Well cementation two-interface cementing strength prediction method and system

Abstract

The invention discloses a well cementation two-interface cementing strength prediction method and system, and belongs to the technical field of oil and gas well engineering well cementation. According to the correlation degree and the weight coefficient of each characteristic parameter, the weighted correlation degree of each characteristic parameter is determined; according to the correlation degree and the weight coefficient of each feature parameter, determining the weighted correlation degree of each feature parameter; determining a screening threshold value of a preset weighted correlation degree, screening out a plurality of characteristic parameters greater than the screening threshold value of the weighted correlation degree from the characteristic parameters, and constructing a training set of a plurality of samples according to the plurality of screened characteristic parameters and the corresponding cementing strength of the well cementation two-interface; and a neural network model is constructed, and a prediction model for predicting the cementing strength of the well cementation two-interface is obtained through training of the training set. By means of the method, the two-interface bonding strength can be accurately predicted.

Description

Technical Field

[0001] The present invention relates to the technical field of well cementing in oil and gas well engineering, and more particularly to a method and system for predicting the bonding strength of the second interface in well cementing. Background Art

[0002] Well cementing is one of the key links in the construction of oil and gas wells. Its main purpose is to reinforce the wellbore, seal off complex formations that are prone to collapse and leakage, so as to ensure subsequent safe drilling and reasonable oil and gas production, etc. In the downhole annulus sealing system, the interface between the cement sheath and the formation is usually referred to as the second interface in well cementing, and the quality of the bonding of the second interface in well cementing is crucial for the safety of the entire oil and gas wellbore.

[0003] The magnitude of the bonding strength of the second interface in well cementing reflects the quality of its bonding, which refers to the strength of the bonding formed between the sealing material in the annulus and the formation. Determining the bonding strength of the second interface in well cementing is of great significance for ensuring well cementing quality, preventing annulus pressure buildup and interlayer crossflow, optimizing well cementing construction, and improving oil production efficiency, etc. It is one of the key parameters for ensuring wellbore integrity and maintaining the production life of oil and gas wells.

[0004] In the existing methods for determining the bonding strength of the second interface in well cementing, although the empirical formula method is intuitive and simple, its applicability in special cases is relatively low; the laboratory test method requires a lot of time and resources and incurs high costs; the physical model method has high interpretability, but the modeling and analysis processes are relatively complex, and high requirements are placed on the accuracy and reliability of the model; the numerical simulation method has defects such as high computational complexity, difficulty in parameter selection, and difficulty in validating the model.

[0005] However, the parameters affecting the bonding strength of the second interface in well cementing in deep formations include geological factors, drilling factors, and cement slurry performance parameters, etc. Due to the limitations of geological conditions, there are complex non-linear relationships between various parameters, resulting in difficulty in parameter selection and inability to accurately and efficiently predict the bonding strength of the second interface in well cementing. Summary of the Invention

[0006] Aiming at the problems existing in the above fields, the present invention proposes a method and system for predicting the bonding strength of the second interface in well cementing. By determining the weight coefficient of each characteristic parameter and the weighted correlation degree of the correlation degree, the characteristic parameters are screened according to the magnitude of the weighted correlation degree, making the relationship between the characteristic parameters and the bonding strength of the second interface in well cementing more objective. A training set of multiple samples will be constructed based on the screened multiple characteristic parameters and the corresponding bonding strength of the second interface in well cementing, which is used to train the constructed neural network model to obtain a prediction model for predicting the bonding strength of the second interface in well cementing, and predict the bonding strength of the second interface in well cementing, which can fully explore the hidden potential relationships between different types of parameters and the bonding strength of the second interface in well cementing, and achieve accurate prediction of the bonding strength of the second interface.

[0007] To solve the above technical problems, the present invention discloses a method for predicting the bonding strength of the second interface in cementing, comprising the following steps:

[0008] Taking multiple geological parameters, drilling parameters and cementing parameters of the target well obtained as characteristic parameters, and taking the bonding strength of the second interface in cementing of the target well obtained by testing as the target parameter;

[0009] Obtaining multiple sample values of each characteristic parameter; by determining the absolute value of the difference between each sample value of each characteristic parameter and the corresponding target parameter, respectively obtaining the correlation coefficient between each sample value of each characteristic parameter and the corresponding target parameter; performing weighted averaging on the multiple sample values of each characteristic parameter and the correlation coefficient of the corresponding target parameter to determine the correlation degree of the target parameter corresponding to each characteristic parameter;

[0010] By determining the proportion of each sample value of each characteristic parameter in the total sample values of all characteristic parameters, obtaining the entropy value of each characteristic parameter; by performing weighted averaging on the entropy value of each characteristic parameter and the entropy values of all characteristic parameters, determining the weight coefficient of each characteristic parameter;

[0011] According to the correlation degree and weight coefficient of the target parameter corresponding to each characteristic parameter, determining the weighted correlation degree of each characteristic parameter; determining a preset screening threshold for the weighted correlation degree, and screening out multiple characteristic parameters whose weighted correlation degree is greater than the weighted correlation degree screening threshold;

[0012] According to the multiple characteristic parameters after screening and the corresponding bonding strength of the second interface in cementing, constructing a training set of multiple samples; constructing a neural network model, and obtaining a prediction model for predicting the bonding strength of the second interface in cementing through training with the training set.

