A real-time data compression and transmission method for a while-drilling acoustic wave forward-looking long-range detector

Abstract

The invention discloses a real-time data compression and transmission method for a while-drilling acoustic wave forward-looking long-range detector. The method comprises the following steps: firstly, an encoder and a decoder of a QCS-Net network are deployed; during the drilling process, down-sampling compression and nonlinear mapping are performed on acoustic wave forward-looking long-range detection data through a convolution layer and an activation function of the encoder to obtain compressed floating-point acoustic wave forward-looking long-range detection data; then, the floating-point acoustic wave forward-looking long-range detection data is converted into a binary bit stream through a quantization algorithm; then, the binary bit stream is converted into floating-point data, and the original signal size dimension is restored through a transposed convolution in a decoder, and the original signal is further reconstructed and the quality is improved through a convolution layer with a residual structure, so as to obtain the final high-fidelity acoustic wave forward-looking long-range detection data.

Description

Technical Field

[0001] The invention belongs to the technical field of acoustic logging while drilling, and more specifically, relates to a real-time data compression and transmission method for an acoustic forward-looking long-range detector while drilling. Background Art

[0002] In the exploration of oil and gas resources, the forward-looking long-range detection technology while drilling is praised as one of the "20 new technologies with great development potential" in the oil and gas industry. It is a high-tech logging technology that uses the acoustic wave reflection mechanism to timely obtain the physical characteristics of the well wall formation in front of the drill bit while the drill bit is drilling. The acoustic wave forward-looking long-range detection technology while drilling uses ultrasonic waves to obtain logging data of the well wall formation information within the fan-shaped range in front of the drill bit during drilling, and transmits the logging results to the ground by means of downhole storage circuits to store the logging information or mud pulses, etc., which can timely feedback underground information. Analysis of these logging data can obtain formation information such as well wall cracks and holes, which plays a positive role in optimizing drilling decisions, improving reservoir drilling rate, and reducing operating costs.

[0003] The logging images or other forms of signals obtained based on the acoustic forward-looking long-range detection data while drilling are usually stored in the edge device memory of the drilling end due to the influence of communication bandwidth. The forward-looking long-range detection data can only be sent to the ground end after a certain operation stage or all operations are completed. In other words, it is difficult to obtain acoustic forward-looking long-range detection data in real time, which goes against the original intention of real-time analysis of geological structures in logging while drilling. In order to improve the processing speed of logging while drilling data, compression algorithms such as DWT and DCT are proposed to compress logging data, but the compression of these methods is relatively low and still cannot meet the requirements of real-time compressed transmission of logging data.

[0004] In recent years, with the development of deep learning technology, deep learning technology has also been used in the compression of images or other data, and has shown great advantages. However, up to now, the oil and gas drilling field is limited by many factors such as the actual production environment and hardware platform limitations, and there are few compressed sensing and transmission methods used for logging data compression and transmission. At the same time, there is no data quantization method that can match hardware transmission. In addition, the existing compressed sensing methods still have problems such as poor hardware support, inability to match hardware transmission protocols, low compression ratio, and poor reconstruction quality. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a real-time data compression and transmission method for a downhole acoustic wave forward-looking long-range detector, which effectively reduces the data transmission cost through data quantization-dequantization at the downhole end, and improves the reconstruction quality of high compression ratio data by reconstructing the compressed data at the ground server end.

