A method for predicting sand production in oil wells based on production dynamics and ESP operating parameters.

Abstract

This invention discloses a method for predicting sand production in oil wells based on production dynamics and ESP (Electric Submersible Pump) operating parameters, comprising the following steps: S1: a primary sand production factor correlation analysis strategy based on oil well production dynamic parameters; S2: a secondary sand production pattern prediction strategy based on ESP pump current. On one hand, this invention constructs a multi-dimensional dynamic parameter evaluation system by integrating grey relational analysis and analytic hierarchy process (AHP), establishing a quantitative correlation model between production parameter fluctuations and sand production levels, and realizing dynamic screening and weight allocation of sensitive factors of operating parameters. On the other hand, based on the temporal feature capture capability of LSTM networks, a two-dimensional feature space of pump current change rate and amplitude is constructed. By deeply mining the coupling relationship between current waveform and sand particle transport, a dynamically adaptive sand production state identification mechanism is established.

Description

Technical Field

[0001] This invention belongs to the field of petroleum engineering technology, and in particular relates to a method for predicting sand production in oil wells based on production dynamics and electric submersible pump operating parameters. Background Technology

[0002] The efficient development of loose sandstone reservoirs in offshore oilfields faces severe challenges, with sand production in wells being a key issue affecting safe and stable production. Under high-speed extraction conditions, due to low reservoir cementation and drastic formation pressure changes, sand particles enter the wellbore with the fluid, easily leading to a series of production accidents such as equipment wear, reduced pump efficiency, and shortened pump maintenance cycles. Relevant statistics show that sand production-related submersible pump failures account for over 90% of all pump maintenance incidents, reducing the maintenance cycle from 300 days to 192 days, severely restricting the economic benefits of offshore oilfields. Therefore, accurately predicting sand production risks and taking proactive prevention and control measures are core requirements for achieving intelligent and efficient development of offshore oilfields.

[0003] Currently, sand production prediction technologies are mainly based on rock mechanics models, logging parameters, or empirical formulas, such as the sonic transit time method, the Beaufort-Saufer index method, and the S-index method. These methods establish critical sand production indices using parameters such as sonic transit time and rock elastic modulus. However, the prediction results from a single method vary significantly, requiring comprehensive judgment based on normalized weighting coefficients. In practical applications, the differences in production modes across wells and insufficient adaptability to field data lead to low reliability of prediction results. Furthermore, while coupled production profile and sand-carrying capacity models can determine threshold production levels to guide production by correlating well productivity curves with critical sand-carrying flow rates, they rely on static geological parameters and cannot reflect sand transport patterns under dynamic production conditions in real time. Fiber optic acoustic monitoring and resistance methods can detect sand collision signals in real time, but the equipment is costly, installation is complex, and construction is more difficult due to space constraints on offshore platforms, hindering large-scale offshore applications.

[0004] Therefore, existing sand production prediction methods generally suffer from a disconnect between theoretical models and actual mine conditions. They are largely based on idealized assumptions and fail to fully consider the complex operating conditions of offshore oil wells, such as high gas content, multiphase flow, and temperature and pressure fluctuations, leading to insufficient prediction accuracy. Furthermore, these methods lack integration with dynamic production parameters and fail to effectively utilize the correlation between ESP operating parameters, such as current, frequency, temperature, and discharge rate, and sand production, making real-time monitoring and early warning difficult. In addition, current technologies are still primarily passive sand control methods, focusing on mechanical sand control and chemical sand fixation, without establishing a prediction-based active control mechanism. This results in incomplete capacity release, accelerated equipment wear, and high maintenance costs.

[0005] As a core lifting device in offshore oilfields, the operating parameters of ESPs (Electric Submersible Pumps), such as pump current, vibration frequency, inlet and outlet pressures, are significantly correlated with the sand content in the well fluid. For example, sand particles entering the pump body can cause impeller jamming, and sudden changes in motor load can trigger abnormal current. However, existing research is mostly at the qualitative analysis stage, and on-site handling relies solely on the experience and judgment of on-site operators, without establishing quantitative models of current characteristics and sand production volume and scale to guide actual production. In recent years, research on improving the durability of ESPs by optimizing their operating conditions and structure has gradually unfolded. For example, sand-resistant ESPs have improved their erosion resistance by optimizing impeller materials and structural design, such as hard alloy coatings and open impellers. However, the research and development focus is still on hardware improvements, without establishing a close link between operating parameters and the prediction of sand production conditions and scale.