[0013] Preferably, the determining the correlation degree of the target parameter corresponding to each characteristic parameter specifically includes:

[0014] Obtaining \(x=(x 1 ,x 2 ,\cdots,x n )\) as the characteristic parameter after normalization processing, \(y=(y 1 ,y 2 ,\cdots,y n )\) as the target parameter corresponding to the characteristic parameter \(x\), and \(n\) as the number of samples of each characteristic parameter;

[0015] The correlation degree \(r ij \) of the characteristic parameter \(x\) to the target parameter \(y\) is:

[0016] \(\Delta ij =|y i -x ij |\);

[0017]

[0018] In the formula, x ij is an i-dimensional input vector, representing the sample value of the j-th sample of the i-th characteristic parameter; y j is a 1-dimensional output vector, representing the target parameter value of the j-th sample; Δ ij is the absolute value of the difference between the j-th sample of the characteristic parameter x in the i-th characteristic and the corresponding target parameter y j ; ξ ij is the correlation coefficient between the characteristic parameter x and the target parameter y on the j-th sample in the i-th characteristic; ρ is the discrimination coefficient, ρ ∈ [0, 1]; r ij is the correlation degree between the characteristic parameter x and the target parameter y on the j-th sample in the i-th characteristic.

[0019] Preferably, determining the weight coefficient of each characteristic parameter specifically includes:

[0020] The weight coefficient of the characteristic parameter is obtained by weighted averaging the entropy values of all characteristic parameters:

[0021]

[0022] In the formula, p ij is the proportion of the sample value of the j-th sample of the i-th characteristic parameter corresponding to the characteristic parameter x to the total sample values of all characteristic parameters; m is the number of characteristic parameters; H i is the entropy value of the i-th characteristic parameter; G i is the weight coefficient of the i-th characteristic parameter.

[0023] Preferably, the weighted correlation degree of each characteristic parameter is:

[0024] δ i = G i * r ij ;

[0025] In the formula, δ i is the weighted correlation degree of the weight coefficient G i of the i-th characteristic parameter and the correlation degree r ij ;

[0026] Among them, the greater the weighted correlation degree of the characteristic parameter, the greater the contribution of the characteristic parameter to the prediction result, and the higher the importance of the characteristic parameter.

[0027] Preferably, the multiple geological parameters, drilling parameters, and cementing parameters of the target well include: drilling fluid density, drilling fluid rheology, drilling fluid filtration loss, drilling fluid pH value, filter cake thickness, cement slurry density, cement slurry rheology, compressive strength of cement stone, thickening time of cement slurry, API water loss of cement slurry, spacer fluid density, spacer fluid flushing efficiency, average injection and displacement return velocity, and settlement stability.

[0028] Preferably, the multiple characteristic parameters after screening are those with a weighted correlation degree greater than the weighted correlation degree screening threshold, specifically including:

[0029] Determine the preset screening threshold of the weighted correlation degree according to the value of the weighted correlation degree of the weight coefficient and correlation degree of each characteristic parameter;

[0030] According to the preset screening threshold of the weighted correlation degree, screen out multiple characteristic parameters with a weighted correlation degree greater than the weighted correlation degree screening threshold, and screen out cement slurry density, cement slurry rheology, compressive strength of cement stone, thickening time of cement slurry, drilling fluid density, spacer fluid flushing efficiency, and drilling fluid pH value from the characteristic parameters;

[0031] Take cement slurry density, cement slurry rheology, compressive strength of cement stone, thickening time of cement slurry, drilling fluid density, spacer fluid flushing efficiency, and drilling fluid pH value as the multiple characteristic parameters after screening.

[0032] Preferably, the construction of the neural network model specifically includes:

[0033] Determine that S = [(x 1 , y 1 ), (x 2 , y 2 ),..., (x n , y n )] is the training set of the multiple characteristic parameters after screening, where x i is an m-dimensional characteristic parameter, and y i is the bonding strength of the second cementing interface corresponding to the i-th characteristic parameter;

[0034] According to the multiple characteristic parameters after screening and the corresponding bonding strength of the second cementing interface, calculate the probability value of the sample point of the output parameter through the SoftMax classification function, and divide the sample category of the input parameter according to the probability, where the SoftMax classification function is:

[0035]

[0036] In the formula, y t is the value of the t-th node in the output layer, that is, the sample value of the t-th sample point of the output parameter;

[0037] Optimize the error value of the sample points by setting the error function through the LM optimization algorithm to obtain the error value of the t-th sample point after optimization, where the error function is:

[0038]

[0039] In the formula, x t is the sample value of the t-th sample point of the input parameter, y t is the sample value of the t-th sample point of the output parameter, and w is the weight vector; e t 2 (w) is the error value of the t-th sample point of the output parameter;

[0040] The initial weight vector w = w k has a random value in the interval [-1, 1]. Determine the training error function value E(w k+1 ) corresponding to the (k + 1)-th iteration by calculating the vector w k+1 composed of the weights of the (k + 1)-th iteration;

[0041] The calculation process of the vector w k+1 is as follows:

[0042] w k+1 = w k + Δw;

[0043] Δw = [F T (w)F(w) + μI] -1 F T (w)E(w);

[0044]

[0045] In the formula, w k is the vector composed of the weights of the k-th iteration; Δw is the weight increment; μ is the damping coefficient; I is the identity matrix; F(w) is the Jacobian matrix of the weights; is the partial derivative of the error value with respect to the weight vector;

[0046] By repeating the iteration, calculate the training error function value E(w k+1 ). When E(w k+1 ) is less than E(w k ), then retain the value of E(w k+1 ) and decrease the damping coefficient and continue the iteration; when E(w k+1 ) is greater than E(w k ), then increase the damping coefficient μ (μ = μ·β) and recalculate the training error until E(w k+1 ) is less than E(w kStop the calculation when (condition), and optimize the sample points of each output parameter;

[0047] By repeating the iteration and calculating the value of the training error function, determine the lowest value of the training error function E(w k ) min The optimal structural parameters of the model below, where the structural parameters include the number of hidden layers and neurons;

[0048] Based on the number of unit nodes in the input layer and the output layer, determine the number of unit nodes in the hidden layer as:

[0049]

[0050] where q is the number of unit nodes in the hidden layer, a is the number of unit nodes in the input layer, b is the number of unit nodes in the output layer, and c is a constant;

[0051] Use the training set to train neurons with different numbers of hidden layers in the neural network model. By adjusting the structural parameters of the neural network model, determine the optimal number of hidden layers and neurons;

[0052] Substitute the optimal number of hidden layers and neurons into the neural network model to obtain a prediction model for predicting the cementing bond strength at the second interface.