[0006] To achieve the above-mentioned purpose, the present invention provides a method for real-time data compression and transmission of a while-drilling acoustic wave forward-looking long-range detector, characterized in that it comprises the following steps:

[0007] (1) Deploy the encoder in the fully trained quantized compressed sensing network QCS-Net on the drilling end, and deploy the decoder on the ground server end; when powered on, initialize and load the network parameters of QCS-Net first;

[0008] Set the number of binary data bits N when compressing data;

[0009] (2) Obtain observation values ​​of acoustic wave forward-looking far-sighted detection data;

[0010] (2.1) During the drilling detection process, the acoustic wave data or imaging data received by the acoustic wave forward-looking detector is recorded as the acoustic wave forward-looking long-distance detection data X;

[0011] (2.2) According to the preset time cutoff window, the acoustic wave forward far-sighted detection data X is sliced ​​to obtain the acoustic wave forward far-sighted detection data sequence of a certain data dimension, which is recorded as X = {x1, x2, …, x i ,…,x n}, where x i represents the i-th piece of acoustic wave forward-looking far detection data, and n represents the number of slices in X;

[0012] (2.3) traverse each piece of acoustic wave forward-looking detection data x in X in turn i , and input to the encoder, and the compressed data y is obtained through two consecutive convolution downsampling operations in the encoder. i ;

[0013] y i =Conv 4×4 (Conv 8×8 (x i ))

[0014] Among them, Conv 4×4 、Conv 8×8 Respectively represent convolution operations with convolution kernel sizes of 4×4 and 8×8;

[0015] (2.4) Through the activation function HardSigmoid, the compressed data y i Perform nonlinear mapping to obtain the observation value of the compressed and complete acoustic wave forward-looking far-reaching detection data

[0016]

[0017] Among them, p i,j Represents x i The value of the jth element in ;

[0018] (3) Convert the observation value of the acoustic wave forward-looking far-reaching detection data into a bit stream;

[0019] (3.1), the observed value Flatten into a one-dimensional vector;

[0020]

[0021] Among them, Flatten() represents the flattening operation;

[0022] (3.2), flattened one-dimensional vector Carry on 2 N The enlarged and integer parts are rounded and truncated to obtain the integer part of the enlarged compressed data

[0023]

[0024] Among them, Round() represents the rounding operation;

[0025] (3.3) Compress the integer part of the data Convert to binary data, and then intercept the lower N bits of the binary data as the binary acoustic wave remote detection bit stream y i,b ;

[0026] (4) Use the binary sound wave to remotely detect the bit stream y i,b Transmit to the ground server, and then convert the bit stream y i,b Convert to floating point data;

[0027] (4.1) On the ground server, the binary sound wave remote detection bit stream y i,b Convert to decimal data to get floating point data y i,f ;

[0028] (4.2), for floating point data y i,f Further through scaling and resizing operations, the dequantized compressed data y is obtained i,dq ;

[0029] y i,dq =Reshape(y i,f / 2 N )

[0030] Among them, Reshape() represents the scaling and resizing operation;

[0031] (5) Use the decoder in the well-trained quantized compressed sensing network QCS-Net to decode and reconstruct the acoustic wave forward-looking long-range detection data in real time;

[0032] (5.1), dequantize the compressed data y i,dq As the input of the decoder, the decoded data y is obtained through two consecutive deconvolution upsampling operations in the decoder. i,d ;

[0033] y i,d =ReLU{DeConv 8×8 [DeConv 4×4 (y i,dq )]}

[0034] Among them, DeConv 4×4 、DeConv 8×8 They represent deconvolution operations with kernel sizes of 4×4 and 8×8 respectively;

[0035] (5.2), for the decoded data y i,d Nonlinear reconstruction to obtain acoustic wave forward-looking far detection data y i,R ;

[0036]

[0037] Among them, ResBlock() represents the convolution-activation function, and ReLU() represents the rectified linear unit activation function;

[0038] (6) Determine whether the traversal of each piece of data in the acoustic wave forward far-sighted detection data X is complete. If the traversal is complete, the iteration ends and the acoustic wave forward far-sighted detection data Y={y 1,R y 2,R ,…,y i,R ,…,y n,R}; Otherwise, let i=i+1 and return to step (2.3).

