Online prediction method and system for drilling rate during drilling process based on multi-source information fusion

Abstract

The present invention proposes a method and system for online drilling speed prediction during the drilling process based on multi-source information fusion. The method comprises: a drilling data preprocessing stage, in which drilling pressure, rotational speed, and torque are filtered; a drilling speed modeling optimization stage, in which a hybrid bat algorithm is used to improve the model structure of a restricted Boltzmann machine and a back-propagation neural network, forming a novel hybrid bat algorithm optimization-restricted Boltzmann machine-back-propagation neural network, and using this algorithm to establish an offline drilling speed prediction model for complex formation environments; and a drilling speed prediction model update stage, in which lithologic changes and time intervals are used as update conditions, and a sliding window strategy is used based on the fusion of multi-source formation and drilling information to update the drilling speed prediction model, thereby achieving online drilling speed prediction. The present invention has the beneficial effects of helping to shorten the drilling cycle, reduce operating costs, achieve high-precision drilling speed prediction, and is effective, laying an important foundation for intelligent optimization and control of the drilling process.

Description

Technical Field

[0001] The present invention relates to the field of geological drilling engineering, and in particular to a method and system for online prediction of drilling speed during a drilling process based on multi-source information fusion. Background Art

[0002] my country's total resource and energy consumption has increased annually, and the supply-demand imbalance for strategic resources and energy has surged, with a high degree of external dependence. Ensuring resource and energy security is crucial for national economic development and strategic security. With the continued exploitation of shallow and mid-level resources and the discovery of vast quantities of deeply buried resources, deep geological exploration and development have become inevitable. However, the deep-earth environment is complex and harsh, with diverse lithologies. The frequent encounters with alternating soft and hard formations and rock fragmentation lead to significant drilling characteristics, such as multiple source variables, low-quality drilling data, strong nonlinearity, and dramatic lithologic variations. These characteristics severely impact drilling efficiency and overall profitability. Furthermore, drilling rate is a key parameter determining drilling efficiency. Establishing an accurate drilling rate prediction model can help reduce drilling cycle time and costs. Therefore, research on drilling rate prediction for deep geological drilling processes needs to be accelerated.

[0003] Existing ROP prediction research can be categorized into two types: offline and online. Offline ROP prediction models utilize static modeling using industrial data from nearby well sites or regions, achieving good prediction results when the formation environment remains relatively unchanged. Online ROP prediction models dynamically update models using streaming industrial data during the drilling process, offering real-time prediction capabilities and high accuracy. However, these models often use time intervals or drilling depths as update indicators, fail to fully consider formation lithology information, and remain ineffective in addressing complex formation environments. Therefore, to address these current challenges, a high-precision online ROP prediction model that comprehensively integrates formation and drilling information is needed to provide technical support for intelligent optimization and control of deep geological drilling processes.

[0004] The above content is only used to assist in understanding the technical solution of the present invention and does not constitute an admission that the above content is prior art. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention proposes an online prediction method and system for drilling speed during the drilling process based on multi-source information fusion. The method mainly uses the hybrid bat algorithm to improve the model structure of the restricted Boltzmann machine and the back propagation neural network, and establishes an online prediction model of drilling speed by fusing multi-source information of the formation and drilling. The prediction accuracy and capture ability of the model can be effectively improved, thereby achieving energy saving and efficiency improvement in the drilling process.

[0006] To achieve the above technical objectives, the present invention adopts a technical solution: an online prediction method for drilling speed during drilling based on multi-source information fusion, comprising the following steps:

[0007] S1: Selecting drilling pressure, rotational speed and torque as inputs of the drilling speed prediction model, using outlier removal and wavelet filtering analysis to screen and filter the drilling data, using lithology identification technology to obtain the formation drillability identification value and judge the formation lithology changes;

[0008] S2: Introducing a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network algorithm, through three steps: hybrid bat algorithm optimization, restricted Boltzmann machine training, and back propagation neural network training and reverse fine-tuning, to establish an offline drilling rate prediction model for complex formation environments;

[0009] S3: Using lithologic changes and time intervals as update conditions for the ROP prediction model, a sliding window strategy is used based on formation drillability information and drilling process information to update the ROP prediction model in real time.

[0010] S4: Input the actual drilling pressure, rotation speed, and torque into the updated drilling speed prediction model to obtain the drilling speed at the next moment.