[0053] It also includes dividing the training set and test set of the prediction data using the K-fold cross-validation method, inputting the test set into the prediction model, and outputting the prediction result of the cementing bond strength at the second interface.

[0054] Preferably, it also includes a cementing bond strength prediction system at the second interface, characterized by including:

[0055] A parameter acquisition module, which is used to take multiple geological parameters, drilling parameters, and cementing parameters of the target well as characteristic parameters, and take the cementing bond strength at the second interface of the target well obtained through testing as the target parameter;

[0056] A characteristic parameter correlation degree determination module, which is used to obtain multiple sample values of each characteristic parameter; by determining the absolute value of the difference between each sample value of each characteristic parameter and the corresponding target parameter, respectively obtain the correlation coefficient between each sample value of each characteristic parameter and the corresponding target parameter; perform weighted averaging on the multiple sample values of each characteristic parameter and the correlation coefficients of the corresponding target parameters to determine the correlation degree of the target parameter corresponding to each characteristic parameter;

[0057] A characteristic parameter weight coefficient determination module, which is used to obtain the entropy value of each characteristic parameter by determining the proportion of each sample value of each characteristic parameter in the total sample values of all characteristic parameters; by performing weighted averaging on the entropy value of each characteristic parameter and the entropy values of all characteristic parameters, determine the weight coefficient of each characteristic parameter;

[0058] A feature parameter screening module, configured to determine the weighted correlation degree of each feature parameter according to the correlation degree and weight coefficient of the target parameter corresponding to each feature parameter; determine a screening threshold for the preset weighted correlation degree, and screen out multiple feature parameters whose weighted correlation degree is greater than the weighted correlation degree screening threshold.

[0059] A prediction module, configured to construct a training set of multiple samples according to the screened multiple feature parameters and the corresponding cementing second interface bonding strength; construct a neural network model, and obtain a prediction model for predicting the cementing second interface bonding strength through training with the training set.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] The cementing second interface bonding strength prediction method proposed by the present invention, by determining the target parameter and the feature parameter, and by determining the correlation degree of the target parameter corresponding to each feature parameter and the weight coefficient of each feature parameter, and then determining the weighted correlation degree of the weight coefficient and the correlation degree of each feature parameter, can combine the entropy method with the grey correlation analysis, effectively solve the problem that the local feature correlation coefficient has a great influence on the correlation degree, improve the reliability of the grey correlation analysis, and make the relationship between the feature parameter and the cementing second interface bonding strength more objective. Screen the features according to the magnitude of the weighted correlation degree value, construct a training set of multiple samples according to the screened multiple feature parameters and the corresponding cementing second interface bonding strength, train the constructed neural network model through the training set, and then obtain a prediction model for predicting the cementing second interface bonding strength. Predict the cementing second interface bonding strength through this prediction model, which can fully explore the hidden potential relationship between different types of parameters and the bonding strength, realize the accurate prediction of the cementing second interface bonding strength, and has the advantages of high accuracy, low cost, good real-time performance and self-adaptability, and has great potential advantages in practical applications. Description of the Drawings

[0062] Figure 1 It is a flow chart of the cementing second interface bonding strength prediction method of the present invention;

[0063] Figure 2 It is a framework diagram of a cementing second interface bonding strength prediction method provided by an embodiment of the present invention;

[0064] Figure 3 It is a determination coefficient diagram of the cementing second interface bonding strength prediction model provided by an embodiment of the present invention;

[0065] Figure 4 It is a correct prediction deviation diagram of the cementing second interface bonding strength prediction model provided by an embodiment of the present invention on the test set. Detailed Embodiments

[0066] Next, in combination with the accompanying drawings in the embodiments of the present invention, Figures 1 - 4 the technical solutions in the embodiments of the present invention will be clearly and completely described. It should be understood that the terms described in the present invention are only for describing specific embodiments and are not used to limit the present invention.

[0067] As Figure 1 shown, the present invention proposes a method for predicting the bonding strength of the second interface of cementing, including the following steps:

[0068] S1: Take multiple geological parameters, drilling parameters, and cementing parameters of the target well obtained as characteristic parameters, and take the bonding strength of the second interface of cementing of the target well obtained by testing as the target parameter;

[0069] S2: Obtain multiple sample values of each characteristic parameter; by determining the absolute value of the difference between each sample value of each characteristic parameter and the corresponding target parameter, respectively obtain the correlation coefficient between each sample value of each characteristic parameter and the corresponding target parameter; perform weighted averaging on the multiple sample values of each characteristic parameter and the correlation coefficient of the corresponding target parameter to determine the correlation degree of the target parameter corresponding to each characteristic parameter;

[0070] S3: Obtain the entropy value of each characteristic parameter by determining the proportion of each sample value of each characteristic parameter in the total sample values of all characteristic parameters; perform weighted averaging on the entropy value of each characteristic parameter and the entropy values of all characteristic parameters to determine the weight coefficient of each characteristic parameter;

[0071] S4: According to the correlation degree and weight coefficient of the target parameter corresponding to each characteristic parameter, determine the weighted correlation degree of each characteristic parameter; determine the screening threshold of the preset weighted correlation degree, and screen out multiple characteristic parameters whose weighted correlation degree is greater than the weighted correlation degree screening threshold;

[0072] S5: According to the multiple characteristic parameters after screening and the corresponding bonding strength of the second interface of cementing, construct a training set of multiple samples; construct a neural network model, and obtain a prediction model for predicting the bonding strength of the second interface of cementing through training with the training set.

[0073] Specifically, it also includes preprocessing the characteristic parameters and target parameters.

[0074] Eliminate the missing values of all geological parameters, drilling parameters, cementing parameters of the target well obtained and the bonding strength of the second interface of cementing obtained by testing.