[0006] With the advancement of digitalization in offshore oilfields, the demand for real-time data acquisition and intelligent analysis is growing, providing a new approach for integrating ESP (Electric Submersible Pump) operating parameters with sand production prediction. However, existing technologies still have shortcomings in multi-source data fusion, failing to deeply integrate multi-dimensional parameters such as ESP current, vibration, and temperature with formation properties and production dynamics data, making it difficult to construct high-precision prediction models. Furthermore, traditional critical sand production indicators, such as sonic transit time ≥105 μs / ft, are mostly fixed values ​​and do not dynamically adjust with production pressure differentials and fluid properties, leading to delayed early warnings. Existing monitoring systems rely on offline data or periodic logging, which cannot meet the needs of offshore platforms for real-time early warning and rapid decision-making.

[0007] In summary, existing sand production prediction technologies, due to limitations in theoretical models and insufficient utilization of dynamic data, are insufficient to meet the demands of efficient offshore oilfield development. This invention, by deeply exploring the intrinsic correlation between ESP operating parameters and sand production, constructs a real-time, accurate prediction system, providing technical support for intelligent sand control in offshore oil wells, and offering significant economic and safety benefits. Summary of the Invention

[0008] The problem this invention aims to solve is to provide a method for predicting sand production in oil wells based on production dynamics and ESP operating parameters. This method innovatively integrates ESP operating data with formation dynamic parameters to establish a sand production risk classification and early warning mechanism. It also adaptively adjusts the prediction threshold based on historical data and real-time operating conditions to improve the model's generalization ability.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by this invention is: a method for predicting sand production in oil wells based on production dynamics and electric submersible pump operating parameters, comprising the following steps: S1: A correlation analysis strategy for primary sand production factors based on dynamic parameters of oil well production; S2: A prediction strategy for secondary sand discharge patterns based on the pump current of an electric submersible pump.

[0010] Furthermore, S1 includes the following steps: S11: Record the general data of a sand-producing well in an oilfield, including static geological data and production dynamic data. The recording period is from normal production before sand production to the point where there are obvious signs of sand production. S12: Using grey relational analysis and analytic hierarchy process as the main analytical methods, the correlation and weight of factors affecting the sand content of sand-producing wells are analyzed. S13: Based on the strength of correlation or parameter weight, rank the main controlling factors affecting the sand content of sand-producing wells and determine the characteristics of factors associated with the sand content.

[0011] Furthermore, in S11, the criterion for determining obvious signs of sand production is that the sand content of the well is greater than 0.01% according to laboratory tests.

[0012] Furthermore, S12 includes the following steps: S121: Determine the analysis sequence, using the general parameters of typical sand-producing wells as reference sequences Y={Y( k ) | k = 1, 2, … , n The actual profile parameters of the sand-producing wells recorded were used as the comparison sequence X. i ={X i ( k ) | k = 1, 2,… , n}, i = 1, 2, … , m Establish a basic data table for grey relational analysis of sand wells; S122: Dimensionless processing, which normalizes or eliminates dimensions from data; S123: Calculate the grey relational coefficient; S124: Calculate the degree of correlation; S125: Establish a hierarchical structure model; S126: Establish a pairwise comparison matrix; S127: Calculate the weight vector and perform a consistency check; S128: Calculate the combined weight vector and perform a combined consistency test.

[0013] Furthermore, S2 includes the following steps: S21: Record the operating data of the electric submersible pump of a sand-producing well in an oilfield. The recording period starts from normal production before sand production and stops when there are obvious signs of sand production. S22: Based on the results obtained from the primary sand production factor correlation analysis strategy, the factors with strong correlation with the sand content in the test are selected, and a long short-term memory neural network model, i.e., the LSTM model, is established. S23: Perform data preprocessing on the production dynamic data of the recorded oil wells; S24: Set the parameters of the LSTM model; S25: Input the parameters, take the pump current as the research object, take the factors that are strongly correlated with the sand content in the test as independent variables, train the LSTM model, and divide the training set and validation set. S26: Output the pump current prediction result, analyze the change of the pump current within a certain time range, and calculate the pump current early warning threshold; S27: Establish a sand production early warning mechanism.