[0011] Furthermore, the specific process of step S1 is:

[0012] S11: Based on engineering experience, an outlier elimination method is used to screen the drilling data, wherein the drilling data includes bit weight, rotation speed, torque and drilling speed. The measurement range of the drilling data is:

[0013]

[0014] Wherein, WOB is the weight on bit, its unit is KN, RPM is the rotation speed, its unit is rpm, Torque is the torque, its unit is Nm, ROP is the drilling speed, its unit is cm / min;

[0015] S12: Use wavelet filtering to filter out the peaks and burrs in the drilling data. The wavelet transform expression is:

[0016]

[0017] Among them, W f (a, b) is the drilling characteristic information after wavelet forward transform, f(t) is the original drilling data, a is the expansion factor, b is the scale factor, t is the time, ψ() is the wavelet basis function, and the wavelet inverse transform expression is:

[0018]

[0019] Among them, g(t) is the drilling data after wavelet inverse transformation, c ψ is the wavelet factor;

[0020] S13: Mark the drilled cores and analyze them using lithology identification technology in a suitable environment to obtain formation drillability information. At the same time, determine in real time whether the formation lithology has changed based on the formation drillability information at consecutive moments.

[0021] Furthermore, step S2 specifically includes the following process:

[0022] S21: Introducing the trial-and-error method and hybrid bat algorithm to optimize the hyperparameters of the drilling speed prediction model, including batch size, number of training iterations, number of hidden layer neurons, learning rate, and momentum;

[0023] S22: Extract key features from drilling data by training a restricted Boltzmann machine. The restricted Boltzmann machine includes a visual layer v and a hidden layer h. The energy function under a given state (v, h) is defined as:

[0024]

[0025] Among them, E θ (v,h) is the energy function of the restricted Boltzmann machine, θ={w ij ,c i ,d j} are the model parameters of the restricted Boltzmann machine, p and q are the number of neurons in the visual layer and the hidden layer, c i and d j is the bias of the i-th visual layer neuron and the j-th hidden layer neuron, v i is the i-th visual layer neuron, h j is the jth hidden layer neuron, w ij is the weight between the i-th visual layer neuron and the j-th hidden layer neuron, i = 1, 2…, p, j = 1, 2…, q;

[0026] Given a drilling training set For nth k The log-likelihood function L(θ) of the restricted Boltzmann machine on K is:

[0027]

[0028] Where r is the rth drilling data sample, r=1,2…,n k , n k is the total number of samples, v r is the rth visual layer, P(v r ) is the visual layer v r The marginal probability distribution of ;

[0029] S23: The output features of the restricted Boltzmann machine are used as the input of the back-propagation neural network. The back-propagation neural network consists of two parts: forward propagation and back-propagation. The forward propagation can be expressed as:

[0030]

[0031] in, is the drilling speed prediction value output by the forward pass, f() is the activation function, g r is the output layer bias, k represents the kth hidden layer node, k=1,2…,n, n is the total number of hidden layer nodes, w kr is the connection weight between the hidden layer and the output layer, x k is the hidden layer node;

[0032] At the same time, the network structure is continuously adjusted using back propagation. The error function of the back propagation neural network can be expressed as:

[0033]

[0034] Among them, E y is the mean square error of the back propagation neural network, n k is the total number of samples, is the predicted value of drilling speed, y r is the actual value of drilling speed, and then the gradient descent method is used to update the parameters of the back propagation neural network (w kr ,g r ).

[0035] Furthermore, the specific process of step S3 is:

[0036] Taking lithologic changes and time intervals as model update conditions, the sliding window strategy is used to update the drilling rate prediction model in real time based on the fusion of multi-source information of formation and drilling. The model update condition is expressed as:

[0037]

[0038] Among them, u() is the switch variable for updating the drilling speed prediction model, Δu r is the state variable updated by lithologic change, Δu t is the state variable updated at time interval, r s and r s+1 is the formation drillability corresponding to two consecutive moments, r z is the lithologic change update threshold, t a and t a+1 is the time point corresponding to two consecutive moments, t b is the time interval update threshold, when Δu r Greater than or equal to r zWhen Δu t Equal to t b When the drilling speed prediction model is updated, it will continuously learn from the actual drilling data based on time.

[0039] When any of the above conditions is met, the drilling rate prediction model will be updated and predicted in real time.