[0075] Detect outliers in the data by the Z-score method, and the calculation formula is:

[0076]

[0077] where xi is a data point; is the mean of all data points; σ is the standard deviation of all data points.

[0078] Z i The absolute value of represents the distance between the score within the standard deviation range and the overall mean. The threshold is set to 2.5. When |Z i | is greater than 2.5, it can be determined as an outlier.

[0079] It also includes normalizing and preprocessing each characteristic parameter. The maximum and minimum normalization method is used to map the original values of each characteristic parameter to the interval [0, 1]. The normalized data is:

[0080]

[0081] In the formula, x i is the sample value of the i-th characteristic parameter; x max and x min are the maximum and minimum values in the sample value data respectively; f(x i ) is the normalized data, that is, the data of the i-th characteristic parameter after normalization.

[0082] The multiple geological parameters, drilling parameters and cementing parameters of the target well obtained include: drilling fluid density, drilling fluid rheology, drilling fluid filtration loss, drilling fluid pH value, filter cake thickness, cement slurry density, cement slurry rheology, compressive strength of cement stone, thickening time of cement slurry, API water loss of cement slurry, spacer fluid density, spacer fluid flushing efficiency, average injection and displacement return velocity and settlement stability.

[0083] In step S2, determine the correlation degree of the target parameter corresponding to each characteristic parameter, specifically including:

[0084] Obtain x = (x 1 , x 2 ,..., x n ) as the characteristic parameter after normalization processing, y = (y 1 , y 2 ,..., y n ) as the target parameter corresponding to the characteristic parameter x, and n is the number of samples of each characteristic parameter;

[0085] The correlation degree r ij of the characteristic parameter x to the target parameter y is:

[0086] Δ ij = |y i - x ij |;

[0087]

[0088] In the formula, x ij is an i-dimensional input vector, representing the sample value of the j-th sample of the i-th characteristic parameter; y j is a 1-dimensional output vector, representing the target parameter value of the j-th sample; Δ ij is the absolute value of the difference between the j-th sample of the characteristic parameter x in the characteristic parameter i and the corresponding target parameter y j ; ξ ij is the correlation coefficient between the characteristic parameter x and the target parameter y on the j-th sample in the characteristic i; ρ is the discrimination coefficient, ρ ∈ [0, 1]; r ij is the correlation degree between the characteristic parameter x and the target parameter y on the j-th sample in the characteristic i.

[0089] In step S3, determine the weight coefficient of each characteristic parameter, specifically including:

[0090] The weight coefficient of the characteristic parameter is obtained by weighted averaging the entropy values of all characteristic parameters:

[0091]

[0092] In the formula, p ij is the proportion of the sample value of the j-th sample of the i-th characteristic parameter corresponding to the characteristic parameter x in the total sample of all characteristic parameters; m is the number of characteristic parameters; H i is the entropy value of the i-th characteristic parameter; G i is the weight coefficient of the i-th characteristic parameter.

[0093] In step S4, determine the weighted correlation degree of each characteristic parameter as:

[0094] δ i = G i * r ij ;

[0095] In the formula, δ i is the weighted correlation degree of the weight coefficient G i of the i-th characteristic parameter and the correlation degree r ij ;

[0096] Among them, the greater the weighted correlation degree of the characteristic parameter, the greater the contribution of the characteristic parameter to the prediction result, and the higher the importance of the characteristic parameter.

[0097] Determine the screening threshold of the preset weighted correlation degree, and screen out multiple characteristic parameters with a weighted correlation degree greater than the weighted correlation degree screening threshold, specifically including:

[0098] Determine the screening threshold of the preset weighted correlation degree according to the value of the weighted correlation degree of the weight coefficient and the correlation degree of each characteristic parameter.

[0099] According to the preset screening threshold of weighted correlation degree, multiple characteristic parameters with weighted correlation degree greater than the weighted correlation degree screening threshold are screened out, and from the characteristic parameters, slurry density, slurry rheology, compressive strength of cement stone, thickening time of slurry, drilling fluid density, flushing efficiency of spacer fluid, and pH value of drilling fluid are screened out.

[0100] Take the slurry density, slurry rheology, compressive strength of cement stone, thickening time of slurry, drilling fluid density, flushing efficiency of spacer fluid, and pH value of drilling fluid as the multiple characteristic parameters after screening.

[0101] In step S6, a neural network model is constructed, specifically including:

[0102] Determine that S = [(x 1 , y 1 ), (x 2 , y 2 ),..., (x n , y n )] is the training set of multiple characteristic parameters after screening, where x i is an m-dimensional characteristic parameter, and y i is the bonding strength of the second interface of cementing corresponding to the i-th characteristic parameter;

[0103] According to the multiple characteristic parameters after screening and the corresponding bonding strength of the second interface of cementing, calculate the probability value of the sample point of the output parameter through the SoftMax classification function, and divide the sample category of the input parameter according to the probability, where the SoftMax classification function is:

[0104]

[0105] In the formula, y t is the value of the t-th node of the output layer, that is, the sample value of the t-th sample point of the output parameter;

[0106] Set the error function through the LM optimization algorithm to optimize the error value of the sample point, and obtain the optimized error value of the t-th sample point, where the error function is:

[0107]

[0108] In the formula, x t is the sample value of the t-th sample point of the input parameter, y t is the value of the t-th sample point of the output parameter, w is the weight vector; e t 2 (w) is the error value of the t-th sample point of the output parameter;

[0109] The initial weight vector w = w kA random value within the range of [-1, 1], and the vector w composed of the weights for the (k + 1)-th iteration is calculated k+1 , and the value of the training error function E(w k+1 ) corresponding to the (k + 1)-th iteration is determined;

[0110] The vector w k+1 is calculated as follows:

[0111] w k+1 = w k + Δw;

[0112] Δw = [F T (w)F(w) + μI] -1 F T (w)E(w);

[0113]

