A method and system for predicting drilling loss rate based on geological and engineering parameters

Abstract

The present invention provides a method and system for predicting drilling loss rate based on geological and engineering parameters, which belongs to the field of drilling for oil and gas exploration and development. The method comprises: step one: obtaining parameters that affect the downhole loss rate and collecting the drilling loss rate; step two: processing the parameters that affect the downhole loss rate to obtain processed parameters; step three: establishing a neural network; step four: using the processed parameters and the drilling loss rate to train and verify the neural network to obtain a drilling loss rate prediction model; step five: using the loss rate prediction model to predict the drilling to be predicted to obtain the drilling loss rate of the drilling to be predicted. The present invention realizes artificial intelligence prediction of drilling loss rate based on geological and engineering parameters, solves the problem of large errors in the prediction of drilling site loss rate based solely on expert experience and theoretical calculation methods, reduces the risk of drilling well leakage, and achieves the purpose of safe and efficient drilling.

Description

Technical Field

[0001] The present invention belongs to the field of oil and gas exploration and development drilling, and in particular relates to a method and system for predicting drilling loss rate based on geological and engineering parameters. Background Art

[0002] Oil and gas drilling is a high-investment, high-risk operation, with downhole drilling fluid loss being a common risk. Drilling often encounters fractures, cavities, or highly porous formations, leading to varying degrees of downhole fluid loss. Loss not only increases drilling costs but is also a major constraint to safe and efficient drilling. Accurately predicting the downhole loss rate is key to reducing the incidence of lost circulation and the time and complexity of lost circulation treatment. Accurately predicting the downhole loss rate based on known formation information and drilling parameters allows for proactive preventive measures, such as adjusting the drilling fluid density or increasing the amount of plugging material added to the drilling fluid, to mitigate the risk of loss. However, failure to accurately predict the downhole loss rate can lead to serious losses due to misjudgment, compromising drilling safety, or to the reckless increase in plugging material dosage, increasing drilling fluid costs.

[0003] Chinese patent publication CN110766192A discloses a deep learning-based drilling lost circulation prediction system and method, which uses a VGG convolutional neural network model for deep learning. The system and method include: 1) acquiring drilling production operation data; 2) preprocessing the acquired data to obtain sample data for deep learning; and 3) feature extraction and classification training. The system uses a VGG convolutional neural network model for deep learning, and obtains a trained drilling lost circulation prediction model through feature extraction and classification training. 4) Result output and result confirmation: Use the trained drilling leakage prediction model to verify the real-time verification data and give the leakage judgment result; Chinese patent publication CN110439534A discloses a seismic data leakage prediction method, which relates to the field of seismic data application technology for oil and gas field exploration and development, including the following steps: a. High-resolution processing; b. Fine layer comparison; c. Automatic fold identification: Obtain formation curvature along the seismic layer; d. Fold property judgment: Determine whether it is an anticline or a syncline based on the positive and negative changes in curvature; e. Extract amplitude difference: Calculate the amplitude difference in all directions within the target layer range with the center of the micro-fold; f. Calculate the micro-fold amount: According to the three parameters of curvature positive and negative signs, curvature size and amplitude difference value, comprehensively calculate the micro-fold amount; g. Micro-fold classification: Classify the micro-fold according to the calculated micro-fold amount; h. Well leakage point prediction: According to Well trajectory coordinates, picking up possible well leakage point locations from micro-fold distribution data; Chinese patent publication CN111191836A discloses a well leakage prediction method, device and equipment, the method comprising: obtaining data to be predicted; the data to be predicted includes at least two of the inlet flow, outlet flow, total pool volume, standpipe pressure and drill bit torque of any drilling well; inputting the data to be predicted into a first feature extraction model, and obtaining a first feature vector of the data to be predicted after processing; the first feature extraction model is obtained by training an asymmetric convolutional neural network model using historical data with labels; based on the first feature vector of the data to be predicted and the first feature vector of each historical data, respectively calculating the first similarity between the data to be predicted and each historical data; based on the labels corresponding to the top N historical data with the highest first similarity, determining the well leakage prediction result corresponding to the data to be predicted.

[0004] There are two traditional methods for predicting lost circulation risk: (1) expert experience method, where experts make predictions based on experience by observing the formation and drilling parameters; and (2) theoretical calculation based on a small number of parameters using rock mechanics and fluid mechanics formulas.

[0005] The limitations of traditional methods are as follows: (1) Because there are many factors affecting well leakage, involving a large number of parameters such as formation, drilling, drilling fluid, etc., the changes in these parameters are random and accidental, and expert experience methods are difficult to accurately predict; (2) The theoretical formula calculation method is difficult to take into account the influencing factors, but there are errors in the acquisition of some parameters such as formation leakage rate and formation fracture width, and the accuracy of the calculation results is difficult to guarantee.

[0006] Existing methods have great limitations and it is difficult to accurately predict the drilling loss rate, which leads to complex well leakage, increases the plugging time, wastes drilling fluid and plugging materials, increases drilling costs, and easily causes complex situations downhole. Summary of the Invention

[0007] The purpose of the present invention is to solve the above-mentioned difficulties in the prior art and provide a drilling loss rate prediction method and system based on geological and engineering parameters. The method utilizes oilfield geological information, drilling parameters of existing wells, and lost circulation data of existing wells to establish a drilling loss rate prediction model for the well to be predicted. This model can achieve more accurate prediction and judgment of downhole loss rate, solve the problem of large errors in the drilling site expert experience method and theoretical formula method, improve the prediction accuracy of downhole loss rate, facilitate the early implementation of reasonable leakage prevention measures, and reduce the risk of leakage. The system meets the requirements of safe and efficient drilling and reduces drilling costs.