[0014] Furthermore, S22 includes the following steps: S221: Construct an LSTM cell structure. For each time step, calculate the current state and hidden state according to the formulas for the input gate, forget gate, and output gate. S222: Bidirectional LSTM is used to capture long-period features; S223: The gated recurrent unit extracts short-period features and obtains the reset gate and update gate by combining the input at this position and the output of the upper layer with the Sigmoid activation function; S224: Introducing a spatiotemporal weighted module.

[0015] Furthermore, in S23, the data preprocessing includes denoising, normalization, and data cleaning, so that the data can be analyzed under the same dimension, thereby improving the training effect of the long short-term memory neural network model.

[0016] Furthermore, S25 includes the following steps: S251: Calculate the loss; S252: Calculate the gradient; S253: Update parameters; S254: The training set and validation set are divided proportionally, and LSTM models are trained separately for each division. The error under each division ratio is obtained, and the training set and validation set with the smallest error are selected for analysis.

[0017] Furthermore, in step S26, based on the matching of historical pump current parameters with dynamic time warping, a pump current warning threshold is determined, achieving adaptive threshold adjustment. The algorithm formula is as follows: in: Q —Current sequence, dimensionless; C —Optimal matching historical sequence, dimensionless; threshold —Threshold, dimensionless; μ —Mean value, dimensionless; γ—Adjustment coefficient, used to adjust the influence of the DTW part on the threshold, usually a constant, dimensionless; DTW—Dynamic Time Warping Distance, m; DTW(Q,C)—DTW distance between sequence Q and sequence C, m; L—Sequence length, m.

[0018] Furthermore, in S27, the early warning mechanism is determined by the predicted current, the actual current, and the rate of change of current, with the formula for the rate of change of current as follows: .

[0019] in: k —Number of sampling points; t —Index of the sampling points, dimensionless; I pred,i For the first i The predicted value at time A; I real,i It is the first i The true value at any given moment, A.

[0020] The advantages and positive effects of this invention are: 1. On one hand, this invention constructs a multi-dimensional dynamic parameter evaluation system by integrating grey relational analysis and hierarchical analysis, establishing a quantitative correlation model between production parameter fluctuations and sand output, and realizing dynamic screening and weight allocation of sensitive factors of operating parameters. On the other hand, based on the time-series feature capture capability of LSTM networks, a two-dimensional feature space of pump current change rate and amplitude is constructed. By deeply mining the coupling relationship between current waveform and sand particle movement, a dynamic adaptive sand output state identification mechanism is established.

[0021] 2. This invention provides a primary sand production factor correlation analysis strategy based on dynamic parameters of oil well production, which features comprehensive factor analysis and dynamic parameter correlation.

[0022] 3. This invention provides a two-stage sand production pattern prediction strategy based on the pump current of an electric submersible pump, which features intelligent feature extraction, accurate time series modeling, and real-time anomaly identification. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the overall process of an embodiment of the present invention.

[0024] Figure 2 This is an evaluation diagram of the sand production scale of a sand-producing well in an embodiment of the present invention.

[0025] Figure 3 This is a diagram showing the analysis results of the grey relational analysis method and the hierarchical analysis method in the embodiments of the present invention.

[0026] Figure 4This is a graph showing the overall relationship between key parameters of the electric submersible pump and the sand content in laboratory tests in an embodiment of the present invention.

[0027] Figure 5 This is a graph showing the relationship between the minimum current of the electric submersible pump and the sand content in the test results in an embodiment of the present invention.

[0028] Figure 6 This is a graph showing the relationship between the maximum current of the electric submersible pump and the sand content in the test results in an embodiment of the present invention.

[0029] Figure 7 This is a graph showing the relationship between the fluctuation amplitude of the electric submersible pump and the sand content in the test results in an embodiment of the present invention.

[0030] Figure 8 This is a diagram illustrating the LSTM model training process in an embodiment of the present invention.

[0031] Figure 9 This is a graph showing the predicted changes in pump current of an oil well over the next ten days, as described in an embodiment of the present invention. Detailed Implementation

[0032] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The embodiments of the present invention will be further described below with reference to the accompanying drawings: like Figure 1 As shown, an oil well sand production prediction method based on production dynamics and electric submersible pump operating parameters includes the following steps.