[0040] Furthermore, the method also includes using a model evaluation index to evaluate the prediction performance of the method. The specific formula is:

[0041]

[0042]

[0043] Among them, RMSE is the root mean square error, NRMSE is the normalized root mean square error, and y r is the actual drilling speed, is the drilling speed prediction value, r is the rth drilling data sample, r=1,2…,n k , n k Indicates the total number of samples.

[0044] An online prediction system for drilling speed during drilling based on multi-source information fusion, comprising:

[0045] The data preprocessing module is used to select drilling pressure, rotational speed and torque as inputs to the drilling speed prediction model, screen and filter the drilling data using outlier removal and wavelet filtering analysis, and use lithology identification technology to obtain the formation drillability identification value and judge the changes in the formation lithology;

[0046] A model module was established to introduce a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network algorithm. Through three steps of hybrid bat algorithm optimization, restricted Boltzmann machine training, and back propagation neural network training and reverse fine-tuning, an offline drilling rate prediction model for complex formation environments was established.

[0047] The update model module is used to update the ROP prediction model in real time using lithologic changes and time intervals as the update conditions, and uses a sliding window strategy based on formation drillability information and drilling process information.

[0048] The drilling speed prediction module is used to input the actual drilling pressure, rotation speed and torque into the updated drilling speed prediction model to obtain the drilling speed at the next moment.

[0049] Furthermore, in the data preprocessing module, the process of preprocessing the drilling data is as follows:

[0050] S11: Based on engineering experience, an outlier elimination method is used to screen the drilling data, wherein the drilling data includes bit weight, rotation speed, torque and drilling speed. The measurement range of the drilling data is:

[0051]

[0052] Wherein, WOB is the weight on bit, its unit is KN, RPM is the rotation speed, its unit is rpm, Torque is the torque, its unit is Nm, ROP is the drilling speed, its unit is cm / min;

[0053] S12: Use wavelet filtering to filter out the peaks and burrs in the drilling data. The wavelet transform expression is:

[0054]

[0055] Among them, W f (a, b) is the drilling characteristic information after wavelet forward transform, f(t) is the original drilling data, a is the expansion factor, b is the scale factor, t is the time, ψ() is the wavelet basis function, and the wavelet inverse transform expression is:

[0056]

[0057] Among them, g(t) is the drilling data after wavelet inverse transformation, c ψ is the wavelet factor;

[0058] S13: Mark the drilled cores and analyze them using lithology identification technology in a suitable environment to obtain formation drillability information. At the same time, determine in real time whether the formation lithology has changed based on the formation drillability information at consecutive moments.

[0059] Furthermore, in the model building module, the model building process is as follows:

[0060] S21: Introducing the trial-and-error method and hybrid bat algorithm to optimize the hyperparameters of the drilling speed prediction model, including batch size, number of training iterations, number of hidden layer neurons, learning rate, and momentum;

[0061] S22: Extract key features from drilling data by training a restricted Boltzmann machine. The restricted Boltzmann machine consists of a visual layer v and a hidden layer h, and the energy function under a given state (v, h) is defined as:

[0062]

[0063] Among them, E θ (v,h) is the energy function of the restricted Boltzmann machine, θ={w ij ,c i ,d j} are the model parameters of the restricted Boltzmann machine, p and q are the number of neurons in the visual layer and the hidden layer, c i and d j is the bias of the i-th visual layer neuron and the j-th hidden layer neuron, v i is the i-th visual layer neuron, h j is the jth hidden layer neuron, w ij is the weight between the i-th visual layer neuron and the j-th hidden layer neuron, i = 1, 2…, p, j = 1, 2…, q;

[0064] Given a drilling training set For nth k The log-likelihood function L(θ) of the restricted Boltzmann machine on K is:

[0065]

[0066] Where r is the rth drilling data sample, r=1,2…,n k , n k is the total number of samples, v r is the rth visual layer, P(v r ) is the visual layer v r The marginal probability distribution of ;

[0067] S23: The output features of the restricted Boltzmann machine are used as the input of the back-propagation neural network. The back-propagation neural network consists of two parts: forward propagation and back-propagation. The forward propagation can be expressed as:

[0068]

[0069] in, is the drilling speed prediction value output by the forward pass, f() is the activation function, g r is the output layer bias, k represents the kth hidden layer node, k=1,2…,n, n is the total number of hidden layer nodes, w kr is the connection weight between the hidden layer and the output layer, x k is the hidden layer node;

[0070] At the same time, the network structure is continuously adjusted using back propagation. The error function of the back propagation neural network can be expressed as:

[0071]

[0072] Among them, E y is the mean square error of the back propagation neural network, n k is the total number of samples, is the predicted value of drilling speed, y ris the actual value of drilling speed, and then the gradient descent method is used to update the parameters of the back propagation neural network (w jr ,g r ).