[0008] The present invention is achieved through the following technical solutions:

[0009] A first aspect of the present invention provides a method for predicting drilling loss rate based on geological and engineering parameters, the method comprising:

[0010] Step 1: Obtain parameters that affect the downhole loss rate and collect the drilling loss rate;

[0011] Step 2: Process the parameters that affect the downhole leakage rate to obtain processed parameters;

[0012] Step 3: Build a neural network;

[0013] Step 4: using the processed parameters and the drilling loss rate to train and verify the neural network to obtain a drilling loss rate prediction model;

[0014] Step 5: Use the loss rate prediction model to predict the drilling well to obtain the drilling loss rate of the drilling well to be predicted.

[0015] A further improvement of the present invention is that the operation of obtaining the parameters affecting the downhole leakage rate in step 1 includes:

[0016] 19 parameters that affect downhole leakage rate are obtained, as follows:

[0017] The following data are collected from the well logging data: vertical depth TVD, maximum horizontal principal stress SH, minimum horizontal principal stress Sh, fracture pressure Pf, pore pressure Pp, weight on bit WOB, rotational speed RPM, rate of penetration ROP, vertical pressure SP, drilling fluid displacement FR, drilling fluid density RHO, drilling fluid plastic viscosity PV, drilling fluid dynamic shear force YP, drilling fluid water loss FV, drilling fluid solid content SC, drilling fluid plugging material content LCMC, lithology LITH, where lithology LITH is an enumeration type;

[0018] The uniaxial compressive strength USC is calculated using the following formula:

[0019]

[0020] The tensile strength TSTR is calculated using the following formula:

[0021]

[0022] In the above two formulas, V cl is the formation shale content ratio obtained by logging, ρ is the formation rock density obtained by logging, V p is the P-wave velocity obtained from logging, V s is the shear wave velocity obtained from well logging.

[0023] A further improvement of the present invention is that the operation of step 2 includes:

[0024] The parameters affecting the downhole leakage rate are smoothed using the following formulas:

[0025]

[0026] Among them, x t For the original data before processing, X t is the data after smoothing, N is the step size, and the value range of N is 3-10;

[0027] The smoothed data is normalized so that the value of each parameter ranges from 0 to 1.

[0028] A further improvement of the present invention is that the operation of step three includes:

[0029] Establish a multi-layer BP neural network: The multi-layer BP neural network consists of an input layer, 2-6 hidden layers, and an output layer;

[0030] The number of neurons in the multi-layer BP neural network is set as follows:

[0031] The number of neurons in the input layer is 19;

[0032] The number of neurons in the last hidden layer is 1, and the number of neurons in each other hidden layer is given by the formula Determine, where i is the number of neurons in the input layer, j is the number of neurons in the output layer, and a is a constant between 1 and 10;

[0033] The number of neurons in the output layer is 1.

[0034] A further improvement of the present invention is that the multi-layer BP neural network includes one input layer, three hidden layers and one output layer.

[0035] A further improvement of the present invention is that the value of k is 10.

[0036] A further improvement of the present invention is that the operation of step five includes:

[0037] 19 parameters of the well to be logged are obtained, and the same operation as step 2 is performed on the 19 parameters to obtain processed parameters, and then the processed parameters are input into the drilling loss rate prediction model, and the drilling loss rate prediction model outputs the drilling loss rate of the well to be predicted.

[0038] A further improvement of the present invention is that the method further comprises, after step 5: determining the downhole loss risk based on the drilling loss rate of the well to be predicted, and taking measures to reduce the drilling loss rate if the drilling loss rate is greater than a set threshold;

[0039] The measures include: increasing the LCMC content of the drilling fluid plugging material, reducing the density of the drilling fluid, or reducing the drilling fluid displacement.

[0040] A second aspect of the present invention provides a drilling loss rate prediction system based on geological and engineering parameters, the system comprising:

[0041] A parameter acquisition unit, used to acquire parameters affecting the downhole leakage rate and collect the drilling leakage rate;

[0042] a data processing unit connected to the parameter acquisition unit, and configured to process the parameters affecting the downhole leakage rate to obtain processed parameters;

[0043] A neural network building unit, used for building a neural network;

[0044] A model acquisition unit is connected to the parameter acquisition unit, the data processing unit, and the neural network establishment unit, and is used to train and verify the neural network using the processed parameters and the drilling loss rate to obtain a drilling loss rate prediction model;

[0045] Prediction unit: connected to the model acquisition unit, used to predict the drilling loss rate of the predicted well using the leakage rate prediction model to obtain the drilling loss rate of the predicted well;

[0046] Output unit: connected to the prediction unit, used to output the drilling loss rate of the well to be predicted.

[0047] According to a third aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores at least one computer-executable program, and when the at least one program is executed by the computer, the computer executes the steps in the method for predicting drilling loss rate based on geological and engineering parameters.

