A method for identifying flow patterns of horizontal oil-gas-water three-phase flow

Abstract

The present invention belongs to the technical field of petroleum engineering measurement, and specifically relates to a method for identifying flow patterns of horizontal oil, gas and water three-phase flows. 1. A high-speed camera is used to collect flow pattern images of horizontal oil, gas and water three-phase flows; 2. An improved aggregate color feature YX, an improved local texture synthesis feature HX and an improved frequency domain feature KX are used to construct corresponding flow pattern features; 3. An improved multi-feature fusion method is used to fuse flow pattern image features and construct a multi-level fusion feature MLFKF; 4. GBDT is used to establish a horizontal oil, gas and water three-phase flow pattern identification model to identify the flow pattern of horizontal oil, gas and water three-phase flows. Through the flow pattern diagram of the horizontal oil, gas and water three-phase flow, a multi-level fusion feature is constructed to improve the existing horizontal oil, gas and water three-phase flow pattern identification accuracy and improve the existing horizontal oil, gas and water three-phase flow pattern identification efficiency.

Description

Technical field:

[0001] The invention belongs to the field of petroleum engineering and measurement technology, and in particular relates to a method for identifying flow patterns of horizontal oil-gas-water three-phase flows. Background technology:

[0002] At present, my country's oilfield exploitation has entered the middle and late stages, and the underground oil-gas-water three-phase flow is extremely common. The flow pattern of the oil-gas-water three-phase flow can be divided according to the geometric shape of the oil-gas-water three-phase flow during the flow process. The identification of the flow pattern of the oil-gas-water three-phase flow is an important part of the current oilfield dynamic monitoring field. Its identification accuracy affects the accuracy of the subsequent flow parameter measurement of the oil-gas-water three-phase flow. The flow pattern of the oil-gas-water three-phase flow depends on the fluid density, viscosity, pipe diameter, etc. For the installed oilfield pipelines, the flow pattern of the oil-gas-water three-phase flow is mainly affected by the flow rate and phase content of each phase. Horizontal pipes are one of the main pipeline forms in the oilfield exploitation process. Therefore, improving the identification accuracy of the horizontal oil-gas-water three-phase flow pattern plays an important role in oilfield monitoring and related instrument design.

[0003] Machine learning is an effective flow pattern identification method, and its identification accuracy is highly dependent on the features constructed by its target object. In view of the complex and diverse flow patterns of horizontal oil-gas-water three-phase flow, the low accuracy of conventional single feature identification, and the low identification efficiency of multi-feature fusion methods. Based on the existing identification methods of color features, local texture features, and frequency domain features, the color features, local texture features, and frequency domain features are improved according to the characteristics of horizontal oil-gas-water three-phase flow. An improved multi-feature fusion method is proposed to construct multi-level fusion features, improve the existing horizontal oil-gas-water three-phase flow pattern identification accuracy, and improve the existing horizontal oil-gas-water three-phase flow pattern identification efficiency. Summary of the invention:

[0004] The purpose of the present invention is to improve the existing horizontal oil-gas-water three-phase flow pattern identification accuracy, improve the existing horizontal oil-gas-water three-phase flow pattern identification efficiency, and provide a horizontal oil-gas-water three-phase flow pattern identification method based on multi-level fusion features.

[0005] The technical solution adopted by the present invention is: a method for identifying the flow pattern of a horizontal oil-gas-water three-phase flow based on multi-level fusion features. The method for identifying the flow pattern of a horizontal oil-gas-water three-phase flow comprises the following steps:

[0006] Step 1: Set different working conditions of horizontal oil-gas-water three-phase flow through oil phase, water phase and gas phase flow control valves, use high-speed camera device to collect flow pattern images of horizontal oil-gas-water three-phase flow under different working conditions, and establish flow pattern image database LX of horizontal oil-gas-water three-phase flow through manual flow pattern division.

[0007] In the above steps, in the horizontal oil-gas-water three-phase flow pattern image database LX, the pixel point distribution of the pth image LXp is expressed as follows:

[0008]

[0009] Where m is the total number of pixel rows in the image LXp; j is the jth row of pixel rows in the image LXp;

[0010] n is the total number of columns of LXp pixels in the image; i is the i-th column of LXp pixels in the image;

[0011] a j,i is the pixel at the jth row and ith column of the image LXp.

[0012] Step 2: Based on the horizontal oil-gas-water three-phase flow pattern image LXp collected by a high-speed camera, the improved aggregate color feature, the improved local texture synthesis feature and the improved frequency domain synthesis feature are used to construct the corresponding features.

[0013] The improved aggregate color feature, the improved local texture synthesis feature, and the improved frequency domain feature construction method are as follows:

[0014] S1: In order to further improve the ability of color features of horizontal oil-gas-water three-phase flow to express spatial information, the present invention proposes an improved method for constructing aggregate color features: histograms are established for the RGB color space and HSV color space channels of the horizontal oil-gas-water three-phase flow pattern image LXp, and feature fusion is performed using the feature vector inner product method to construct an improved aggregate color feature YX.