[0073] Furthermore, in the update model module, the update model process is:

[0074] Taking lithologic changes and time intervals as model update conditions, the sliding window strategy is used to update the drilling rate prediction model in real time based on the fusion of multi-source information of formation and drilling. The model update condition is expressed as:

[0075]

[0076] Among them, u() is the switch variable for updating the drilling speed prediction model, Δu r is the state variable updated by lithologic change, Δu t is the state variable updated at time interval, r s and r s+1 is the formation drillability corresponding to two consecutive moments, r z is the lithologic change update threshold, t a and t a+1 is the time point corresponding to two consecutive moments, t b is the time interval update threshold, when Δu r Greater than or equal to r z When Δu t Equal to t b When the drilling speed prediction model is updated, it will continuously learn from the actual drilling data based on time.

[0077] When any of the above conditions is met, the drilling rate prediction model will be updated and predicted in real time.

[0078] Furthermore, in the drilling rate prediction module, the model evaluation index is used to evaluate the prediction performance of the method. The specific formula is:

[0079]

[0080]

[0081] Among them, RMSE is the root mean square error, NRMSE is the normalized root mean square error, and y r is the actual drilling speed, is the drilling speed prediction value, r is the rth drilling data sample, r=1,2…,n k , n k Indicates the total number of samples.

[0082] The beneficial effects of the present invention based on its technical solution are:

[0083] (1) Drilling data is screened and filtered by using outlier removal and wavelet filtering technology, and lithology identification technology is used to identify formation drillability information, providing effective data support for drilling rate modeling and model updating;

[0084] (2) An offline drilling rate prediction model is established using a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network. This model has strong learning and nonlinear fitting capabilities and is more suitable for complex formation environments.

[0085] (3) Taking the lithologic changes and time intervals as the model updating conditions, the drilling speed prediction model is updated in real time through the sliding window strategy based on the fusion of multi-source information of formation and drilling, which improves the prediction accuracy of drilling speed and is conducive to the application of the present invention in actual production. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 Flowchart of the method according to the embodiment of the present invention;

[0087] Figure 2 The proposed drilling rate prediction model framework;

[0088] Figure 3 Distribution map of drilling data of Dandong Ke Drilling;

[0089] Figure 4 Distribution of filtered Dandong Ke Drilling well drilling data;

[0090] Figure 5 Formation drillability distribution;

[0091] Figure 6 Comparison of actual drilling rate and predicted drilling rate. DETAILED DESCRIPTION

[0092] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments.

[0093] This embodiment provides an online prediction method for drilling speed during drilling based on multi-source information fusion, and its flow chart and framework diagram are shown in FIG. Figure 1 and Figure 2 The present invention divides the modeling process into three stages: drilling data preprocessing, optimized drilling speed modeling, and drilling speed prediction model update. In the first stage, outlier removal and wavelet filtering techniques are used to filter the drilling pressure, rotation speed, and torque. In the second stage, a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network is used to establish an offline drilling speed prediction model. In the third stage, lithology changes and time intervals are used as update conditions. Based on the fusion of multi-source formation and drilling information, a sliding window strategy is used to update the drilling speed prediction model to achieve online high-precision drilling speed prediction.

[0094] The specific steps are as follows:

[0095] S1: Selecting drilling pressure, rotational speed and torque as inputs of the drilling speed prediction model, using outlier removal and wavelet filtering analysis to screen and filter the drilling data, using lithology identification technology to obtain the formation drillability identification value and judge the formation lithology changes;

[0096] S11: Based on engineering experience, an outlier elimination method is used to screen the drilling data, wherein the drilling data includes bit weight, rotation speed, torque and drilling speed. The measurement range of the drilling data is:

[0097]

[0098] Wherein, WOB is the weight on bit, its unit is KN, RPM is the rotation speed, its unit is rpm, Torque is the torque, its unit is Nm, ROP is the drilling speed, its unit is cm / min;

[0099] S12: Use wavelet filtering to filter out the peaks and burrs in the drilling data. The wavelet transform expression is:

[0100]

[0101] Among them, W f (a, b) is the drilling characteristic information after wavelet forward transform, f(t) is the original drilling data, a is the expansion factor, b is the scale factor, t is the time, ψ() is the wavelet basis function, and the wavelet inverse transform expression is:

[0102]

[0103] Among them, g(t) is the drilling data after wavelet inverse transformation, c ψ is the wavelet factor;

[0104] S13: Mark the drilled cores and analyze them using lithology identification technology in a suitable environment to obtain formation drillability information. At the same time, determine in real time whether the formation lithology has changed based on the formation drillability information at consecutive moments.