[0114] In the formula, w k is the vector composed of the weights for the k-th iteration; Δw is the weight increment; μ is the damping coefficient; I is the identity matrix; F(w) is the Jacobian matrix of the weights; is the partial derivative of the error value with respect to the weight vector;

[0115] By repeating the iteration, the value of the training error function E(w k+1 ) is calculated. When E(w k+1 ) is less than E(w k ), then the value of E(w k+1 ) is retained, and the damping coefficient is decreased and the iteration continues; when E(w k+1 ) is greater than E(w k ), then the damping coefficient μ (μ = μ·β) is increased and the training error is recalculated until E(w k+1 ) is less than E(w k ) and the calculation stops, and the sample points of each output parameter are optimized;

[0116] By repeating the iteration and calculating the value of the training error function, the optimal structural parameters of the model under the lowest training error function value E(w k ) min are determined. The structural parameters include the number of hidden layers and neurons;

[0117] Based on the number of unit nodes in the input layer and the number of unit nodes in the output layer, the number of unit nodes in the hidden layer is determined as:

[0118]

[0119] where q is the number of unit nodes in the hidden layer, a is the number of unit nodes in the input layer, b is the number of unit nodes in the output layer, and c is a constant;

[0120] Train neurons in different numbers of hidden layers of the neural network model using the training set, and determine the optimal number of hidden layers and neurons by adjusting the structural parameters of the neural network model.

[0121] Substitute the optimal number of hidden layers and neurons into the neural network model to obtain a prediction model for the bonding strength of the second interface of cementing.

[0122] Use the K-fold cross-validation method to divide the training set and the test set, input the test set into the prediction model for the bonding strength of the second interface of cementing, and output the prediction result of the bonding strength of the second interface of cementing for the target well.

[0123] Through the coefficient of determination R 2 , evaluate the prediction effect of the prediction model;

[0124]

[0125] Measure the difference between the predicted value and the actual value of the prediction model through the correct prediction deviation Q:

[0126]

[0127] In the formula, y i is the actual value of the bonding strength of the second interface; y ’ i is the predicted value of the bonding strength of the second interface; is the average value of the bonding strength of the second interface.

[0128] The present invention also proposes a prediction system for the bonding strength of the second interface of cementing, including:

[0129] A parameter acquisition module, configured to use multiple geological parameters, drilling parameters, and cementing parameters of the target well obtained as characteristic parameters, and use the bonding strength of the second interface of cementing of the target well obtained by testing as the target parameter;

[0130] A characteristic parameter correlation degree determination module, configured to obtain multiple sample values of each characteristic parameter; by determining the absolute value of the difference between each sample value of each characteristic parameter and the corresponding target parameter, respectively obtain the correlation coefficient between each sample value of each characteristic parameter and the corresponding target parameter; perform weighted averaging on the multiple sample values of each characteristic parameter and the correlation coefficients of the corresponding target parameters to determine the correlation degree of the target parameter corresponding to each characteristic parameter;

[0131] A characteristic parameter weight coefficient determination module, configured to obtain the entropy value of each characteristic parameter by determining the proportion of each sample value of each characteristic parameter in the total sample values of all characteristic parameters; perform weighted averaging on the entropy value of each characteristic parameter and the entropy values of all characteristic parameters to determine the weight coefficient of each characteristic parameter;

[0132] A feature parameter screening module, configured to determine the weighted correlation degree of each feature parameter according to the correlation degree and weight coefficient of the target parameter corresponding to each feature parameter; determine a screening threshold for the preset weighted correlation degree, and screen out multiple feature parameters whose weighted correlation degree is greater than the weighted correlation degree screening threshold;

[0133] A prediction module, configured to construct a training set of multiple samples according to the multiple screened feature parameters and the corresponding cementing second interface bond strength; construct a neural network model, and obtain a prediction model for predicting the cementing second interface bond strength through training with the training set.

[0134] The method for predicting the cementing second interface bond strength proposed by the present invention determines the weighted correlation degree of each feature parameter, screens each feature parameter according to the value of the weighted correlation degree, screens out multiple feature parameters whose weighted correlation degree is greater than the preset weighted correlation degree screening threshold, and then obtains the multiple screened feature parameters. According to the multiple screened feature parameters and the corresponding cementing second interface bond strength, a training set of multiple samples is constructed, and the neural network model constructed is trained through the training set to obtain a prediction model for predicting the cementing second interface bond strength, which can fully explore the hidden potential relationship between different types of parameters and the bond strength, and realize the accurate prediction of the cementing second interface bond strength.

[0135] Embodiment

[0136] To verify the effectiveness of the method proposed by the present invention, the following embodiments are provided by the present invention for detailed analysis.

[0137] As Figure 2 shown, it is a framework diagram of the method for predicting the second interface bond strength provided by this embodiment, including the following steps:

[0138] Step 1: Obtain the feature parameters and target parameters of the target well

[0139] From the on-site drilling data, select well sections with good cementing effects, collect a total of 267 groups of data on geology, drilling, and cementing in each range, and obtain a total of 14 data on cementing second interface bond strength, geological parameters, drilling parameters, and cementing parameters through indoor experimental tests, that is, from S 1 to S 14 They are respectively: S 1 Drilling fluid density (g / cm 3 )、S 2 Drilling fluid rheology (mPa.s)、S 3 Drilling fluid filtration loss (ml)、S 4 Drilling fluid pH value、S 5 Filter cake thickness (mm)、S 6 Cement slurry density (g / cm 3)、S 7 Rheology of cement slurry (mPa·s), S 8 Compressive strength of cement stone (MPa), S 9 Thickening time of cement slurry (min), S 10 API fluid loss of cement slurry (ml), S 11 Density of spacer fluid (g / cm 3 )、S 12 Flushing efficiency of spacer fluid, S 13 Average return velocity of injection and displacement (m 3 / min) and S 14 Settling stability (g / cm 3 ), taking S 1 to S 14 as characteristic parameters, and the target parameter D is the bonding strength of the second interface of cementing.

