Drillability prediction method based on particle swarm optimization neural network and microstructure

Abstract

The present invention discloses a method for predicting drillability based on a particle swarm optimization neural network and a mesostructure, comprising: obtaining a rock sample thin section image of a rock sample to be tested, calculating the mesostructure parameters of the rock sample thin section image; constructing a rock drillability prediction model based on a particle swarm optimization neural network, and inputting the mesostructure parameters of the rock sample thin section image into the rock drillability prediction model to obtain a predicted value of rock drillability. The present invention can better mine the potential laws and features in the data and improve the accuracy and efficiency of prediction.

Description

Technical Field

[0001] The invention relates to the technical field of petroleum exploration and development, and in particular to a drillability prediction method based on particle swarm optimization neural network and microstructure. Background Art

[0002] The study of rock drillability is crucial for oil and gas exploration and development. A deep understanding of the properties of underground rocks can provide valuable geological information and help accurately evaluate the characteristics of reservoirs and the distribution of oil and gas resources. Such understanding can help optimize exploration positioning, reduce exploration risks, and improve the efficiency and feasibility of resource development. Rock drillability prediction is of great significance in drilling engineering. Accurate prediction of rock drillability can help optimize drilling design and operations and improve drilling efficiency and success rate. By predicting the physical properties of underground rock formations, engineers can better understand the geological conditions that may be encountered during drilling and adopt corresponding drilling strategies. This helps reduce the risk of drilling accidents and improve the safety of drilling. Overall, rock drillability prediction plays a key role in drilling engineering and geological resource development, providing basic geological information for engineering decision-making, thereby promoting scientific and technological progress and sustainable development in the energy industry.

[0003] In the invention of patent number CN115078362B, a method for characterizing drillability based on rock mesostructure is proposed. This method evaluates rock drillability through a single indicator, TC, without comprehensively considering the factors affecting rock drillability, such as circumference, porosity, etc., and does not use a particle swarm optimization neural network to improve the accuracy of the rock drillability prediction model.

[0004] Patent No. CN102606151B proposes a method and device for predicting the drillability of wildcat well rocks before drilling. This method uses the seismic velocity spectrum of the formation to be drilled to pick up the layer velocities of different layers of different rocks in the wildcat well as mutual analysis, and uses BP neural network for training to obtain the relationship between the layer velocity and the drillability of the rock. However, it does not consider the influence of the structure of the rock itself on the drillability.

[0005] In the article "Study on Rock Drillability Logging Calculation Based on BP Neural Network", a practical method based on BP neural network to obtain rock drillability using logging data is proposed. This method uses logging data as input parameters and drillability as output parameters, and uses BP neural network for training to obtain a drillability prediction model based on logging data. However, no prediction model for rock drillability is established based on the structural properties of the rock itself.

[0006] In summary, there is an urgent need for a rock drillability prediction method based on particle swarm optimization neural network, i.e., mesostructure, to comprehensively consider the factors affecting rock drillability and significantly improve the accuracy of the prediction model. Summary of the invention

[0007] The purpose of the present invention is to provide a method for predicting drillability based on a particle swarm optimization neural network and a microstructure, so as to solve the problems existing in the above-mentioned prior art.

[0008] The present invention provides a method for predicting drillability based on a particle swarm optimization neural network and a microstructure, comprising:

[0009] Obtaining a rock sample thin section image of the rock sample to be tested, and calculating the mesoscopic structural parameters of the rock sample thin section image;

[0010] A rock drillability prediction model is constructed based on a particle swarm optimization neural network, and the mesoscopic structural parameters of the rock sample thin section image are input into the rock drillability prediction model to obtain a predicted value of the rock drillability.