[0105] S2: Introducing a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network algorithm, through three steps: hybrid bat algorithm optimization, restricted Boltzmann machine training, and back propagation neural network training and reverse fine-tuning, to establish an offline drilling rate prediction model for complex formation environments;

[0106] S21: Introducing the trial-and-error method and hybrid bat algorithm to optimize the hyperparameters of the drilling speed prediction model, including batch size, number of training iterations, number of hidden layer neurons, learning rate, and momentum;

[0107] S22: Extract key features from drilling data by training a restricted Boltzmann machine. The restricted Boltzmann machine includes a visual layer v and a hidden layer h. The energy function under a given state (v, h) is defined as:

[0108]

[0109] Among them, E θ (v,h) is the energy function of the restricted Boltzmann machine, θ={w ij ,c i ,d j} are the model parameters of the restricted Boltzmann machine, p and q are the number of neurons in the visual layer and the hidden layer, c i and d j is the bias of the i-th visual layer neuron and the j-th hidden layer neuron, v i is the i-th visual layer neuron, h j is the jth hidden layer neuron, w ij is the weight between the i-th visual layer neuron and the j-th hidden layer neuron, i = 1, 2…, p, j = 1, 2…, q;

[0110] Given a drilling training set For nth k The log-likelihood function L(θ) of the restricted Boltzmann machine on K is:

[0111]

[0112] Where r is the rth drilling data sample, r=1,2…,n k , n k is the total number of samples, v r is the rth visual layer, P(v r ) is the visual layer v r The marginal probability distribution of ; At the same time, the contrast divergence algorithm is used to quickly train the restricted Boltzmann machine to ensure the timeliness of the calculation;

[0113] S23: The output features of the restricted Boltzmann machine are used as the input of the back-propagation neural network. The back-propagation neural network consists of two parts: forward propagation and back-propagation. The forward propagation can be expressed as:

[0114]

[0115] in, is the drilling speed prediction value output by the forward pass, f() is the activation function, g ris the output layer bias, k represents the kth hidden layer node, k=1,2…,n, n is the total number of hidden layer nodes, w kr is the connection weight between the hidden layer and the output layer, x k is the hidden layer node;

[0116] At the same time, the network structure is continuously adjusted using back propagation. The error function of the back propagation neural network can be expressed as:

[0117]

[0118] Among them, E y is the mean square error of the back propagation neural network, n k is the total number of samples, is the predicted value of drilling speed, y r is the actual value of drilling speed, and then the gradient descent method is used to update the parameters of the back propagation neural network (w kr ,g r ).

[0119] S3: Using lithologic changes and time intervals as update conditions for the ROP prediction model, a sliding window strategy is used based on formation drillability information and drilling process information to update the ROP prediction model in real time.

[0120] Taking lithologic changes and time intervals as model update conditions, the sliding window strategy is used to update the drilling rate prediction model in real time based on the fusion of multi-source information of formation and drilling. The model update condition is expressed as:

[0121]

[0122] Among them, u() is the switch variable for updating the drilling speed prediction model, Δu r is the state variable updated by lithologic change, Δu t is the state variable updated at time interval, r s and r s+1 is the formation drillability corresponding to two consecutive moments, r z is the lithologic change update threshold, t a and t a+1 is the time point corresponding to two consecutive moments, t b is the time interval update threshold, when Δu r Greater than or equal to r z When Δu t Equal to t b When the drilling speed prediction model is updated, it will continuously learn from the actual drilling data based on time.

[0123] When any of the above conditions is met, the drilling rate prediction model will be updated and predicted in real time.

[0124] S4: Input the actual drilling pressure, rotation speed, and torque into the updated drilling speed prediction model to obtain the drilling speed at the next moment.