[0140] It also includes preprocessing the characteristic parameters and the target parameter

[0141] Removing the missing values in the geological parameters, drilling parameters, cementing parameters and the characteristic parameters of the bonding strength of the second interface of cementing, dealing with the outliers and normalizing them.

[0142] Table 1 Dataset of geological characteristic parameters, drilling characteristic parameters, cementing characteristic parameters and bonding strength characteristic parameters

[0143]

[0144]

[0145] Step 2: Screening of characteristic parameters

[0146] Calculate the correlation degree, weight coefficient and weighted correlation degree between each geological parameter, drilling parameter, cementing parameter and the bonding strength D of the second interface of cementing in turn. By determining the screening threshold of the preset weighted correlation degree, screen out multiple characteristic parameters with a weighted correlation degree greater than the weighted correlation degree screening threshold, and extract the characteristic parameters with a larger weighted correlation degree, including: S 6 Density of cement slurry, S 7 Rheology of cement slurry, S 8 Compressive strength of cement stone, S 9 Thickening time of cement slurry, S 1 Density of drilling fluid, S 12 Flushing efficiency of spacer fluid and S 4 pH value of drilling fluid. The calculation results of the correlation degree, weight coefficient and weighted correlation degree calculated from S 1 to S 14 are shown in Table 2-4 correspondingly.

[0147] Table 2 Calculation results of correlation degree

[0148] Feature <![CDATA[S 1 > <![CDATA[S 2 > <![CDATA[S 3 > <![CDATA[S 4 > <![CDATA[S 5 > <![CDATA[S 6 > <![CDATA[S 7 > Degree of association 0.6873 0.6547 0.6519 0.7082 0.6947 0.8417 0.7618 Feature <![CDATA[S 8 > <![CDATA[S 9 > <![CDATA[S 10 > <![CDATA[S 11 > <![CDATA[S 12 > <![CDATA[S 13 > <![CDATA[S 14 > Degree of association 0.7239 0.7124 0.7071 0.6748 0.6834 0.6458 0.6488

[0149] Table 3 Weight Calculation Results

[0150] Feature <![CDATA[S 1 > <![CDATA[S 2 > <![CDATA[S 3 > <![CDATA[S 4 > <![CDATA[S 5 > <![CDATA[S 6 > <![CDATA[S 7 > Weight 0.0633 0.0461 0.0419 0.0568 0.0547 0.0613 0.0602 Feature <![CDATA[S 8 > <![CDATA[S 9 > <![CDATA[S 10 > <![CDATA[S 11 > <![CDATA[S 12 > <![CDATA[S 13 > <![CDATA[S 14 > Weight 0.0617 0.0621 0.0519 0.0497 0.0597 0.0428 0.0404

[0151] Table 4 Weighted Correlation Degree Calculation Results

[0152]

[0153]

[0154] Step 3: Establish a prediction model for predicting the cementing second interface bonding strength

[0155] The BP neural network model is a multi-layer feedforward network model based on the backpropagation algorithm. It continuously trains the network model through the reverse transmission of errors, continuously adjusts the network weights in the direction of the gradient descent of the relative error function, and approaches the desired target. It is a typical supervised learning algorithm.

[0156] Taking the cement slurry density, cement slurry rheology, compressive strength of cement stone, thickening time of cement slurry, drilling fluid density, flushing efficiency of spacer fluid, and pH value of drilling fluid extracted by screening according to the weighted correlation degree value as the input of the prediction model, and the cementing second interface bonding strength D as the output of the prediction model, that is, the inputs of the prediction model are S 6 , S 7 , S 8 , S 9 , S 1 , S 12 , and S 4 , and the output is the target parameter D.

[0157] Combining the input parameters S 6 , S 7 , S 8 , S 9 , S 1 , S 12 , and S 4 with the target parameter D to jointly form the training data set S. Then the final training data set S = {S 6 , S 7 , S 8 , S 9 , S 1 , S 12 , S 4 , D} = [(x 1 , y 1 ), (x 2 , y 2 ),...,(x n , y n)], where x i is a 7-dimensional characteristic parameter, and y i is the bonding strength of the second interface of the well cementing.

[0158] Use the training set to train the BP neural network model.

[0159] Calculate the output sample probability value through the following SoftMax classification function, and divide the input sample category according to the probability:

[0160]

[0161] In the formula, x t is the sample value of the t-th sample point of the input parameter, and y t is the sample value of the t-th sample point of the target parameter D.

[0162] Calculate the mean square error loss value through the following formula to optimize the t-th sample point of the input parameter:

[0163]

[0164] Optimize the error value of the t-th sample point of the output layer through the error function to obtain the error value of the optimized t-th sample point, where the error function is:

[0165]

[0166] In the formula, n is the number of samples, and the value of n = 267 in this embodiment; w is the weight vector; e t 2 (w) is the error value of the t-th sample point of the output layer.

[0167] The initial weight vector w = w k The value of is a random value within the interval [-1, 1], and calculate the vector w k+1 composed of the weights of the (k + 1)-th iteration as:

[0168] w k+1 = w k + Δw;

[0169] Re-determine the training error function value E(w k+1 ) according to the error value of the optimized t-th sample point and the vector composed of the weights of the (k + 1)-th iteration.

[0170] Calculate the training error of the model according to the values of Δw, F(w), and E(w k+1 ) as follows:

[0171]

[0172] In the formula, w kThe vector composed of the weights for the k-th iteration; Δw is the weight increment; e(w) is the error vector; μ is the damping coefficient; I is the identity matrix; F(w) is the Jacobian matrix of the weights; is the partial derivative of the error value with respect to the weight vector.

[0173] By repeating the iteration, calculate the training error function value E(w k+1 ), when E(w k+1 ) is less than E(w k ), then retain the value of E(w k+1 ) and reduce the damping coefficient and continue the iteration; when E(w k+1 ) is greater than E(w k ), then increase the damping coefficient μ (μ = μ·β) and recalculate the training error until E(w k+1 ) is less than E(w k ) to stop the calculation, that is, to achieve the optimization of each output sample point.