[0048] Compared with the existing technology, the beneficial effects of the present invention are: the present invention realizes artificial intelligence prediction of drilling loss rate based on geological and engineering parameters, overcomes the shortcomings of the existing methods, and solves the problem of large errors in the prediction of drilling site loss rate based solely on expert experience and theoretical calculation methods. By improving the judgment accuracy of downhole loss rate, it is conducive to taking correct leakage prevention measures, reducing the risk of drilling leakage, and achieving the goal of safe and efficient drilling. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 BP neural network model structure;

[0050] Figure 2 BP neuron model;

[0051] Figure 3 A block diagram of the steps of the method of the present invention;

[0052] Figure 4 Schematic diagram of the composition structure of the system of the present invention. DETAILED DESCRIPTION

[0053] The present invention is further described in detail below with reference to the accompanying drawings:

[0054] To accurately predict the downhole loss rate, through extensive theoretical analysis and verification of measured data, the main factors affecting the downhole loss rate and the historical loss rate data of the drilled wells were screened out to form a training parameter set. Based on the above parameter set, a drilling loss rate prediction model for the well to be predicted was established using a multi-layer BP neural network algorithm.

[0055] BP neural network is one of the most widely used neural network algorithms. Its output results are transmitted by forward propagation, and the error is transmitted by back propagation. It usually consists of a three-layer structure consisting of 1 input layer, 1 hidden layer and 1 output layer.

[0056] Because drilling lost circulation is influenced by numerous factors, including formation and engineering, research has shown that increasing the number of neural network layers and optimizing the number of neurons is necessary to improve the accuracy of lost circulation prediction and meet field requirements. This invention, based on an optimized neural network algorithm model and utilizing geological information and drilling parameters from actual well drilling, achieves a more accurate prediction and assessment of downhole lost circulation rates.

[0057] The present invention provides a method for predicting the drilling loss rate based on geological and engineering parameters, including:

[0058] Step 1: Obtain parameters that affect the downhole loss rate and collect the drilling loss rate;

[0059] Step 2: Process the parameters that affect the downhole leakage rate to obtain processed parameters;

[0060] Step 3: Build a neural network;

[0061] Step 4: using the processed parameters and the drilling loss rate to train and verify the neural network to obtain a drilling loss rate prediction model;

[0062] Step 5: Use the loss rate prediction model to predict the drilling well to obtain the drilling loss rate of the drilling well to be predicted.

[0063] The specific implementation steps are as follows:

[0064] (1) The main factors affecting the downhole loss rate are determined to be 19: vertical depth TVD (m), lithology LITH (enumerated type), maximum horizontal principal stress SH (MPa), minimum horizontal principal stress Sh (MPa), uniaxial compressive strength USC (MPa), tensile strength TSTR (MPa), fracture pressure Pf (MPa), pore pressure Pp (MPa); bit weight WOB (kN), rotation speed RPM (rpm), drilling rate ROP (m / h), vertical pressure SP (MPa), drilling fluid displacement FR (L / s); drilling fluid density RHO (g / cm 3 ), drilling fluid plastic viscosity PV (mPa·s), drilling fluid dynamic shear force YP (Pa), drilling fluid water loss FV (ml), drilling fluid solid content SC (%), drilling fluid plugging material content LCMC (%). The target value of the model calculation is determined to be the drilling loss rate LR (m 3 / h);

[0065] Among them, the uniaxial compressive strength is calculated according to the following formula:

[0066]

[0067] The tensile strength is calculated using the following formula:

[0068]

[0069] In the above two formulas, V cl is the formation shale content ratio obtained by logging, ρ is the formation rock density obtained by logging, V p is the P-wave velocity obtained from logging, V s is the shear wave velocity obtained from well logging.

[0070] The lithology is an enumerated type and can be defined as {sandstone = 0, mudstone = 1, shale = 2, dolomite = 3, limestone = 4, salt-gypsum rock = 5, igneous rock = 6}.

[0071] The values of other parameters can be obtained from the logging data collected on site.

[0072] (2) Set the input parameter inputData1 of the training set and collect the data set x containing 19 parameters from the target block i , as the data for neural network training, the parameter set contains N 19-dimensional vectors x i =[TVD i ,LITH i ,SH i ,Sh i ,USC i ,TSTR i ,Pf i ,Pp i ,WOB i ,RPM i ,ROP i ,SP i ,FR i ,RHO i ,PV i ,YP i ,FV i ,SC i ,LCMC i ], (i=1->N); Set the output target value Target1 of the training set, the target value is the historical leakage rate LR of each well depth that has been drilled i (m 3 / h) is 1-dimensional data, y i =[LR i ], (i=1->N), if there is no leakage at a certain well depth, the leakage rate of the well depth is defined as 0 (m 3 / h);

[0073] (3) Perform smoothing preprocessing on the training set data.

[0074] Since the sensors at the drilling site have certain errors, the values of various parameters often fluctuate. It is necessary to smooth the data in order to more accurately represent the overall trend of the data. The specific method is data moving average processing: x t For the original data before processing, X t is the data after smoothing, N is the step size, the value range of N is 3-10, and the preferred value of N is 5.

[0075] (4) Perform normalization preprocessing on the training set data, dividing each data by the maximum value of the data, that is, X N =X / X max , so that the value range of each parameter is between 0 and 1;

[0076] (5) Establish a multi-layer BP neural network with 4 to 8 layers, preferably a 5-layer BP neural network:

[0077] like Figure 1 and Figure 2 As shown in the figure, the multi-layer BP neural network designed by the present invention for lost circulation prediction consists of an input layer, 2-6 hidden layers (preferably 3 layers), and an output layer. The input layer receives data, and the output layer outputs data. Neurons in the previous layer connect to neurons in the next layer, collecting information transmitted by neurons in the previous layer and passing the value to the next layer through an "activation function."