[0011] Optionally, the process of obtaining microstructure parameters includes:

[0012] forming the rock sample into rock sample slices;

[0013] Using a microscope to obtain single polarization and orthogonal polarization images of the rock sample slice;

[0014] Calculate the mineral particle contours of the single polarized light and orthogonal polarized light images based on ImageJ image processing software;

[0015] Obtaining geometric structure data based on the mineral particle contours;

[0016] The microstructure parameters are obtained based on the geometric structure data.

[0017] Optionally, the mesostructure parameters include: diameter, area, circumference, minimum aspect ratio, roundness, and mesostructure coefficient.

[0018] Optionally, the rock drillability prediction model constructed based on the particle swarm optimization neural network includes obtaining optimal network weights and thresholds based on the particle swarm algorithm; wherein the process of obtaining the optimal network weights and thresholds based on the particle swarm algorithm includes:

[0019] S1, initialize the initial velocity and displacement of the particle;

[0020] S2, update the speed and position of particles based on particle swarm algorithm;

[0021] S3, update the individual optimal value and the global extreme value according to the position of the particle, and calculate the global optimal fitness;

[0022] S4, looping through S1-S3 until the global optimal fitness converges to obtain the optimal network weights and thresholds.

[0023] Optionally, the calculation formula of the global optimal fitness is:

[0024]

[0025] Where n is the number of sampling points, y(t) is the expected output at the tth point, is the actual output at point t.

[0026] Optionally, after the rock drillability prediction model is constructed, a model learning and training process is also included, wherein the process of learning and training the rock drillability prediction model includes:

[0027] Initialize the allowable error and translation factor;

[0028] Conduct drillability test on test rock samples to obtain target output parameters;

[0029] Calculating the mesoscopic structural parameters of the thin section image of the test rock sample to obtain input parameters;

[0030] Inputting the input parameters into the rock drillability prediction model to obtain an actual output value;

[0031] Obtaining an error based on the actual output value and the target output parameter;

[0032] When the error is smaller than the allowable error, a trained rock drillability prediction model is obtained, otherwise relearning and training is performed.

[0033] Optionally, the error is calculated as follows:

[0034]

[0035] In the formula, E k Indicates error.

[0036] The present invention has the following technical effects:

[0037] The present invention provides a rock drillability prediction method based on a particle swarm optimization neural network and a mesostructure. The method has a reliable principle and is easy to operate. It can quickly and accurately predict rock drillability based on downhole cuttings, improve drilling efficiency, and thus greatly improve the development efficiency of oil and gas resources. The method has broad market application prospects.

[0038] The present invention can better mine the potential laws and features in the data and improve the accuracy and efficiency of prediction; the present invention can quickly converge to the optimal solution and reduce the time required for model training; the present invention can provide a scientific basis for engineers to design drilling plans and process flows more finely, thereby reducing drilling risks and costs; the present invention can improve the efficiency and feasibility of resource exploration and development. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0040] Figure 1 A flow chart of a rock drillability prediction method based on a particle swarm optimization neural network and mesostructure in an embodiment of the present invention;

[0041] Figure 2 is a displacement-time curve obtained from a rock drillability experiment in an embodiment of the present invention;

[0042] Figure 3 301 is a comparison result before and after image processing in an embodiment of the present invention; wherein 301 represents a single polarized image of a rock, 302 represents a result of contour division of mineral particles, and 303 represents a result of image processing;

[0043] Figure 4 The figure is a comparison result between the predicted value and the actual value of the rock drillability prediction model in the embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] Embodiment 1

[0046] like Figure 1 As shown, this embodiment discloses a rock drillability prediction method based on particle swarm optimization neural network and microstructure, and the specific implementation process includes: step 1, conducting drillability experiments to obtain target output parameters; step 2, calculating microstructure parameters as input parameters; step 3, setting allowable error, translation factor, etc. to initialize the network; step 4, setting the initial velocity and displacement of the particles, continuously updating the velocity and position of the particles based on the particle swarm algorithm, and then updating the individual optimal value and global extreme value according to the position of the particles, and calculating the global optimal fitness. Determine whether it converges, and finally obtain the optimal network weight and threshold; step 5, perform network learning and training; step 6, determine the relationship between the error and the allowable error, and stop the iteration when the error is less than the allowable error, otherwise re-perform network learning and training; step 7, establish a rock drillability prediction model.