[0034] S1: A strategy for correlation analysis of primary sand production factors based on dynamic parameters of oil well production.

[0035] Specifically, the primary sand production factor correlation analysis strategy includes the following steps: S11: Record the well profile data of a sand-producing well in an oilfield, including static geological data and production dynamic data. The recording period is from normal production before sand production to the point where obvious signs of sand production appear. Preferably, the static geological data includes, but is not limited to, porosity, permeability, clay content, and crude oil viscosity. The production dynamic data includes, but is not limited to, production pressure differential, daily fluid production, daily oil production, water cut, fluid production intensity, oil production intensity, fluid production index, and oil production index. In this embodiment, the criterion for determining that a sand-producing well has obvious signs of sand production is that the sand content of the well is greater than 0.01% according to laboratory tests.

[0036] S12: Using grey relational analysis and analytic hierarchy process as the main analytical methods, the correlation and weight of factors affecting the sand content of sand-producing wells are analyzed.

[0037] Specifically, the main steps in grey relational analysis include: The analysis series is determined using the general parameters of typical sand-producing wells, including porosity, permeability, clay content, crude oil density, crude oil viscosity, production pressure differential, daily fluid production, daily oil production, water cut, fluid production intensity, oil production intensity, fluid production index, and oil production index, as reference series (parent series) Y = {Y( k ) | k = 1, 2, … , n}, using the actual profile parameters of the recorded sand-producing wells as the comparison sequence (subsequence) X i ={X i ( k ) | k = 1, 2, … , n}, i = 1, 2, … , m A basic data table of grey relational analysis for sand wells was established.

[0038] Dimensionless processing: Normalize or eliminate the dimensions of data to avoid uneven weighting among data in the same group caused by large data spans. It can also avoid situations where different dimensions make it inconvenient to compare data or difficult to draw correct conclusions.

[0039] Calculating the grey relational coefficient: The formula for calculating the grey relational coefficient is as follows: k = 1, 2, … , n ; in: ρ —Resolution coefficient, dimensionless; —The i-th comparison sequence is in the i-th... k The correlation coefficient between the indicator and the reference sequence reflects the degree of similarity between the two at that point and is dimensionless. Y ( k —The reference sequence is in the first k The values ​​of each indicator are dimensionless; xi ( k )—No. i The comparison sequence in the th ... k The numerical values ​​of each indicator are dimensionless.

[0040] Calculating the correlation: The correlation is a numerical value representing the degree of association between the comparison sequence and the reference sequence. Its calculation formula is as follows: , k = 1, 2, … , n .

[0041] Specifically, the analysis of the weights of each parameter using the analytic hierarchy process includes the following steps: Establish a hierarchical structure model: Based on the influence relationship with the target factor (sand content in the test), decompose the various factors related to sand production in oil wells into several levels from top to bottom according to different attributes. The decomposition is based on the fact that factors in the same level belong to or influence factors in the upper level, while also dominating or being affected by factors in the lower level.

[0042] Constructing a pairwise comparison matrix: Starting from level 2 of the hierarchical model, for factors within the same level that belong to (or influence) each factor in the previous level, construct a pairwise comparison matrix using pairwise comparisons and comparison scales 1-4, continuing down to the lowest level. Compare the... i The element and the first j When considering the importance of an element relative to a factor in the previous layer, use quantified relative weights. a ij To describe, among which, a ij Values ​​between 1 and 4: (1) a ij =1, element i With elements j They are equally important to the factors at the next higher level. (2) a ij =2, element i Compared to elements j Slightly more important than the factors at the next higher level; (3) a ij =3, element i Compared to elements j Important to the factors at the next higher level; (4) a ij =4, element i Compared to elements j It is far more important than the factors at the next higher level; Establish a paired comparison table of factors influencing sand production in oilfield sand-producing wells, and form a table of factors... a ij The paired comparison matrix A = ( a ij ) n×n .