[0039] The object of the invention of the present invention is achieved in this way:

[0040] The invention discloses a real-time data compression and transmission method for a while-drilling acoustic wave forward-looking long-range detector. The method comprises the following steps: firstly, an encoder and a decoder of a QCS-Net network are deployed; during the drilling operation, down-sampling compression and nonlinear mapping are performed on acoustic wave forward-looking long-range detection data through a convolution layer and an activation function of the encoder to obtain compressed floating-point acoustic wave forward-looking long-range detection data; then, the floating-point acoustic wave forward-looking long-range detection data is converted into a binary bit stream through a quantization algorithm; then, the binary bit stream is converted into floating-point data, and the original signal size dimension is restored through a transposed convolution in a decoder, and the original signal is further reconstructed and the quality is improved through a convolution layer with a residual structure, so as to obtain final high-fidelity acoustic wave forward-looking long-range detection data.

[0041] At the same time, the real-time data compression and transmission method of the drilling acoustic wave forward-looking long-range detector of the present invention also has the following beneficial effects:

[0042] (1) Deploy a lightweight quantized compressed sensing encoder and a data quantization algorithm in the acoustic forward-looking long-range detector while drilling, which can realize intelligent real-time compression of acoustic forward-looking long-range detection data, thereby compressing the acoustic forward-looking long-range detection data with an extremely high compression ratio, greatly improving the data compression multiple, and also greatly reducing the acoustic forward-looking long-range detection data transmission cost;

[0043] (2) A customized data quantization-dequantization algorithm converts floating-point data into a 0-1 bit stream, thereby better supporting hardware transmission protocols and truncating invalid data bits, thereby reducing the transmission bandwidth occupancy of acoustic wave long-range detection data, and being able to transmit floating-point acoustic wave forward-looking long-range detection data in the form of pulses; the dequantization algorithm converts the 0-1 bit stream into floating-point data, thereby realizing real-time decoding of compressed data;

[0044] (3) The decoder part of the quantized compressed sensing network can restore high-fidelity downhole acoustic wave forward-looking long-range detection data in near real-time conditions through real-time nonlinear reconstruction and recovery of the decoded data. This not only improves the data recovery quality and effectively reduces the data reconstruction error, but also effectively improves the logging efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a flow chart of a real-time data compression and transmission method of a while-drilling acoustic wave forward-looking long-range detector of the present invention;

[0046] Figure 2 It is a specific implementation scheme of the acoustic wave forward-looking long-range detector while drilling;

[0047] Figure 3 It is a schematic diagram of the structure of the quantized compressed sensing network;

[0048] Figure 4 It is a schematic diagram of the customized quantization and dequantization algorithm. DETAILED DESCRIPTION

[0049] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0050] Example

[0051] Figure 1 The present invention discloses a flow chart of a real-time data compression and transmission method of a while-drilling acoustic wave forward-looking long-range detector.

[0052] In this embodiment, if Figure 1As shown, the present invention provides a real-time data compression and transmission method for a while drilling acoustic wave forward-looking long-range detector, comprising the following steps:

[0053] S1, such as Figure 2 As shown in the figure, the encoder in the fully trained quantized compressed sensing network QCS-Net is deployed on the drilling end, and the decoder is deployed on the ground server end; when powered on, the network parameters of QCS-Net are initialized and loaded first; the structure of QCS-Net is as follows Figure 3 As shown;

[0054] Set the number of binary data bits N when compressing data;

[0055] S2, obtaining the observation value of the acoustic wave forward far detection data;

[0056] S2.1. During the drilling detection process, the acoustic wave data or imaging data received by the acoustic wave forward-looking detector is recorded as the acoustic wave forward-looking long-range detection data X;

[0057] S2.2, according to the preset time cut-off window, the acoustic wave forward far-sighted detection data X is sliced ​​to obtain the acoustic wave forward far-sighted detection data sequence of a certain data dimension, which is recorded as X = {x1, x2, ..., x i ,…,x n}, where x i represents the i-th piece of acoustic wave forward-looking far detection data, and n represents the number of slices in X;

[0058] S2.3, traverse each piece of acoustic wave forward-looking detection data x in X in turn i , and input it to the encoder, and through two consecutive convolution downsampling operations in the encoder, the effective information features in the acoustic wave forward far detection data are extracted to obtain the downsampled compressed data y i ;

