field production prediction and fluctuation tracing method based on a three-level architecture provided by the present invention solves the shortcomings of traditional modeling in spatiotemporal modeling, and breaks through the limitations of traditional single models in complex industrial scenarios, providing an integrated solution of high-precision prediction and explainable tracing for oil and gas production systems.

[0020] The above content only represents the embodiments of the present invention and should not be considered as limiting the scope of the present invention. Any equivalent structure or equivalent process transformation based on the contents described in the present specification and drawings, as well as direct or indirect applications in other related technical fields, shall be deemed to be included in the scope of patent protection of the present invention.

[0021] Other technical features not explicitly mentioned in this example may be fine-tuned by those skilled in the art based on actual needs. However, it should be noted that it is not necessary for those skilled in the art to adopt these specific details to practice the present invention. In other examples, to avoid obscuring the present invention, well-known components, structures, or parts are not specifically described, and are therefore within the scope of protection of the technical solutions claimed in the claims of the present invention.

Claims

1. The oil and gas field production forecasting and fluctuation tracing method based on a three-level architecture is characterized by: The following steps are involved: Step S10, collect multi-source heterogeneous data based on the large tank-branch line-single well three-level architecture, and perform data cleaning, missing value filling and normalization processing on the multi-source heterogeneous data, wherein the multi-source heterogeneous data includes single well level data, branch line level data, large tank level data and historical fault records; single well level data includes single well data acquisition time, single well number, historical liquid production, historical gas production, historical oil production, historical water production, wellhead pressure, bottom hole flow pressure, wellhead temperature, wellbore radius, water content, gas-liquid ratio, electric pump current, electric pump voltage, electric pump frequency, pumping unit strokes, pumping unit load, reservoir permeability, effective thickness of oil layer, generated pressure difference, Crude oil underground viscosity, crude oil volume coefficient, and oil leakage radius; branch line level data includes branch line data collection time, branch line number, historical branch line flow, pipeline flow, pipeline pressure, pressure differential fluctuation, temperature gradient, maximum allowable fluid flow rate, fluid density, fluid viscosity, pipeline radius, and safety factor; large tank level data includes large tank data collection time, historical tank storage level, pipe pressure, oil temperature, oil density, well tank flow, tank discharge flow, sedimentation rate, and ambient temperature; historical fault records include fault time, fault type, fault level, single well liquid production in the six hours before the fault, branch line flow in the six hours before the fault, and large tank level in the six hours before the fault; Step S20: construct a hierarchical interconnected prediction network through the spatiotemporal attention mechanism and feature alignment to achieve single well prediction, line flow prediction, and large tank prediction, and transfer and correct errors step by step; Step S30: Based on the reservoir material balance and oil production engineering specifications, a loss function for physical constraints is constructed, and a quadratic penalty term is applied to the prediction results that violate the constraints, so that the prediction results are within a reasonable range; Step S40: construct a multi-level alarm classifier, use the exponentially weighted moving average method to calculate the control limits of the prediction residual sequence, and establish a dynamic threshold fluctuation detector. After the fluctuation exceeds the control limit within three consecutive time points, use the decision tree to output the alarm type and alarm level attribution, so as to effectively identify the type of abnormal event and the location where the abnormal event occurs.

2. The oil and gas field production prediction and fluctuation tracing method based on the three-level architecture according to claim 1 is characterized in that: In step S20, the single well production is predicted using the single well prediction module. The single well prediction result is aligned with the line feature through the spatiotemporal attention mechanism to predict the line flow rate. The line prediction result is fused with the large tank feature through the spatiotemporal attention mechanism to predict the large tank data. The single well production prediction is defined as: in represents the predicted production of a single well, represents the single well prediction model, Represents single-well level data; The prediction sequence output by the single well module is aligned with the line features through the spatiotemporal attention mechanism and is defined as: in represents the spatiotemporal attention fusion function from single well to branch line, Indicates line level data, represents the binding of single well data to the spatial encoder, Defined as ,in is element-wise multiplication, where is the spatial position code, is the dot product operation, Indicates the binding of line data and time encoder, Defined as ， in represents the time position code, For transpose operation, by transposing and Perform dot product operation to obtain the correlation score between the single well predicted flow rate and the line level data. Represents the scaling factor, and the scaled correlation score is The attention weight is obtained by operation and the attention weight is combined with Perform dot product operations to scale the original line-level data according to label strength The impact of the original line level data Keep the branch's own information; The line flow prediction is defined as: in Indicates the predicted flow rate of each line. represents the line splitting prediction model; The prediction sequence output by the splitting module is aligned with the large tank features through the spatiotemporal attention mechanism and is defined as: in Represents the spatiotemporal attention fusion function of the branch line to the large tank, Represents the large tank level data, Indicates the binding of line data and spatial encoder, Defined as ,in is the spatial position code, Indicates the binding of large tank data and time encoder, Defined as ,in Represents the time position encoding, by transposing and Perform dot product operation to obtain the correlation score between the predicted flow rate of the split line and the large tank level data. Represents the scaling factor, and the scaled correlation score is The attention weight is obtained by operation and the attention weight is combined with Perform dot product operations to scale the original large tank level data according to label strength The impact of the original large tank level data Keep the big jar's own information; The large tank data prediction is defined as: in Represents the forecast data of the large tank, Represents the large tank prediction model.

3. The oil and gas field production prediction and fluctuation tracing method based on the three-level architecture according to claim 1 is characterized in that: In step S30, the total loss function is defined as: in represents the total loss, represents the mean square error loss, Represents the secondary penalty loss for a single well, which imposes a secondary penalty on the part of the predicted production of a single well that exceeds the upper and lower limits. It represents the secondary penalty loss of the split line, which imposes a secondary penalty on the part of the split line predicted flow rate that exceeds the threshold. The secondary penalty loss of a single well is defined as: in represents the loss weight, represents the activation function, represents the predicted production of a single well, It represents the theoretical maximum production of a single well calculated using Darcy's formula. It represents the minimum single-well production calculated using the minimum economic production. The maximum single-well production is defined as: in represents the theoretical maximum production of a single well, represents the effective permeability of the reservoir, represents the effective thickness of the oil layer, Indicates the production pressure difference, Indicates the underground viscosity of crude oil, represents the crude oil volume coefficient, represents the oil leakage radius, represents the wellbore radius, represents the driving energy term of the radial flow of the fluid, It represents the radial flow resistance when the fluid converges from the drainage boundary to the wellbore. It represents the total resistance term for radial flow of fluid; The minimum single well production is defined as: in, Indicates the minimum production of a single well, represents the annual fixed cost, represents the net price of crude oil, represents the unit variable cost; The split-line quadratic penalty loss is defined as: in represents the secondary penalty loss of the split line, represents the loss weight, represents the activation function, represents the predicted flow of the branch line, Indicates the maximum flow rate allowed in the branch pipeline, which is defined as: in represents the pipe radius, Indicates the maximum allowable flow rate of the fluid. represents the fluid density, represents the safety factor, Indicates the cross-sectional area of the branch pipe.

4. The oil and gas field production prediction and fluctuation tracing method based on the three-level architecture according to claim 1 is characterized in that: In step S40, the residual sequence is defined as: in represents the residual sequence, represents the time step, represents the predicted value, represents the true value; The control limits are defined as: in express The limit value of the time, Indicates the current moment, represents the attenuation factor, express The residual of time, express The smoothed value of the residual sequence at time , is given by definition, Indicates the previous moment.

Images