[0125] The method also includes using model evaluation indicators to evaluate the prediction performance of the method. The specific formula is:

[0126]

[0127]

[0128] Among them, RMSE is the root mean square error, NRMSE is the normalized root mean square error, and y r is the actual drilling speed, is the drilling speed prediction value, r is the rth drilling data sample, r=1,2…,n k , n k The smaller the root mean square error and the normalized root mean square error, the better the fitting effect between the predicted drilling speed and the actual drilling speed, indicating that the prediction ability of the drilling speed prediction model of the present invention is stronger.

[0129] This example uses industrial data from a drilling site in Dandong, Northeast China as an example. The specific process is as follows:

[0130] (1) 122,383 sets of actual drilling data (bit weight, speed, torque and drilling speed) from the scientific drilling wells in Dandong, Northeast China were selected. The data distribution is as follows: Figure 3 As shown in the histogram, the data demonstrates that the quality of the drilling data from Kezui is indeed very low. The inputs (weight on bit, rotational speed, and torque) and output (rate of penetration) of the ROP prediction model exhibit a non-Gaussian distribution and contain a large number of outliers. For example, some WOB and torque measurements exceed 500 kN and 60,000 Nm, respectively, and a few ROP measurements even exceed 6,000 cm / min.

[0131] Two pre-processing techniques are used to screen and filter the drilling data, such as Figure 2 As shown in stage 1 of the data preprocessing, in the first data preprocessing technology, the outlier elimination method is combined with manual experience to filter the drilling data. In the second data preprocessing technology, the wavelet filtering method is used to reduce the noise of the drilling data. The distribution of the filtered Dandongke drilling data is as follows: Figure 4 As shown in the figure, the peaks and burrs in the data are effectively removed. At the same time, the lithology recognition technology is used to identify the formation drillability information. The formation drillability distribution is shown in the figure. Figure 5 As shown in the figure, we can see that the formation drillability identification values are mainly concentrated between 7 and 10, indicating that the formation in the current area is relatively hard, which is not conducive to the drill bit breaking the rock quickly;

[0132] (2) The hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network is used to establish an offline drilling speed prediction model, such as Figure 2 As shown in Stage 2 of the paper, first, a restricted Boltzmann machine is trained to extract key features of the drilling data. Second, the output features of the restricted Boltzmann machine are used as input to the backpropagation neural network, which is then trained and fine-tuned. Finally, a trial-and-error method and a hybrid bat algorithm are introduced to optimize the model hyperparameters. The optimal model hyperparameters for the proposed method are shown in Table 1.

[0133] Table 1 Optimal hyperparameters of the model of the method proposed in this invention

[0134]

[0135]

[0136] (3) Taking the lithologic change and time interval as the model update conditions, the drilling rate prediction model is updated through the sliding window strategy based on the fusion of multi-source information of formation and drilling, such as Figure 2 As shown in stage 3;

[0137] (4) To test the effectiveness of the proposed method, it was compared with seven drilling rate prediction methods (two offline and five online), whose definitions are shown in Table 2. Table 3 and Figure 6 The comparison results of actual drilling rate and predicted drilling rate are shown.

[0138] Table 2 Definitions of the method proposed in this invention and seven drilling rate prediction methods

[0139]

[0140] Table 3 Comparison results between the method proposed in this invention and seven drilling speed prediction methods

[0141]

[0142]

[0143] As can be seen, the method disclosed in this paper has excellent performance in capturing drilling rate trends, with root mean square error and normalized root mean square error of 0.0862 and 3.36%, respectively. Compared with seven drilling rate prediction methods (M1, M2, M3, M4, M5, M6, and M7), the prediction accuracy is at least 13% higher. This shows that the proposed method can meet the needs of drilling projects, help shorten drilling cycles, reduce operating costs, achieve high-precision drilling rate prediction, and is effective, laying an important foundation for intelligent optimization and control of drilling processes.

[0144] The beneficial effects of the present invention are:

[0145] (1) Drilling data is screened and filtered by using outlier removal and wavelet filtering technology, and lithology identification technology is used to identify formation drillability information, providing effective data support for drilling rate modeling and model updating;

[0146] (2) An offline drilling rate prediction model is established using a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network. This model has strong learning and nonlinear fitting capabilities and is more suitable for complex formation environments.

[0147] (3) Taking the lithologic changes and time intervals as the model updating conditions, the drilling speed prediction model is updated in real time through the sliding window strategy based on the fusion of multi-source information of formation and drilling, which improves the prediction accuracy of drilling speed and is conducive to the application of the present invention in actual production.