[0043] Calculate the weight vectors and perform a consistency check: for pairwise comparison matrices A Indicators of inconsistency CI The calculation method is as follows ,according to CI With average random consistency index RIThe random consistency ratio can be obtained. ,when CR When <0.1, determine the pairwise comparison matrix. A It exhibits satisfactory consistency; matrices can also be used. A The maximum eigenvalue and its eigenvector judgment matrix A Consistency, where the largest eigenvalue The corresponding feature vector , k = 1, 2, … , n ,in .

[0044] Calculate the combined weight vector and perform a combined consistency test: Calculate the weight vector of each parameter using the geometric mean method or the normalized column average method.

[0045] S13: Based on the strength of correlation or parameter weights, rank the main controlling factors affecting the sand content in sand-producing wells to determine the characteristics of factors associated with the sand content. Specifically, the strength of correlation is determined by the correlation coefficient in the grey relational analysis method; the larger the correlation coefficient, the stronger the correlation with the reference sequence. In the analytic hierarchy process (AHP), the parameter weights reflect the degree of influence of each factor on the target layer; the larger the parameter weight, the stronger the influence of that parameter on the target layer.

[0046] S2: A prediction strategy for secondary sand discharge patterns based on the pump current of an electric submersible pump.

[0047] Specifically, the strategy for predicting secondary sand production patterns includes the following steps: S21: Record the well profile data for a sand-producing well in a certain oilfield, including production dynamic data and ESP (Electric Submersible Pump) operating data. The data should be recorded from the point of normal production before sand production until obvious signs of sand production appear. Preferably, the production dynamic data includes, but is not limited to, daily fluid production, gas-oil ratio, oil pressure, and flowing pressure; the ESP operating data includes, but is not limited to, pump current, pump frequency, pump inlet pressure, pump outlet pressure, and motor temperature. Similarly, the criterion for determining obvious signs of sand production in a sand-producing well is that the well's sand content, as determined by laboratory analysis, is greater than 0.01%.

[0048] S22: Based on the results obtained from the primary factor correlation analysis strategy, the factors with strong correlation to the sand content in the test were selected, and a Long Short-Term Memory (LSTM) neural network model was established. Specifically, the LSTM model is an improvement on the Recurrent Neural Network (RNN) model, and its framework includes the following: Constructing the LSTM cell structure: For each time... t The current state and hidden state are calculated according to the formulas for the input gate, forget gate, and output gate. Input gate Used to control the current inputx t The impact on the current cell state, where, σ It is the Sigmoid activation function. W i It is the weight matrix of the input gate. b i It is a bias; Forgotten Gate Control the cell state of the previous moment C t-1 Which information needs to be retained? Output gate Control the current unit state C t For the current hidden state h t The impact; Cell state , where the current input candidate value ; Hidden state The final hidden state h t It is determined by the current cell state and the output gate.

[0049] Bidirectional LSTM is used to capture long-period features: for hidden states h t ,have .

[0050] Gated Recurrent Unit (GRU) extracts short-period features: by combining the input at this position and the output of the upper layer with a Sigmoid activation function, a reset gate and an update gate are obtained. The "reset gate + ... tanh "The original information (output from the previous position and input from the current position) is modified to obtain the result after processing the original information. This result is then passed through an update gate and added to the weighted output from the previous position to obtain the output from the current position. For each time period..." t Calculate according to the formula for updating and resetting doors. h t ; Update Gate ; Reset door ; Reset door r t Previous position output h t-1 Enter this location x t Joint decision ; Final output at current position ; Spatiotemporal weighted module ,in , H hist This is the historical state matrix.

[0051] S23: Preprocess the acquired oil well production dynamic data. This preprocessing includes denoising, normalization, and data cleaning. The purpose is to clean the data, remove outliers and noise, and ensure the data is analyzed on the same scale, thereby improving the model's training effectiveness.

[0052] S24: Set model parameters.

[0053] S25: Input parameters, taking pump current as the research object, and factors with strong correlation between the recorded and tested sand content as independent variables, to train the LSTM model and divide it into training and validation sets.

[0054] Specifically, the LSTM model is trained using supervised learning methods based on historical current data and labels of sand discharge events, enabling it to predict future current trends. During training, cross-validation is employed to adjust model parameters and ensure generalization ability. The main steps include: Calculate the loss: Define the loss function L, typically using the mean absolute error. or mean square error As the loss function, where, X The sequence of independent variables, h ( x ) is the prediction function. y This is the actual value.