[0078] The number of neurons is set as follows: the number of neurons in the input layer is the same as the number of input feature variables, that is, 19; the number of neurons in the hidden layer output layer (that is, the last hidden layer) is the same as the number of output target values, that is, 1; the number of neurons in each other hidden layer is determined by the formula Determine, i is the number of neurons in the input layer, j is the number of neurons in the output layer, a is a constant between 1 and 10, so the number of neurons k in each hidden layer ranges from 5 to 15, and the preferred k is 10. The number of neurons in the output layer is 1.

[0079] The transfer function is set as follows: the activation function from the input layer to the hidden layer is the logsig function, the activation function from the previous hidden layer to the next hidden layer is the logsig function, and the activation function from the hidden layer to the output layer is the linear function (purelin function);

[0080] The training method is set to Levenberg-Marquardt algorithm (trainlm), which is the fastest training algorithm for medium-sized networks. The neural network toolbox in MATLAB software is used to establish the multi-layer structure of the BP neural network lost circulation prediction model as follows:

[0081] net=newff(inputData1,Target1,[19,10,10,10,1],{'logsiglogsig',

[0082] 'logsig','logsig','purelin'},'trainlm');

[0083] Among them, newff() is the BP neural network function (this function can be used in the Matlab software, which will not be repeated here), inputData1 is the input data, Target1 is the output target data, [19, 10, 1] means that the number of input layer nodes of the BP neural network is 19, the number of hidden layer nodes is 10, and the number of output layer nodes is 1 logsiglogsig.

[0084] (6) Training the neural network: Set the neural network training parameters and perform training:

[0085] Training goal: mean square error less than 0.0001;

[0086] net.trainparam.goal=0.0001;

[0087] Show results every 100 training times

[0088] net.trainParam.show=100;

[0089] Maximum number of training sessions: 2000.

[0090] net.trainparam.epochs=2000;

[0091] Call the neural network's train function to train the network

[0092] [net,tr]=train(net,inputData1,Target1);

[0093] (7) Verify the neural network model to obtain the drilling loss rate prediction model: the normalized verification parameter set input parameter inputData2 and output target value Target2, call the MATLAB neural network sim function to obtain the neural network model's predicted value simout2 for inputData2:

[0094] simout2=sim(net,inputData2);

[0095] Compare the verified target value Target2 and the neural network predicted simout2;

[0096] If the lost circulation prediction accuracy does not meet the requirements, the neural network parameters are modified and retrained until the accuracy meets the requirements. The resulting neural network is then used as the drilling lost circulation rate prediction model. The methods for training and validating neural networks are mature technologies and will not be detailed here.

[0097] (8) Using the drilling loss rate prediction model, a new set of normalized input parameters inputData3 is predicted to obtain the loss rate value simout3:

[0098] simout3=sim(net,inputData3);

[0099] (9) According to the drilling loss rate value simout3 predicted by the drilling loss rate prediction model, the downhole loss risk is judged. If the loss rate value is large (i.e., greater than the set threshold), measures are taken, such as increasing the drilling fluid plugging material content LCMC, reducing the drilling fluid density, and reducing the drilling fluid displacement.

[0100] The present invention also provides a prediction of drilling loss rate based on geological and engineering parameters, such as Figure 4 As shown, the system includes:

[0101] The parameter acquisition unit 10 is used to acquire parameters that affect the downhole leakage rate and collect the drilling leakage rate;

[0102] The data processing unit 20 is connected to the parameter acquisition unit 10 and is used to process the parameters affecting the downhole leakage rate to obtain processed parameters;

[0103] A neural network establishing unit 30, for establishing a neural network;

[0104] The model acquisition unit 40 is connected to the parameter acquisition unit 10, the data processing unit 20, and the neural network establishment unit 30, respectively, and is used to train and verify the neural network using the processed parameters and the drilling loss rate to obtain a drilling loss rate prediction model;

[0105] The prediction unit 50 is connected to the model acquisition unit 40 and is used to predict the drilling loss rate of the well to be predicted using the leakage rate prediction model to obtain the drilling loss rate of the well to be predicted.

[0106] Output unit 60: connected to prediction unit 50, used to output the drilling loss rate of the well to be predicted.

[0107] The embodiments of the present invention are as follows:

[0108] Example 1:

[0109] like Figure 3 As shown, the method includes:

[0110] (1) Determine 19 geological and engineering parameters that affect downhole drilling fluid loss rates;

[0111] (2) Set the input parameter inputData of the training set and collect the data set x containing 19 parameters from the target block i , as the data for neural network training, the parameter data set contains N 19-dimensional vectors x i =[TVD i ,LITH i ,SH i ,Sh i ,USC i ,TSTR i ,Pf i ,Pp i ,WOB i ,RPM i ,ROP i ,SP i ,FR i ,RHO i ,PV i ,YP i ,FV i ,SC i ,LCMC i ], (i=1->N), where N=52036;

[0112] (3) Perform smoothing preprocessing on the training set data. Since there are certain errors in the drilling site sensors themselves, the values of various parameters often fluctuate. The data needs to be smoothed to more accurately represent the overall trend of the data. The processing method is data moving average processing: x t For the original data before processing, X t is the data after smoothing, N is the step size, N ranges from 3 to 10, and the preferred value of N is 5. The processed data are shown in Table 1.

[0113]

[0114] Table 1

[0115] According to whether the well has actually been lost and the loss rate when it occurs, the output target value outputTarget of the training set and the target value drilling loss rate y are set. i is 1-dimensional data, y i =[LR i ], (i=1->N), where N=52036, as shown in Table 2.