[0148] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0149] The serial numbers of the embodiments of the present invention are for descriptive purposes only and do not represent superiority or inferiority of the embodiments. In a unit claim that lists several means, several of these means may be embodied by the same item of hardware. The use of the terms first, second, and third, etc., does not denote any order and should be construed as identifiers.

[0150] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for online prediction of drilling speed during drilling process based on multi-source information fusion, characterized in that: The following steps are involved: S1: Selecting drilling pressure, rotational speed and torque as inputs of the drilling speed prediction model, using outlier removal and wavelet filtering analysis to screen and filter the drilling data, using lithology identification technology to obtain the formation drillability identification value and judge the formation lithology changes; S2: Introducing a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network algorithm, an offline drilling rate prediction model for complex formation environments is established through three steps: hybrid bat algorithm optimization, restricted Boltzmann machine training, and back propagation neural network training and reverse fine-tuning. Specifically: S21: Introducing the trial-and-error method and hybrid bat algorithm to optimize the hyperparameters of the drilling speed prediction model, including batch size, number of training iterations, number of hidden layer neurons, learning rate, and momentum; S22: Extract key features from drilling data by training a restricted Boltzmann machine, which includes a visual layer and hidden layers , defining a given state The energy function is: in, is the energy function of the restricted Boltzmann machine, are the model parameters of the restricted Boltzmann machine, and is the number of neurons in the visual layer and hidden layer, and For the i visual layer neurons and the j The bias of the hidden layer neurons, For the i Visual layer neurons, For the j hidden layer neurons, For the i visual layer neurons and the j The weights between the neurons in the hidden layer, , ; S3: Using lithologic changes and time intervals as update conditions for the ROP prediction model, a sliding window strategy is used based on formation drillability information and drilling process information to update the ROP prediction model in real time. Taking lithologic changes and time intervals as model update conditions, the sliding window strategy is used to update the drilling rate prediction model in real time based on the fusion of multi-source information of formation and drilling. The model update condition is expressed as: in, ( ) is the switch variable for updating the drilling speed prediction model, The state variables updated for lithologic changes, is the state variable updated at time intervals, and is the formation drillability corresponding to two consecutive moments, Update thresholds for lithologic changes, and is the time point corresponding to two consecutive moments, is the time interval update threshold, when Greater than or equal to When , the drilling rate prediction model will be updated according to the formation lithology information; when equal When the drilling speed prediction model is updated, it will continuously learn from the actual drilling data based on time. When any of the above conditions is met, the drilling rate prediction model will be updated and predicted in real time; S4: Input the actual drilling pressure, rotation speed, and torque into the updated drilling speed prediction model to obtain the drilling speed at the next moment.

2. The method for online prediction of drilling speed during drilling based on multi-source information fusion according to claim 1, characterized in that: Step S1 specifically The following processes are included: S11: Based on engineering experience, an outlier elimination method is used to screen the drilling data, wherein the drilling data includes bit weight, rotation speed, torque and drilling speed. The measurement range of the drilling data is: Among them, WOB is the bit weight, and its unit is , RPM is the speed, its unit is , Torque is torque, its unit is , ROP is the drilling speed, its unit is ; S12: Use wavelet filtering to filter out the peaks and burrs in the drilling data. The wavelet transform expression is: in, is the drilling characteristic information after wavelet forward transform, is the original drilling data, is the expansion factor, is the scale factor, t For time, is the wavelet basis function, and the inverse wavelet transform expression is: in, is the drilling data after wavelet inverse transformation, is the wavelet factor; S13: Mark the drilled cores and analyze them using lithology identification technology in a suitable environment to obtain formation drillability information. At the same time, determine in real time whether the formation lithology has changed based on the formation drillability information at consecutive moments.

3. The method for online prediction of drilling speed during drilling based on multi-source information fusion according to claim 1, characterized in that: Step S2 specifically further The following processes are included: Given a drilling training set , For the The restricted Boltzmann machine is used to drill into the data. K The log-likelihood function on for: in, For the Drilling data samples, , is the total number of samples, For the A visual layer, For the visual layer The marginal probability distribution of ; The output features of the restricted Boltzmann machine are used as the input of the back-propagation neural network. The back-propagation neural network consists of two parts: forward propagation and back-propagation. The forward propagation can be expressed as: in, is the predicted drilling speed, is the activation function, is the output layer bias, Indicates the hidden layer nodes, , is the total number of hidden layer nodes, is the connection weight between the hidden layer and the output layer, is the hidden layer node; At the same time, the network structure is continuously adjusted using back propagation. The error function of the back propagation neural network can be expressed as: in, is the mean square error of the back-propagation neural network, is the total number of samples, is the predicted drilling speed, is the actual value of drilling speed, and then the gradient descent method is used to update the parameters of the back propagation neural network .