[0047] As a preferred solution of this embodiment, the specific implementation process of each step includes:

[0048] Step 1: Conduct drillability experiment and obtain target output parameters. Select rock samples to conduct rock drillability experiment, calculate and organize the experimental data into a unified Excel table as the target output parameters in the original data. As a preferred solution of this embodiment, select rock samples to conduct rock drillability experiment. The rock sample size is 50mm in diameter × 100mm in height. The rock drillability experiment obtains the displacement-time curve of the micro drill bit during the drilling process, such as Figure 2 As shown in the figure, according to the "Rock Drillability Testing and Classification for Petroleum and Natural Gas Drilling Engineering" (SY / T 5426-2016), the effective drilling time is calculated and organized into a unified Excel table as the target output parameter in the original data. The formula for calculating rock drillability through effective drilling time is as follows: K d =log2(t)

[0049] Where K d represents the drillability of rock, and t represents the effective drilling time.

[0050] Step 2: Calculate the mesostructure parameters as input parameters. Make thin slices of the rock samples used in the rock drillability experiment. The thickness of the slice is 0.03 mm. Observe and take single polarized and orthogonal polarized images of each rock through a microscope, calibrate the scale in ImageJ, and then surround each mineral particle with a number of data points. Calculate the position of each data point, connect two adjacent data points with a straight line to obtain the outline of the surrounded mineral particle, calculate the geometric structure of the mineral particle according to the outline, export the geometric structure data, use Python, ImageJ and Matlab software to process the image and calculate the mesostructure parameters, and use the calculated mesostructure parameters as the input parameters in the original data. As a preferred solution of this embodiment, use Matlab software to program the mesostructure parameters to be calculated according to the original definition, use Matlab software to read the geometric structure data and calculate the mesostructure parameters. The mesostructure parameters include diameter, area, circumference, minimum aspect ratio, roundness, mesostructure coefficient, etc. The calculated mesostructure parameters are used as the input parameters in the original data, such as Figure 3 Shown is the comparison result before and after image processing in an embodiment of the present invention.

[0051] The microstructure coefficient TC is calculated using the following formula:

[0052]

[0053] Where TC is the mesostructure coefficient; AW is the mineral filling coefficient, N0 and N1 are the number of particles with aspect ratio (Feret's definition) lower and higher than the preset value (generally 2), SF0 is the arithmetic mean of the discriminated shape factors of all particles, AR1 is the arithmetic mean of the discriminated aspect ratios of all particles, and AF1 is the angle factor of the particles.

[0054] Step 3: Initialize the network by setting the allowable error, translation factor, etc. As a preferred solution of this embodiment, the allowable error is set to 0.01 and the translation factor is set to 0.5.

[0055] Step 4: Get the optimal network weights and thresholds based on the particle swarm algorithm. Set the initial velocity and displacement of the particles, continuously update the velocity and position of the particles based on the particle swarm algorithm, then update the individual optimal value and global extreme value according to the position of the particles, and calculate the global optimal fitness. Determine whether it converges, and finally get the optimal network weights and thresholds. The fitness calculation formula is as follows:

[0056]

[0057] Where n is the number of sampling points, y(t) is the expected output at the tth point, is the actual output at point t.