[0174] Subsequently, according to the input 7 feature data and 1 output parameter, determine that the number of input layer unit nodes is 7 and the number of output layer unit nodes is 1.

[0175] Calculate the value of the number of hidden layer unit nodes q and train different numbers of hidden layer neurons;

[0176]

[0177] where q is the number of hidden layer unit nodes, a is the number of input layer unit nodes, b is the number of output layer unit nodes, and c is a constant.

[0178] By repeating the iteration and calculating the training error function value, determine the lowest training error function value E(w k ) min of the optimal structural parameters of the model.

[0179] Use the 5-fold cross-validation method to divide the dataset, divide the dataset into a training set and a test set according to a ratio, and train the BP neural network model according to the above steps to determine the optimal number of hidden layers and neurons. The optimal structural parameters of the finally obtained model are as follows:

[0180] Table 5 Optimal Structural Parameters of the Cementing Second Interface Bond Strength Prediction Model

[0181] Parameter Range Optimal value Number of hidden layers (1,4) 2 Number of neurons (16,128) 32

[0182] Re-enter the optimal number of hidden layers and neurons into the BP neural network, determine the optimal parameter structure of the model, and obtain the trained cementing second interface bond strength prediction model.

[0183] The test set is input into the prediction model to predict the bonding strength of the second interface of the cementing, and the prediction results are output.

[0184] The prediction model of the bonding strength of the second interface of the cementing is verified and evaluated by the coefficient of determination and the correct prediction deviation.

[0185] For example, a target well in a certain oilfield in Zhongyuan is tested. The data samples collected on site are input into the neural network model for training, and finally a relatively high coefficient of determination and a relatively low correct prediction deviation are shown as in Table 6.

[0186] Table 6 Prediction effect

[0187]

[0188]

[0189] As Figure 3 and Figure 4 shown, they are respectively the coefficient of determination diagram of the prediction model of the bonding strength of the second interface of the cementing provided by the embodiment of the present invention and the correct prediction deviation diagram of the prediction model on the test set. The closer the value of the coefficient of determination is to 1, the stronger the explanatory power of the model; the lower the correct prediction deviation, the higher the prediction accuracy of the prediction model, indicating that the prediction effect of the prediction model is better.

[0190] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

[0191] In addition, unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the art to which the present invention belongs. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods related to the documents. In case of conflict with any incorporated document, the content of this specification shall prevail.

Claims

1. A method for predicting the bonding strength of cementing interfaces, characterized in that: The following steps are involved: The obtained multiple geological parameters, drilling parameters and cementing parameters of the target well are used as characteristic parameters, and the cementing strength of the cementing interface of the target well obtained by testing is used as the target parameter; Obtain multiple sample values ​​of each characteristic parameter; obtain the correlation coefficient between each sample value of each characteristic parameter and the corresponding target parameter by determining the absolute value of the difference between each sample value of each characteristic parameter and the corresponding target parameter; perform weighted averaging on the correlation coefficients of the multiple sample values ​​of each characteristic parameter and the corresponding target parameter to determine the correlation degree of the target parameter corresponding to each characteristic parameter; The entropy value of each characteristic parameter is obtained by determining the proportion of each sample value of each characteristic parameter to the total sample value of all characteristic parameters; the weight coefficient of each characteristic parameter is determined by taking a weighted average of the entropy value of each characteristic parameter and the entropy values ​​of all characteristic parameters; Determine the weighted correlation of each feature parameter according to the correlation and weight coefficient of the target parameter corresponding to each feature parameter; determine a preset screening threshold of the weighted correlation, and screen out multiple feature parameters whose weighted correlation is greater than the screening threshold of the weighted correlation; According to the screened multiple characteristic parameters and the corresponding cementing interface bonding strength, a training set of multiple samples is constructed; a neural network model is constructed, and a prediction model for predicting the cementing interface bonding strength of the cementing interface is obtained by training the training set.

2. The method for predicting the bonding strength of cementing interfaces according to claim 1, characterized in that: Determining the correlation degree of the target parameter corresponding to each characteristic parameter specifically includes: Get x=(x1,x2,...,x n ) is the normalized feature parameter, y=(y1,y2,...,y n ) is the target parameter corresponding to the feature parameter x, and n is the number of samples of each feature parameter; The correlation degree r of feature parameter x to target parameter y ij for: In the formula, x ij is an i-dimensional input vector, representing the sample value of the j-th sample of the i-th feature parameter; y j is a 1-dimensional output vector, representing the target parameter value of the j-th sample; Δ ij is the jth sample of feature parameter x in feature parameter i and the corresponding target parameter y j The absolute value of the difference between ij is the correlation coefficient between feature parameter x and target parameter y on the jth sample in feature i; ρ is the resolution coefficient, ρ∈[0,1]; r ij is the correlation between feature parameter x and target parameter y on the jth sample in feature i.

3. The method for predicting the bonding strength of cementing interfaces according to claim 2, characterized in that: Determining the weight coefficient of each characteristic parameter specifically includes: The weight coefficient of the feature parameter is obtained by weighted averaging the entropy values ​​of all feature parameters: In the formula, p ij is the proportion of the sample value of the jth sample of the i-th feature parameter corresponding to the feature parameter x to the sample value of the total samples of all feature parameters; m is the number of feature parameters; H i is the entropy value of the i-th characteristic parameter; G i is the weight coefficient of the i-th feature parameter.

4. The method for predicting the bonding strength of cementing interfaces according to claim 3, characterized in that: The weighted correlation of each characteristic parameter is determined as follows: d i =G i *r ij ; In the formula, δ i is the weight coefficient G of the i-th feature parameter i and the correlation r ij The weighted correlation of Among them, the greater the weighted correlation of the feature parameter, the greater the contribution of the feature parameter to the prediction result, and the higher the importance of the feature parameter.