[0116] Parameter name code name 1 2 3 4 5 6 7 8 9 … 52036 <![CDATA[True vertical depth TVD( m )]]> TVD 3520 3525 3530 3535 3540 3545 3550 3555 3560 … 6780 <![CDATA[Underground leakage rate ( m 3 / h)]]> LR 0 0 35 0 0 0 16 0 10 … 0

[0117] Table 2

[0118] (4) The training set data is normalized and preprocessed, that is, the 19 main factors affecting the drilling loss rate of the downhole formation are normalized, that is, each parameter is divided by the maximum value of the parameter, so that the value range of each parameter is between 0 and 1, as shown in Tables 3 and 4.

[0119] Parameter name code name 1 2 3 4 5 6 7 8 9 … 52036 <![CDATA[True vertical depth TVD( m )]]> TVD 0.52 0.52 0.52 0.52 0.52 0.52 0.52 0.52 0.53 … 1.00 Lithology LITH (enumeration type) LITH 0.00 0.00 0.25 0.25 0.25 0.50 0.50 0.25 0.00 … 1.00 <![CDATA[Maximum horizontal principal stress SH (MP a )]]> SH 0.49 0.42 0.52 0.42 0.57 0.53 0.54 0.51 0.57 … 1.00 <![CDATA[Minimum horizontal principal stress Sh (MP a )]]> Sh 0.46 0.42 0.48 0.43 0.52 0.52 0.54 0.52 0.53 … 1.00 <![CDATA[Uniaxial compressive strength USC (MP a )]]> USC 0.94 0.73 0.63 0.59 0.58 0.56 0.53 0.52 0.82 … 1.00 <![CDATA[Tensile strength TSTR (MP a )]]> TSTR 0.91 0.87 0.71 0.71 0.60 0.63 0.57 0.52 0.92 … 1.00 <![CDATA[Rupture pressure Pf (MP a )]]> Pf 0.50 0.49 0.52 0.50 0.51 0.52 0.52 0.51 0.52 … 1.00 <![CDATA[Pore pressure P p (MPa a )]]> <![CDATA[P p ]]> 0.51 0.53 0.51 0.53 0.53 0.53 0.52 0.52 0.53 … 1.00 Weight on bit (WOB) WOB 0.99 1.00 0.95 0.94 0.99 0.99 0.96 1.00 0.98 … 0.97 <![CDATA[Rotational speed RPM( rpm )]]> RPM 0.83 0.83 0.83 0.83 0.83 0.83 0.83 1.00 1.00 … 1.00 <![CDATA[Rate of Penetration ROP( m / h)]]> ROP 0.97 0.92 0.95 1.00 0.92 0.94 0.97 0.97 0.94 … 0.95 <![CDATA[Standing pressure SP (MP a )]]> SP 0.79 0.79 0.79 0.79 0.79 0.79 0.82 0.82 0.82 … 1.00 <![CDATA[Drilling fluid displacement FR (L / s )]]> FR 0.98 0.98 0.98 0.98 0.98 0.98 1.00 1.00 1.00 … 0.98 <![CDATA[Drilling fluid density RHO( g / cm 3 )]]> RHO 0.98 0.98 0.98 0.98 0.98 0.97 0.97 0.97 0.97 … 1.00 <![CDATA[Plastic viscosity PV of drilling fluid ( m P a / s )]]> PV 1.00 0.96 0.92 0.92 0.92 0.96 0.96 1.00 1.00 … 1.00 <![CDATA[Yield point YP of drilling fluid (P a )]]> YP 1.00 1.00 0.88 0.88 0.88 0.88 0.88 1.00 1.00 … 1.00 <![CDATA[Filter loss of drilling fluid FV( m l)]]> FV 1.00 1.00 1.00 1.00 1.00 0.95 0.95 0.95 0.95 … 0.95 Drilling fluid solid content SC (%) SC 0.93 0.93 0.93 0.93 0.93 1.00 1.00 1.00 1.00 … 1.00 Drilling fluid plugging material content LCMC (%) LCMC 0.60 0.60 0.60 0.60 0.60 0.60 0.60 0.80 0.80 … 1.00

[0120] Table 3

[0121] Parameter name code name 1 2 3 4 5 6 7 8 9 … 52036 <![CDATA[Underground leakage rate ( m 3 / h)]]> LR 0.00 0.00 1.00 0.00 0.00 0.00 0.46 0.00 0.29 … 0.00

[0122] Table 4

[0123] (5) Establish a BP neural network: Set the number of neurons: the number of neurons in the input layer, hidden layer, and output layer are [19, 10, 10, 10, 1] (i.e., 1 input layer, 3 hidden layers, and 1 output layer); set the transfer function: the logsig function from the input layer to the hidden layer, and the purelin function from the hidden layer to the output layer; set the training method to trainlm; set the weight value of each parameter;

[0124] net=newff(inputData,outputTarget,[19,10,10,10,101],{'logsig','logsig','logsig','purelin'},'trainlm');

[0125] inputWeights=net.IW{M1,M2,M3,M4,M5,M6};

[0126] Among them, newff() is the BP neural network function, inputData is the input data, outputTarget is the output target data, [19, 10, 10, 10, 101] means that the BP neural network has 19 input layer nodes, 3 hidden layers, 10 nodes in each hidden layer, and 1 output layer node, {'logsig', 'purelin'} means that the transfer functions from the input layer to the hidden layer and from the hidden layer to the output layer use the logsig function and the purelin function respectively; inputWeights is the weight value setting of each parameter.