4. The method for online prediction of drilling speed during drilling based on multi-source information fusion according to claim 1, characterized in that: The method also includes using model evaluation indicators to evaluate the prediction performance of the method. The specific formula is: in, is the root mean square error, is the normalized root mean square error, is the actual drilling speed, is the predicted drilling rate, For the Drilling data samples, , is the total number of samples.

5. A prediction system for an online prediction method of drilling speed during drilling based on multi-source information fusion, characterized in that: The system applies the prediction method according to claim 1, comprising: The data preprocessing module is used to select drilling pressure, rotational speed and torque as inputs to the drilling speed prediction model, screen and filter the drilling data using outlier removal and wavelet filtering analysis, and use lithology identification technology to obtain the formation drillability identification value and judge the changes in the formation lithology; A model module was established to introduce a hybrid bat algorithm optimization-restricted Boltzmann machine-back propagation neural network algorithm. Through three steps of hybrid bat algorithm optimization, restricted Boltzmann machine training, and back propagation neural network training and reverse fine-tuning, an offline drilling rate prediction model for complex formation environments was established. The update model module is used to update the ROP prediction model in real time using lithologic changes and time intervals as the update conditions, and uses a sliding window strategy based on formation drillability information and drilling process information. The drilling speed prediction module is used to input the actual drilling pressure, rotation speed and torque into the updated drilling speed prediction model to obtain the drilling speed at the next moment.

6. The online prediction system for drilling speed during drilling based on multi-source information fusion according to claim 5, characterized in that: In the data preprocessing module, the process of preprocessing drilling data is as follows: S11: Based on engineering experience, an outlier elimination method is used to screen the drilling data, wherein the drilling data includes bit weight, rotation speed, torque and drilling speed. The measurement range of the drilling data is: Among them, WOB is the bit weight, and its unit is , RPM is the speed, its unit is , Torque is torque, its unit is , ROP is the drilling speed, its unit is ; S12: Use wavelet filtering to filter out the peaks and burrs in the drilling data. The wavelet transform expression is: in, is the drilling characteristic information after wavelet forward transform, is the original drilling data, is the expansion factor, is the scale factor, t For time, is the wavelet basis function, and the inverse wavelet transform expression is: in, is the drilling data after wavelet inverse transformation, is the wavelet factor; S13: Mark the drilled cores and analyze them using lithology identification technology in a suitable environment to obtain formation drillability information. At the same time, determine in real time whether the formation lithology has changed based on the formation drillability information at consecutive moments.

7. The online prediction system for drilling speed during drilling based on multi-source information fusion according to claim 5, characterized in that: In the model building module, the model building process includes: Given a drilling training set , For the The restricted Boltzmann machine is used to drill into the data. K The log-likelihood function on for: in, For the Drilling data samples, , is the total number of samples, For the A visual layer, For the visual layer The marginal probability distribution of ; The output features of the restricted Boltzmann machine are used as the input of the back-propagation neural network. The back-propagation neural network consists of two parts: forward propagation and back-propagation. The forward propagation can be expressed as: in, is the predicted drilling speed, is the activation function, is the output layer bias, Indicates the hidden layer nodes, , is the total number of hidden layer nodes, is the connection weight between the hidden layer and the output layer, is the hidden layer node; At the same time, the network structure is continuously adjusted using back propagation. The error function of the back propagation neural network can be expressed as: in, is the mean square error of the back-propagation neural network, is the total number of samples, is the predicted drilling speed, is the actual value of drilling speed, and then the gradient descent method is used to update the parameters of the back propagation neural network .

8. The online prediction system for drilling speed during drilling based on multi-source information fusion according to claim 5, characterized in that: The system also includes the use of model evaluation indicators to evaluate the prediction performance of the system. The specific formula is: in, is the root mean square error, is the normalized root mean square error, is the actual drilling speed, is the predicted drilling rate, For the Drilling data samples, , Indicates the total number of samples.

Images