[0059] y i =Conv 4×4 (Conv 8×8 (x i ))

[0060] Among them, Conv 4×4 、Conv 8×8 Respectively represent convolution operations with convolution kernel sizes of 4×4 and 8×8;

[0061] S2.4, through the activation function HardSigmoid to compress the data y i Perform nonlinear mapping to obtain the observation value of the compressed and complete acoustic wave forward-looking far-reaching detection data

[0062]

[0063] Among them, p i,j Represents x i The value of the jth element in ;

[0064] S3, converting the observation value of the acoustic wave forward far detection data into a bit stream;

[0065] S3.1. Observation value Flatten into a one-dimensional vector;

[0066]

[0067] Among them, Flatten() represents the flattening operation;

[0068] S3.2. Flattened one-dimensional vector Carry on 2 N The enlarged and integer parts are rounded and truncated to obtain the integer part of the enlarged compressed data

[0069]

[0070] Among them, Round() represents the rounding operation;

[0071] S3.3. Compress the integer part of the data Convert to binary data, and then intercept the lower N bits of the binary data as the binary acoustic wave remote detection bit stream y i,b , so that invalid 0 data can be truncated, reducing data transmission bandwidth occupancy and matching hardware transmission protocols;

[0072] In this implementation, the data compression effect is at the sampling rate S R (default is 0.01) and the number of quantization bits N, the current compression multiple R can be calculated: Taking image data as an example, the input image has 96 pixels in length and width, has one grayscale channel, and the data type is floating point. The total amount of data to be transmitted is 96×96×1×32=294912. After the image is compressed 100 times by the encoder (sampling rate S R =0.01), the number of quantization bits is set to N=8, then the data that needs to be transmitted is only 3×3×10×8=720 data points, and the compression factor R=294912 / 720≈400. It can be seen that the compression factor has exceeded 400 times, which greatly reduces the data transmission cost.

[0073] S4, the binary sound wave remote detection bit stream y i,b Transmitted to the ground server, and then the bit stream y is converted into i,b Convert to floating point data;

[0074] S4.1. On the ground server, the binary sound wave remote detection bit stream y i,b As the input of the dequantization algorithm, the binary bit stream is converted into decimal data to obtain floating point data y i,f ;

[0075] S4.2. For floating point data y i,f Further through scaling and resizing operations, the dequantized compressed data y is obtained i,dq ;

[0076] y i,dq =Reshape(y i,f / 2 N )

[0077] Among them, Reshape() represents the scaling and resizing operation;

[0078] S5, using the decoder in the well-trained quantized compressed sensing network QCS-Net to decode and reconstruct the acoustic wave forward-looking far-sighted detection data in real time;

[0079] S5.1. Dequantize the compressed data y i,dq As the input of the decoder, the decoded data y is obtained through two consecutive deconvolution upsampling operations in the decoder. i,d ;

[0080] y i,d =ReLU{DeConv 8×8 [DeConv 4×4 (y i,dq )]}

[0081] Among them, DeConv 4×4 、DeConv 8×8 They represent deconvolution operations with kernel sizes of 4×4 and 8×8 respectively;

[0082] S5.2, decoded data y i,d Nonlinear reconstruction;

[0083] In this embodiment, the further decoded acoustic wave remote detection data y i,d When performing nonlinear reconstruction, a convolution module consisting of two convolution-ReLU layers with a residual structure is used to improve the quality of the acoustic wave detection data, so as to finally obtain high-fidelity acoustic wave forward-looking long-range detection data y i,R ;

[0084]

[0085] Among them, ResBlock() represents the convolution-activation function, and ReLU() represents the rectified linear unit activation function;

[0086] S6, determine whether the traversal of each piece of data in the acoustic wave forward far detection data X is completed, if the traversal is completed, the iteration ends, and the acoustic wave forward far detection data Y={y 1,R y 2,R ,…,y i,R ,…,y n,R}; Otherwise, let i=i+1 and return to step S2.3.