[0127] (6) Training the neural network: Set the neural network training parameters and perform training:

[0128] Training goal: mean square error less than 0.0001;

[0129] net.trainparam.goal=0.0001;

[0130] Show results every 400 training times

[0131] net.trainParam.show=400;

[0132] Maximum number of training times: 15,000.

[0133] net.trainparam.epochs=15000;

[0134] Call the neural network's train function to train the network

[0135] [net,tr]=train(net,inputData,outputData);

[0136] (7) Verify the neural network model: Collect and organize another part of the drilled data as a validation set to verify whether the model's calculation results are consistent with the actual situation. The normalized validation parameter set includes the input parameter inputData2 and the output target value Target2. Call the MATLAB neural network sim function to obtain the neural network model's predicted value simout2 for inputData2:

[0137] simout2=sim(net,inputData2);

[0138] Comparing the parameter prediction results simout2 of the neural network model for the validation set with the target value Target2 of the validation set, it can be seen from the comparison of Table 5 and Table 6 that the average consistency rate between the results predicted by the neural network model and the actual well loss rate is greater than 90%, indicating that the neural network model can be used to predict the downhole loss rate in this area.

[0139] Parameter name code name 1 2 3 4 5 6 7 8 9 … 3130 <![CDATA[True Vertical Depth (TVD) m )]]> TVD 4120 4120 4120 4120 4120 4120 4120 4120 4120 … 6920 Lithology LITH (enumeration type) LITH 1 0 0 1 1 2 1 1 2 … 4 <![CDATA[Maximum horizontal principal stress SH (MP a )]]> SH 65.25 55.48 68.96 56.12 76.14 70.64 72.03 68.28 76.49 … 132.98 <![CDATA[Minimum horizontal principal stress Sh (MP a )]]> Sh 50.16 45.09 51.89 45.54 55.18 55.69 58.02 55.21 57.12 … 106.80 <![CDATA[Uniaxial compressive strength USC (MP a )]]> USC 64.95 51.66 44.35 41.46 40.76 39.28 37.57 36.59 57.50 … 70.06 <![CDATA[Tensile strength TSTR (MP a )]]> TSTR 7.46 6.60 5.46 5.90 4.79 4.70 4.65 3.92 7.27 … 7.29 <![CDATA[Rupture pressure Pf (MP a )]]> Pf 75.44 73.49 77.01 73.78 76.33 77.16 77.09 75.79 77.48 … 148.02 <![CDATA[Pore pressure P p (MP a )]]> <![CDATA[P p ]]> 40.39 41.59 40.54 41.19 40.90 41.31 40.48 40.14 41.36 … 77.60 Weight on bit (WOB) WOB 145 145 145 144 148 141 144 145 145 … 154 <![CDATA[Rotational speed RPM( rpm )]]> RPM 50 50 50 50 50 50 55 55 55 … 55 <![CDATA[Rate of Penetration ROP( m / h)]]> ROP 2.41 2.27 2.39 2.50 2.31 2.26 2.33 2.47 2.39 … 2.37 <![CDATA[Standing pressure SP (MP a )]]> SP 23 23 22 22 23 22 23 23 22 … 25 <![CDATA[Drilling fluid displacement FR (L / s )]]> FR 55 55 55 55 55 55 56 56 56 … 55 <![CDATA[Drilling fluid density RHO( g / cm 3 )]]> RHO 1.35 1.35 1.35 1.35 1.35 1.34 1.34 1.34 1.34 … 1.35 <![CDATA[Plastic viscosity PV of drilling fluid ( m P a / s )]]> PV 25 24 23 23 23 24 24 25 25 … 25 <![CDATA[Yield point YP of drilling fluid (P a )]]> YP 7 8 7 7 8 8 8 8 8 … 7 <![CDATA[Filter loss of drilling fluid FV( m l)]]> FV 3.8 4.0 4.0 4.0 3.8 3.6 3.6 3.6 3.6 … 3.8 Drilling fluid solid content SC (%) SC 18 18 18 18 18 19 19 19 19 … 20 Drilling fluid plugging material content LCMC (%) LCMC 3 3 3 4 4 4 4 4 4 … 5

[0140] Table 5

[0141]

[0142] Table 6

[0143] (8) Prediction of the drilling loss rate of the well being drilled: The data of the well being drilled are collected and sorted as the input parameters of the prediction set, as shown in Table 7, to predict the possible drilling loss rate of the well being drilled. The neural network is used to predict the input parameter inputData3 of the prediction set, and the predicted value simout3 is obtained. As can be seen from Table 8, it is predicted that there is a 12m loss rate at a vertical depth of 3525m in the well. 3 / h of lost circulation, and 19m at a vertical depth of 3555m 3 / h of well leakage:

[0144] simout3=sim(net,inputData3);