[0055] Gradient calculation: Starting from the last time step, calculate the gradient for each time step backward, taking into account the influence of each gate.

[0056] Update parameters: Use gradient descent Update the weights and biases of the preceding time steps.

[0057] Preferably, the training set and validation set are divided into ratios of 90:10, 85:15, 80:20, and 75:25 respectively, and LSTM models are trained separately for each ratio. The error under each ratio is obtained, and the training set and validation set combination with the smallest error is selected for analysis.

[0058] S26: Output pump current prediction results, analyze the changes in pump current within a certain time range, and calculate the pump current warning threshold. Specifically, a pump current historical parameter matching based on Dynamic Time Warping (DTW) is designed to determine the pump current warning threshold, realizing adaptive threshold adjustment. The main algorithm is as follows: in: Q —Current sequence, dimensionless; C —Optimal matching historical sequence, dimensionless; threshold —Threshold, dimensionless; μ —Mean value, dimensionless; γ —Adjustment coefficient, used to adjust the influence of the DTW part on the threshold, usually a constant, dimensionless; DTW—Dynamic Time Warping Distance, m; DTW(Q,C)—DTW distance between sequence Q and sequence C, m; L—Sequence length, m.

[0059] S27: Establish a sand production early warning mechanism. Specifically, the early warning mechanism is determined by the predicted current, actual current, and current change rate, with the current change rate... ; in: k —Number of sampling points; t —Index of the sampling points, dimensionless; I pred,i For the first i The predicted value at time A; I real,i It is the first i The true value at any given moment, A.

[0060] Preferably, this embodiment designs a three-level early warning mechanism based on the real-time changes in pump current: when the pump current change exceeds the early warning threshold by more than 10%, a blue early warning is automatically triggered, and it is recommended to adjust production measures; when the pump current change exceeds 20%, a yellow early warning is triggered, and it is recommended to adjust production measures, and if necessary, sand prevention measures should be taken to reduce the risk of sand discharge; when the pump current change exceeds 30%, a red early warning is triggered, and the early warning system detects potential sand discharge hazards and recommends to take sand prevention, sand flushing and other measures to reduce losses.

[0061] The present invention will now be described in detail with reference to specific embodiments: S1: The first-level sand production factor analysis strategy based on oil well production dynamic parameters, the steps are as follows: S11: Fifty sand-producing wells from a certain oilfield in the Bohai Sea were selected, and their well profile data were recorded, including static geological data (porosity, permeability, clay content, crude oil viscosity) and production dynamic data (production pressure difference, daily fluid production, daily oil production, water cut, fluid production intensity, oil production intensity, fluid production index, oil production index). The data was recorded from the time of normal production before sand production until there were obvious signs of sand production.

[0062] S12: Based on the sand content measured in oil well tests and the impact of sand production on oil production, an evaluation and analysis of the sand production scale was conducted. The sand production scale of oil wells in this oilfield was divided into three categories, as shown in Table 1. Figure 2 ; Table 1 Classification of Sand Production Scale of Sand-Producing Wells S13: Using grey relational analysis and analytic hierarchy process as the main analytical methods, the correlation and weight of factors affecting the sand content of sand-producing wells are analyzed, and a grey relational basic data table for sand-producing wells, a pairwise comparison table of sand-producing influencing factors, and a pairwise comparison matrix are established.

[0063] S14: The main controlling factors affecting the sand content of sand-producing wells are ranked, as shown in Table 2 and... Figure 3 .

[0064] Table 2. Analysis results of grey relational analysis and analytic hierarchy process. According to the grey relational analysis method, the main controlling factors related to sand content analysis, ranked from highest to lowest correlation, are: fluid production index, oil production intensity, oil production index, fluid production intensity, daily oil production, daily fluid production, pressure differential, water cut, permeability, porosity, crude oil viscosity, and mud content. According to the analytic hierarchy process (AHP), the main controlling factors related to sand content analysis, ranked from highest to lowest correlation based on the weight of the parameter and the target layer (sand content analysis), are: fluid production index, oil production intensity, oil production index, fluid production intensity, daily oil production, daily fluid production, pressure differential, water cut, permeability, porosity, crude oil viscosity, and mud content. The conclusions from both analytical methods are similar. Based on the above results, the factors related to sand content analysis are classified according to their correlation: the strongest correlation factors are fluid production index, oil production index, oil production intensity, and fluid production intensity; the moderate correlation factors are daily fluid production, daily oil production, production pressure differential, and water cut; and the weakest correlation factors are porosity, permeability, mud content, and crude oil viscosity.