5. The method for predicting the bonding strength of cementing interfaces according to claim 4, characterized in that: The multiple geological parameters, drilling parameters and cementing parameters of the target well include: drilling fluid density, drilling fluid rheology, drilling fluid filtration loss, drilling fluid pH value, filter cake thickness, cement slurry density, cement slurry rheology, cement stone compressive strength, cement slurry thickening time, cement slurry API water loss, spacer fluid density, spacer fluid flushing efficiency, average injection return rate and sedimentation stability.

6. The method for predicting the bonding strength of cementing interfaces according to claim 5, characterized in that: The step of screening out multiple characteristic parameters whose weighted correlation is greater than a weighted correlation screening threshold specifically includes: Determining a preset screening threshold of the weighted correlation degree according to the weight coefficient of each characteristic parameter and the value of the weighted correlation degree of the correlation degree; According to a preset weighted correlation screening threshold, multiple characteristic parameters whose weighted correlation is greater than the weighted correlation screening threshold are screened out, and cement slurry density, cement slurry rheology, cement stone compressive strength, cement slurry thickening time, drilling fluid density, spacer fluid flushing efficiency and drilling fluid pH value are screened out from the characteristic parameters; Cement slurry density, cement slurry rheology, cement stone compressive strength, cement slurry thickening time, drilling fluid density, spacer fluid flushing efficiency and drilling fluid pH value are selected as multiple characteristic parameters.

7. The method for predicting the bonding strength of cementing interfaces according to claim 6, characterized in that: The construction of the neural network model specifically includes: Determine S = [(x1, y1), (x2, y2), ..., (x n ,y n )] is the training set of multiple feature parameters after screening, where x i is the characteristic parameter of m dimension, y i is the cementing strength of the cementing interface corresponding to the i-th characteristic parameter; According to the screened multiple characteristic parameters and the corresponding cementing strength of the cementing interface, the probability value of the sample point of the output parameter is calculated by the SoftMax classification function, and the sample category of the input parameter is divided according to the probability, where the SoftMax classification function is: In the formula, y t is the value of the tth node in the output layer, that is, the sample value of the tth sample point of the output parameter; The error function is set by the LM optimization algorithm to optimize the error value of the sample point, and the error value of the tth sample point after optimization is obtained, where the error function is: In the formula, x t is the sample value of the tth sample point of the input parameter, y t is the sample value of the tth sample point of the output parameter, w is the weight vector; e t 2 (w) is the error value of the tth sample point of the output parameter; Initial weight vector w = w k The value of is a random value in the interval [-1,1], and the vector w composed of the weights of the k+1th iteration is calculated k+1 , determine the training error function value E(w k+1 ); Vector w k+1 The calculation process is: In k+1 =in k +Δw; Δw=[F T (w)F(w)+μI] -1 F T (w)E(w); In the formula, w k is the vector of weights of the kth iteration; Δw is the weight increment; μ is the damping coefficient; I is the unit matrix; F(w) is the Jacobian matrix of the weights; is the partial derivative of the error value with respect to the weight vector; By repeated iterations, the training error function value E(w k+1 ), when E(w k+1 ) is less than E(w k ), then retain E(w k+1 ) value and reduce the damping coefficient Continue to iterate; when E(w k+1 ) is greater than E(w k ), then increase the damping coefficient μ (μ = μ·β) and recalculate the training error until E(w k+1 ) is less than E(w k ) and stop the calculation, and optimize the sample points of each output parameter; By repeatedly iterating and calculating the training error function value, the lowest training error function value E(w k ) min The optimal structural parameters of the model include the number of hidden layers and neurons; The number of hidden layer unit nodes is determined by the number of input layer unit nodes and the number of output layer unit nodes: Among them, q is the number of hidden layer unit nodes, a is the number of input layer unit nodes, b is the number of output layer unit nodes, and c is a constant; Use the training set to train neurons in different numbers of hidden layers of the neural network model, and determine the optimal number of hidden layers and neurons by adjusting the structural parameters of the neural network model; The optimal number of hidden layers and neurons is substituted into the neural network model to obtain a prediction model for predicting the bonding strength of the cementing interface.

8. The method for predicting the bonding strength of cementing interfaces according to claim 7, characterized in that: It also includes using the K-fold cross validation method to divide the prediction data into a training set and a test set, inputting the test set into the prediction model, and outputting the prediction result of the cementing strength of the cementing interface.

9. A cementing interface bonding strength prediction system, characterized in that: include: A parameter acquisition module is used to use the acquired multiple geological parameters, drilling parameters and cementing parameters of the target well as characteristic parameters, and to use the cementing strength of the second interface of the target well acquired through testing as a target parameter; The characteristic parameter association degree determination module is used to obtain multiple sample values ​​of each characteristic parameter; obtain the association coefficient between each sample value of each characteristic parameter and the corresponding target parameter by determining the absolute value of the difference between each sample value of each characteristic parameter and the corresponding target parameter; and perform weighted averaging on the association coefficients of the multiple sample values ​​of each characteristic parameter and the corresponding target parameter to determine the association degree of the target parameter corresponding to each characteristic parameter; The feature parameter weight coefficient determination module is used to obtain the entropy value of each feature parameter by determining the proportion of each sample value of each feature parameter to the total sample value of all feature parameters; and determine the weight coefficient of each feature parameter by taking a weighted average of the entropy value of each feature parameter and the entropy values ​​of all feature parameters; The feature parameter screening module is used to determine the weighted correlation of each feature parameter according to the correlation and weight coefficient of the target parameter corresponding to each feature parameter; determine a preset screening threshold of the weighted correlation, and screen out multiple feature parameters whose weighted correlation is greater than the weighted correlation screening threshold; The prediction module is used to construct a training set of multiple samples based on the screened multiple characteristic parameters and the corresponding cementing interface bonding strength; construct a neural network model, and obtain a prediction model for predicting the cementing interface bonding strength through training the training set.

Images