[0145] Parameter name code name 1 2 3 4 5 6 7 8 9 … 468 <![CDATA[True vertical depth TVD( m )]]> TVD 3950 3955 3960 3965 3970 3975 3980 3985 3990 … 6290 Lithology LITH (enumeration type) LITH 1 0 0 1 1 2 1 1 2 … 4 <![CDATA[Maximum horizontal principal stress SH (MP a )]]> SH 64.90 56.04 69.44 56.49 76.32 69.96 71.81 67.52 76.10 … 132.84 <![CDATA[Minimum horizontal principal stress Sh (MP a )]]> Sh 50.31 44.60 51.58 45.80 55.32 55.94 57.50 55.35 56.82 … 107.55 <![CDATA[Uniaxial compressive strength USC (MP a )]]> USC 65.23 51.60 44.57 41.20 40.89 39.89 37.83 36.15 57.17 … 70.04 <![CDATA[Tensile strength TSTR (MP a )]]> TSTR 6.73 6.60 6.11 5.69 5.17 4.57 4.10 3.82 7.21 … 8.16 <![CDATA[Rupture pressure Pf (MP a )]]> Pf 75.60 74.00 76.90 73.98 75.70 76.93 76.82 75.77 77.49 … 148.51 <![CDATA[Pore pressure P p (MP a )]]> <![CDATA[P p ]]> 39.93 41.08 40.06 41.96 41.23 41.22 40.60 40.81 41.62 … 77.68 Weight on bit (WOB) WOB 153 145 149 149 149 150 153 154 147 … 163 <![CDATA[Rotational speed RPM( rpm )]]> RPM 50 50 50 50 50 50 55 55 55 … 55 <![CDATA[Rate of Penetration ROP( m / h)]]> ROP 2.42 2.28 2.33 2.50 2.37 2.41 2.30 2.41 2.24 … 2.31 <![CDATA[Standing pressure SP (MP a )]]> SP 21 21 22 22 23 23 23 23 23 … 25 <![CDATA[Drilling fluid displacement FR (L / s )]]> FR 55 55 55 55 55 55 56 56 56 … 55 <![CDATA[Drilling fluid density RHO( g / cm 3 )]]> RHO 1.33 1.33 1.33 1.33 1.34 1.34 1.34 1.34 1.34 … 1.32 <![CDATA[Plastic viscosity PV of drilling fluid ( m P a / s )]]> PV 23 23 23 23 23 24 24 25 25 … 25 <![CDATA[Yield point YP of drilling fluid (P a )]]> YP 7 7 7 7 8 8 8 8 9 … 8 <![CDATA[Filter loss of drilling fluid FV( m l)]]> FV 3.8 4.0 4.0 4.0 3.8 3.6 3.6 3.6 3.6 … 3.8 Drilling fluid solid content SC (%) SC 16 16 16 16 16 16 16 17 17 … 17 Drilling fluid plugging material content LCMC (%) LCMC 3 3 3 4 4 4 4 4 4 … 5

[0146] Table 7

[0147] Parameter name code name 1 2 3 4 5 6 7 8 9 … 468 <![CDATA[True vertical depth TVD( m )]]> TVD 3520 3525 3530 3535 3540 3545 3550 3555 3560 … 6780 <![CDATA[Underground leakage rate ( m 3 / h)]]> LR 0 12 0 0 0 0 0 19 0 … 0

[0148] Table 8

[0149] (9) The prediction results calculated by the neural network model (12m at a vertical depth of 3525m) 3 / h of lost circulation, and 19m at a vertical depth of 3555m 3 / h of well leakage), it is judged that the risk of downhole leakage is high, and measures should be taken in advance before drilling to a vertical depth of 3525m. Anti-leakage measures such as increasing the content of anti-leakage plugging materials in the drilling fluid can be taken.

[0150] Example 2:

[0151] The number of hidden layers in Example 1 is changed to 2, and other settings are the same as in Example 1.

[0152] Example 3:

[0153] The number of hidden layers in Example 1 is changed to 4, and other settings are the same as in Example 1.

[0154] Example 4:

[0155] The number of neurons in each hidden layer in Example 1 is changed to 5, and other settings are the same as in Example 1.

[0156] Example 5:

[0157] The number of neurons in each hidden layer in Example 1 is changed to 15, and other settings are the same as in Example 1.

[0158] Comparative Example 1:

[0159] The number of hidden layers in Example 1 is changed to 1, and other settings are the same as in Example 1.

[0160] Comparative Example 2:

[0161] The number of hidden layers in Example 1 is changed to 7, and other settings are the same as in Example 1.

[0162] Comparative Example 3:

[0163] The number of neurons in each hidden layer in Example 1 is changed to 2, and other settings are the same as in Example 1.

[0164] Comparative Example 4:

[0165] The number of neurons in each hidden layer in Example 1 is changed to 20, and other settings are the same as in Example 1.

[0166] Evaluation indicators:

[0167] Prediction accuracy and model training time are selected as evaluation indicators of model effectiveness.

[0168] As shown in Table 9, the model with 3 hidden layers and 10 neurons in each hidden layer set in Example 1 has the highest prediction accuracy, and the training time is controlled within a relatively short time range.

[0169]

[0170]

[0171] Table 9

[0172] The present invention provides a method and system for predicting drilling loss rates based on geological and engineering parameters. These methods overcome the shortcomings of existing methods and fully utilize oilfield geological information, past well data, and actual drilling data to determine 19 geological, rock mechanics, and engineering parameters that influence drilling loss rates, including vertical depth, lithology, uniaxial compressive strength, weight on bit, rotational speed, drilling fluid displacement, and drilling fluid density. Training and validation parameter sets are established based on past well data, and a prediction parameter set is established based on ongoing drilling data. The data is then smoothed and normalized using a moving average. A multi-layer BP neural network is established, consisting of one input layer, two to six hidden layers, and one input layer. Model parameters are set and the neural network is trained based on the training set until convergence accuracy meets the required accuracy. Model calculations are performed using the validation set to determine whether the prediction accuracy meets the required accuracy. Based on the prediction parameter set, the neural network model is used to predict the input parameter set to be predicted, resulting in a predicted loss rate value. The predicted loss rate value is used to determine the magnitude of the downhole loss risk. If the loss rate value is high, preventive measures are implemented. Through computer model calculations based on neural network algorithms, the prediction accuracy of downhole leakage rate can be improved, and the problem of large prediction errors of drilling site expert experience and theoretical formula methods can be solved. This is conducive to taking reasonable leakage prevention measures in advance, reducing the risk of leakage, and achieving the goal of safe and efficient drilling. It is expected to significantly reduce the complexity and time of well leakage.