[0065] S2: Prediction of secondary sand production patterns based on electric submersible pump current, the steps are as follows: S21: Select 39 sand-producing wells in a certain oilfield in Bohai Sea and record their well profile data, including production dynamic data (daily fluid production, gas-oil ratio, oil pressure, and flowing pressure) and electric submersible pump operating data (pump current, pump frequency, pump inlet pressure, pump outlet pressure, and motor temperature). The data collection period is from normal production before sand production to when there are obvious signs of sand production.

[0066] S22: Classifying by pump current, we discuss the production situation of oil wells under three modes: pump current decrease before sand production, pump current stabilization, and pump current increase.

[0067] S23: Statistical analysis was performed on the relationship between sand content and minimum current, maximum current, and fluctuation range. The resulting graph is shown below. Figures 4-7 Wells with high sand production (>0.5%) were found to have minimum and maximum currents concentrated in the range of 20-40 A, with fluctuations mainly concentrated in the range of 0-50%, and sand production scale was classified as Class III.

[0068] S24: Based on the relationship between the three modes of pump current decrease / stable / increase and the sand content in the laboratory, set the model parameters and establish an LSTM model.

[0069] S25: Input parameters, using pump current as the research object, and the recorded daily liquid production, gas-oil ratio, oil pressure, flow pressure, pump frequency, pump inlet pressure, pump outlet pressure, and motor temperature as independent variables, to train an LSTM model, dividing it into training and validation sets. The model training process is as follows. Figure 8 .

[0070] S26: Output prediction results. The model can predict the pump current of the electric submersible pump for the next 10 days, such as... Figure 9 As shown, the mean absolute error of the training set is 0.78A, and the mean absolute error of the validation set is 1.41A.

[0071] S27: The model can automatically issue real-time warnings based on the predicted pump current threshold and the actual pump current in production. If the pump current change exceeds the warning threshold by more than 10%, a blue warning will be automatically triggered, and production measures will be adjusted accordingly. If the pump current change exceeds 20%, a yellow warning will be triggered, and production measures will be adjusted accordingly. If necessary, sand prevention measures should be taken to reduce the risk of sand discharge. If the pump current change exceeds 30%, a red warning will be triggered. The warning system has detected potential sand discharge hazards and recommends sand prevention and flushing measures to reduce losses.

[0072] The advantages and positive effects of this invention are: 1. On one hand, this invention constructs a multi-dimensional dynamic parameter evaluation system by integrating grey relational analysis and hierarchical analysis, establishing a quantitative correlation model between production parameter fluctuations and sand output, and realizing dynamic screening and weight allocation of sensitive factors of operating parameters. On the other hand, based on the time-series feature capture capability of LSTM networks, a two-dimensional feature space of pump current change rate and amplitude is constructed. By deeply mining the coupling relationship between current waveform and sand particle movement, a dynamic adaptive sand output state identification mechanism is established.

[0073] 2. This invention provides a primary sand production factor correlation analysis strategy based on dynamic parameters of oil well production, which features comprehensive factor analysis and dynamic parameter correlation.

[0074] 3. This invention provides a two-stage sand production pattern prediction strategy based on the pump current of an electric submersible pump, which features intelligent feature extraction, accurate time series modeling, and real-time anomaly identification.

[0075] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for predicting sand production in oil wells based on production dynamics and electric submersible pump operating parameters, characterized in that: Includes the following steps, S1: A correlation analysis strategy for primary sand production factors based on dynamic parameters of oil well production; S2: A prediction strategy for secondary sand discharge patterns based on the pump current of an electric submersible pump.

2. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters according to claim 1, characterized in that: S1 includes the following steps: S11: Record the general data of a sand-producing well in an oilfield, including static geological data and production dynamic data. The recording period is from normal production before sand production to the point where there are obvious signs of sand production. S12: Using grey relational analysis and analytic hierarchy process as the main analytical methods, the correlation and weight of factors affecting the sand content of sand-producing wells are analyzed. S13: Based on the strength of correlation or parameter weight, rank the main controlling factors affecting the sand content of sand-producing wells and determine the characteristics of factors associated with the sand content.

3. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters according to claim 2, characterized in that: In S11, the criterion for determining obvious signs of sand production is that the sand content of the well is greater than 0.01% according to laboratory tests.

4. A method for predicting sand production in oil wells based on production dynamics and ESP operating parameters, as described in claim 2 or 3, characterized in that: S12 includes the following steps: S121: Determine the analysis sequence, using the general parameters of typical sand-producing wells as reference sequences Y={Y( k ) | k = 1,2, … , n The actual profile parameters of the sand-producing wells recorded were used as the comparison sequence X. i ={X i ( k ) | k = 1, 2, … , n }, i = 1, 2, … , m Establish a basic data table for grey relational analysis of sand wells; S122: Dimensionless processing, which normalizes or eliminates dimensions from data; S123: Calculate the grey relational coefficient; S124: Calculate the degree of correlation; S125: Establish a hierarchical structure model; S126: Establish a pairwise comparison matrix; S127: Calculate the weight vector and perform a consistency check; S128: Calculate the combined weight vector and perform a combined consistency test.

5. A method for predicting sand production in oil wells based on production dynamics and ESP operating parameters, as described in claim 2 or 3, characterized in that: S2 includes the following steps: S21: Record the operating data of the electric submersible pump of a sand-producing well in an oilfield. The recording period starts from normal production before sand production and stops when there are obvious signs of sand production. S22: Based on the results obtained from the primary sand production factor correlation analysis strategy, the factors with strong correlation with the sand content in the test are selected, and a long short-term memory neural network model, i.e., the LSTM model, is established. S23: Perform data preprocessing on the production dynamic data of the recorded oil wells; S24: Set the parameters of the LSTM model; S25: Input the parameters, take the pump current as the research object, take the factors that are strongly correlated with the sand content in the test as independent variables, train the LSTM model, and divide the training set and validation set. S26: Output the pump current prediction result, analyze the change of the pump current within a certain time range, and calculate the pump current early warning threshold; S27: Establish a sand production early warning mechanism.

6. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters according to claim 5, characterized in that: S22 includes the following steps: S221: Construct an LSTM cell structure. For each time step, calculate the current state and hidden state according to the formulas for the input gate, forget gate, and output gate. S222: Bidirectional LSTM is used to capture long-period features; S223: The gated recurrent unit extracts short-period features and obtains the reset gate and update gate by combining the input at this position and the output of the upper layer with the Sigmoid activation function; S224: Introducing a spatiotemporal weighted module.

7. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters as described in claim 5, characterized in that: In step S23, the data preprocessing includes denoising, normalization, and data cleaning, so that the data can be analyzed under the same dimension, thereby improving the training effect of the long short-term memory neural network model.

8. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters according to claim 5, characterized in that: S25 includes the following steps: S251: Calculate the loss; S252: Calculate the gradient; S253: Update parameters; S254: The training set and validation set are divided proportionally, and LSTM models are trained separately for each division. The error under each division ratio is obtained, and the training set and validation set with the smallest error are selected for analysis.

9. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters according to claim 5, characterized in that: In step S26, based on the matching of historical pump current parameters using dynamic time warping, a pump current warning threshold is determined, achieving adaptive threshold adjustment. The algorithm formula is as follows: in: Q —Current sequence, dimensionless; C —Optimal matching historical sequence, dimensionless; threshold —Threshold, dimensionless; μ —Mean value, dimensionless; γ —Adjustment coefficient, used to adjust the influence of the DTW part on the threshold, usually a constant, dimensionless; DTW—Dynamic Time Warping Distance, m; DTW(Q,C)—DTW distance between sequence Q and sequence C, m; L—Sequence length, m.

10. The method for predicting sand production in oil wells based on production dynamics and ESP operating parameters according to claim 5, characterized in that: In S27, the early warning mechanism is determined by the predicted current, the actual current, and the rate of change of current, and the formula for the rate of change of current is as follows: in: k —Number of sampling points; t —Index of the sampling points, dimensionless; I pred,i For the first i The predicted value at time A; I real,i It is the first i The true value at any given moment, A.

Images