[0058] Step 5: Perform network learning training, using the microstructure parameters obtained in step 2 as input parameters and the drillability obtained in step 1 as output parameters, and perform network learning training through continuous iteration. As a preferred solution of this embodiment, a data set is created with the microstructure parameters as input parameters and the drillability as output parameters. The data set is defined as (x i ,y i ), the output of the training set is in, The input received by the lth neuron in the output layer is β l ,θ l is the threshold of the lth neuron in the output layer, wh l is the connection weight between the hth neuron in the hidden layer and the lth neuron in the output layer, b h is the output of the hth neuron in the hidden layer. The training set accounts for 80% of the dataset, the test set accounts for 20%, and the random seed is 10. The gradient of the loss function with respect to the weight and bias is calculated by back propagation. The gradient term of the output layer neuron is The gradient term of the hidden layer neuron is Then the weights and biases are updated in the reverse direction of the network, and the value of the loss function is reduced by continuously iterating network learning training.

[0059] Step 6: Error judgment. Calculate the error between the actual output of the network and the expected output. The error is The relationship between the judgment error and the allowable error is determined. When the error is less than the allowable error, the iteration is stopped. Otherwise, the network learning training is repeated until the error is less than the allowable error.

[0060] Step 7: Establish a rock drillability prediction model. By inputting the mesoscopic structural parameters into the established rock drillability prediction model, the predicted value of rock drillability can be calculated.

[0061] The present invention is used to analyze the characteristics of the microscopic structural parameters of the rock and establish a rock drillability prediction model. Figure 4 It can be found that the maximum error of rock drillability prediction is 18.57%, which meets the engineering needs and has a good prospect for promotion and application.

[0062] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for predicting drillability based on particle swarm optimization neural network and microstructure, characterized in that: The following steps are involved: Obtaining a rock sample thin section image of the rock sample to be tested, and calculating the mesoscopic structural parameters of the rock sample thin section image; A rock drillability prediction model is constructed based on a particle swarm optimization neural network, and the mesoscopic structural parameters of the rock sample thin section image are input into the rock drillability prediction model to obtain a predicted value of the rock drillability; The process of obtaining microstructure parameters includes: forming the rock sample into rock sample slices; Using a microscope to obtain single polarization and orthogonal polarization images of the rock sample slice; Calculate the mineral particle contours of the single polarized light and orthogonal polarized light images based on ImageJ image processing software; Obtaining geometric structure data based on the mineral particle contours; Obtaining microstructure parameters based on the geometric structure data; The microstructure parameters include: diameter, area, perimeter, minimum aspect ratio, roundness, and microstructure coefficient; The rock drillability prediction model based on the particle swarm optimization neural network is constructed, which includes obtaining the optimal network weights and thresholds based on the particle swarm algorithm; wherein the process of obtaining the optimal network weights and thresholds based on the particle swarm algorithm includes: S1, initialize the initial velocity and displacement of the particle; S2, update the speed and position of particles based on particle swarm algorithm; S3, update the individual optimal value and the global extreme value according to the position of the particle, and calculate the global optimal fitness; S4, looping through S1-S3 until the global optimal fitness converges and the optimal network weights and thresholds are obtained; The calculation expression of the microstructure coefficient is: Wherein, TC is the mesostructure coefficient; AW is the mineral filling coefficient, N0 and N1 are the number of particles with aspect ratios lower than and higher than the preset values, SF0 is the arithmetic mean of the discernible shape factors of all particles, AR1 is the arithmetic mean of the discernible aspect ratios of all particles, and AF1 is the angle factor of the particles; The calculation formula of the global optimal fitness is: Where n is the number of sampling points, y(t) is the expected output at the tth point, is the actual output at point t; After the rock drillability prediction model is constructed, a model learning and training process is also included, wherein the process of learning and training the rock drillability prediction model includes: Initialize the allowable error and translation factor; Conduct drillability test on test rock samples to obtain target output parameters; Calculating the mesoscopic structural parameters of the thin section image of the test rock sample to obtain input parameters; Inputting the input parameters into the rock drillability prediction model to obtain an actual output value; Obtaining an error based on the actual output value and the target output parameter; When the error is less than the allowable error, a trained rock drillability prediction model is obtained, otherwise relearning and training is performed; The calculation formula of the error is: In the formula, E k Indicates error.

Images