[0173] Finally, it should be noted that the above technical solution is only one embodiment of the present invention. For those skilled in the art, it is easy to make various types of improvements or modifications based on the application methods and principles disclosed in the present invention, and it is not limited to the method described in the above specific embodiment of the present invention. Therefore, the method described above is only preferred and does not have a restrictive meaning.

Claims

1. A method for predicting drilling loss rate based on geological and engineering parameters, characterized by: The method comprises: Step 1: Obtain parameters that affect the downhole loss rate and collect the drilling loss rate; Step 2: Process the parameters that affect the downhole leakage rate to obtain processed parameters; Step 3: Build a neural network; Step 4: using the processed parameters and the drilling loss rate to train and verify the neural network to obtain a drilling loss rate prediction model; Step 5: Use the loss rate prediction model to predict the drilling well to obtain the drilling loss rate of the well to be predicted; The operation of step three includes: Establish a multi-layer BP neural network: The multi-layer BP neural network consists of an input layer, 2-6 hidden layers, and an output layer; The number of neurons in the multi-layer BP neural network is set as follows: The number of neurons in the input layer is 19; The number of neurons in the last hidden layer is 1, and the number of neurons in each other hidden layer is given by the formula Determine, where i is the number of neurons in the input layer, j is the number of neurons in the output layer, and a is a constant between 1 and 10; The number of neurons in the output layer is 1; The operation of obtaining the parameters affecting the downhole leakage rate in step 1 includes: 19 parameters that affect downhole leakage rate are obtained, as follows: The following data are collected from the well logging data: vertical depth TVD, maximum horizontal principal stress SH, minimum horizontal principal stress Sh, fracture pressure Pf, pore pressure Pp, weight on bit WOB, rotational speed RPM, drilling rate ROP, vertical pressure SP, drilling fluid displacement FR, drilling fluid density RHO, drilling fluid plastic viscosity PV, drilling fluid dynamic shear force YP, drilling fluid water loss FV, drilling fluid solid content SC, drilling fluid plugging material content LCMC, lithology LITH, uniaxial compressive strength USC, tensile strength TSTR, where lithology LITH is an enumeration type; The uniaxial compressive strength USC is calculated using the following formula: The tensile strength TSTR is calculated using the following formula: In the above two formulas, V cl is the formation shale content ratio obtained by logging, ρ is the formation rock density obtained by logging, V p is the P-wave velocity obtained from logging, V s is the shear wave velocity obtained from well logging; After step five, the method further includes: determining the downhole loss risk based on the drilling loss rate of the predicted well; if the drilling loss rate is greater than a set threshold, taking measures to reduce the drilling loss rate; the measures include: increasing the LCMC content of the drilling fluid plugging material, reducing the drilling fluid density, or reducing the drilling fluid displacement.

2. The method for predicting drilling loss rate based on geological and engineering parameters according to claim 1, characterized in that: The operation of step 2 includes: The parameters affecting the downhole leakage rate are smoothed using the following formulas: Among them, x t For the original data before processing, X t is the data after smoothing, N is the step size, and the value range of N is 3-10; The smoothed data is normalized so that the value of each parameter ranges from 0 to 1.

3. The method for predicting drilling loss rate based on geological and engineering parameters according to claim 1, characterized in that: The multi-layer BP neural network includes one input layer, three hidden layers and one output layer.

4. The method for predicting drilling loss rate based on geological and engineering parameters according to claim 1, characterized in that: The value of k is 10.

5. The method for predicting drilling loss rate based on geological and engineering parameters according to claim 1, characterized in that: The operation of step five includes: 19 parameters of the well to be logged are obtained, and the same operation as step 2 is performed on the 19 parameters to obtain processed parameters, and then the processed parameters are input into the drilling loss rate prediction model, and the drilling loss rate prediction model outputs the drilling loss rate of the well to be predicted.

6. A drilling loss rate prediction system based on geological and engineering parameters, for implementing the method according to any one of claims 1 to 5, characterized in that: The system comprises: A parameter acquisition unit, used to acquire parameters affecting the downhole leakage rate and collect the drilling leakage rate; a data processing unit connected to the parameter acquisition unit, and configured to process the parameters affecting the downhole leakage rate to obtain processed parameters; A neural network building unit, used for building a neural network; A model acquisition unit is connected to the parameter acquisition unit, the data processing unit, and the neural network establishment unit, and is used to train and verify the neural network using the processed parameters and the drilling loss rate to obtain a drilling loss rate prediction model; Prediction unit: connected to the model acquisition unit, used to predict the drilling loss rate of the predicted well using the leakage rate prediction model to obtain the drilling loss rate of the predicted well; Output unit: connected to the prediction unit, used to output the drilling loss rate of the well to be predicted.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one computer-executable program, which, when executed by the computer, enables the computer to perform the steps of the method for predicting drilling loss rate based on geological and engineering parameters as described in any one of claims 1 to 5.

Images