[0087] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A method for real-time data compression and transmission of a while-drilling acoustic wave forward-looking long-range detector, characterized in that: The steps include: (1) Deploy the encoder in the fully trained quantized compressed sensing network QCS-Net on the drilling end, and deploy the decoder on the ground server end; when powered on, initialize and load the network parameters of QCS-Net first; Set the number of binary data bits N when compressing data; (2) Obtain observation values ​​of acoustic wave forward-looking far-sighted detection data; (2.1) During the drilling detection process, the acoustic wave data or imaging data received by the acoustic wave forward-looking detector is recorded as the acoustic wave forward-looking long-distance detection data X; (2.2) According to the preset time cutoff window, the acoustic wave forward far-sighted detection data X is sliced ​​to obtain the acoustic wave forward far-sighted detection data sequence of a certain data dimension, which is recorded as X = {x1, x2, …, x i ,…,x n }, where x i represents the i-th piece of acoustic wave forward-looking far detection data, and n represents the number of slices in X; (2.3) traverse each piece of acoustic wave forward-looking detection data x in X in turn i , and input to the encoder, and the compressed data y is obtained through two consecutive convolution downsampling operations in the encoder. i ; y i =Conv 4×4 (Conv 8×8 (x i )) Among them, Conv 4×4 、Conv 8×8 Respectively represent convolution operations with convolution kernel sizes of 4×4 and 8×8; (2.4) Through the activation function HardSigmoid, the compressed data y i Perform nonlinear mapping to obtain the observation value of the compressed and complete acoustic wave forward-looking far-reaching detection data Among them, p i,j Represents x i The value of the jth element in ; (3) Convert the observation value of the acoustic wave forward-looking far-reaching detection data into a bit stream; (3.1), the observed value Flatten into a one-dimensional vector; Among them, Flatten() represents the flattening operation; (3.2), flattened one-dimensional vector Carry on 2 N The enlarged and integer parts are rounded and truncated to obtain the integer part of the enlarged compressed data Among them, Round() represents the rounding operation; (3.3) Compress the integer part of the data Convert to binary data, and then intercept the lower N bits of the binary data as the binary acoustic wave remote detection bit stream y i,b ; (4) Use the binary sound wave to remotely detect the bit stream y i,b Transmit to the ground server, and then convert the bit stream y i,b Convert to floating point data; (4.1) On the ground server, the binary sound wave remote detection bit stream y i,b Convert to decimal data to get floating point data y i,f ; (4.2), for floating point data y i,f Further through scaling and resizing operations, the dequantized compressed data y is obtained i,dq ; and i,dq =Reshape(and i,f / 2 N ) Among them, Reshape() represents the scaling and resizing operation; (5) Use the decoder in the well-trained quantized compressed sensing network QCS-Net to decode and reconstruct the acoustic wave forward-looking long-range detection data in real time; (5.1), dequantize the compressed data y i,dq As the input of the decoder, the decoded data y is obtained through two consecutive deconvolution upsampling operations in the decoder. i,d ; and i,d =ReLU{DeConv 8×8 [DeConv 4×4 (and i,dq )]} Among them, DeConv 4×4 、DeConv 8×8 They represent deconvolution operations with kernel sizes of 4×4 and 8×8 respectively; (5.2), for the decoded data y i,d Nonlinear reconstruction to obtain acoustic wave forward-looking far detection data y i,R ; Among them, ResBlock() represents the convolution-activation function, and ReLU() represents the rectified linear unit activation function; (6) Determine whether the traversal of each piece of data in the acoustic wave forward far-sighted detection data X is complete. If the traversal is complete, the iteration ends and the acoustic wave forward far-sighted detection data Y={y 1,R y 2,R ,…,y i,R ,…,y n,R }; Otherwise, let i=i+1 and return to step (2.3).

Images