[0015] The improved aggregate color feature YX is expressed as:

[0016] YX={yx 1 ,yx 2 ,yx 3 ,…yx evyx ,…,yx evYmax}

[0017] Where evYmax is the maximum grayscale level of the pixels in the image LXp after the RGB and HSV channel dimensions are normalized and sorted from small to large;

[0018] yx evyx It is the feature value with gray level evyx in the improved aggregate color feature YX.

[0019] yx evyx It is expressed as:

[0020]

[0021] In the formula, C evyx Y RGB For the image LXp, the vector group with RGB channel grayscale level evyx;

[0022] C evyx Y HSV For the image LXp, the vector group whose HSV channel grayscale is evyx.

[0023] C evyx Y RGB It is expressed as:

[0024]

[0025] In the formula, R evyx is the number of pixels with gray level evyx in the R channel space of the image LXp;

[0026] G evyx is the number of pixels with gray level evyx in the G channel space of the image LXp;

[0027] B evyx is the number of pixels with gray level evyx in the B channel space of the image LXp;

[0028] R yr is the gray value of the pixel numbered yr in the image LXp, R channel space;

[0029] G yg is the gray value of the pixel numbered yg in the G channel space of the image LXp;

[0030] B yb It is the gray value of the pixel numbered yb in the B channel space of the image LXp.

[0031] C evyx Y HSV It is expressed as:

[0032]

[0033] In the formula, H evyx is the number of pixels with gray level evyx in the image LXp, H channel space;

[0034] S evyx is the number of pixels with gray level evyx in the image LXp, S channel space;

[0035] V evyx is the number of pixels with gray level evyx in the image LXp, V channel space;

[0036] H yh is the gray value of the pixel numbered yh in the image LXp, H channel space;

[0037] S ysis the gray value of the pixel numbered ys in the image LXp, S channel space;

[0038] V yv It is the gray value of the pixel numbered yv in the image LXp, V channel space.

[0039] S2: Local texture features have the advantages of rotation invariance, grayscale invariance, low computational complexity, and insensitivity to illumination changes. However, due to the complex flow pattern of horizontal oil-gas-water three-phase flow, traditional local texture features are difficult to characterize the flow pattern texture of oil-gas-water three-phase flow. To address this problem, the present invention proposes a local angle feature and a local distance feature, and uses a feature splicing method to perform feature fusion, thereby constructing an improved local texture synthesis feature.

[0040] The improved local texture synthesis feature construction method is as follows:

[0041] The improved local texture synthesis feature HX is expressed as:

[0042] HX={hx 1 ,hx 2 ,hx 3 ,…,hx evhx ,…,hx evHmax}

[0043] Where evHmax is the maximum eigenvalue of the eigenvalues ​​in the improved local texture synthesis feature HX sorted from small to large;

[0044] hx evhx The number of feature blocks whose improved local texture synthesis feature value is evh in the image LXp.

[0045] The horizontal oil-gas-water three-phase flow pattern image LXp, after grayscale processing, the distribution of feature blocks in the image is expressed as:

[0046]

[0047] Where N is the total number of feature block columns in the image LXp; I is the Ith column of the feature block in the image LXp;

[0048] M is the total number of feature block rows in the image LXp; J is the Jth row of feature blocks in the image LXp;

[0049] A J,I It is the feature block of the Jth row and the Ith column of the image LXp.

[0050] hx evhx It is expressed as:

[0051]

[0052] Where hc is the feature block number in the image LXp;

[0053] It is the improved local texture synthesis feature value of the hc-th feature block in the image LXp.

[0054] Improved local texture synthesis feature value It is expressed as:

[0055]

[0056] In the formula, is the local angle eigenvalue of the hcth feature block in the image LXp;

[0057] is the local distance eigenvalue of the hcth feature block in the image LXp;

[0058] P A is the total number of local angle feature description points of the hc-th feature block in the image LXp;

[0059] P B is the total number of local distance feature description points of the hc-th feature block in the image LXp;

[0060] P quan for The total number of feature description points.

[0061] Local angle eigenvalue It is expressed as:

[0062]

[0063] In the formula, p a is the local distance of the hcth feature block in the image LXp to the pth a Characteristic description points;

[0064] Co is the gray value of the central pixel of the hcth feature block in the image LXp;

[0065] θ is the angle index value with Co as the origin; R is the distance index value with Co as the origin;

[0066] C θ,R is the gray value of the pixel with Co as the origin, angle index value θ, and distance index value R;

[0067] is the local distance pth with the angle index value θ of the hcth feature block in the image LXp a The feature values ​​of the feature description points.

[0068] It is expressed as:

[0069]

[0070] In the formula, w(G θ )for The dynamic weight of g(G θ,R )for Proportional group value.

[0071] w(G θ ) and g(G θ,R ) are respectively expressed as:

[0072]

[0073] LG B P PB,R It is expressed as:

[0074]

[0075] In the formula, p b is the local angle of the hcth feature block in the image LXp b Characteristic description points;

[0076] is the local angle p with Co as the origin and distance index value R b Characteristic values ​​of the description points;

[0077] G R is the average grayscale value of all pixels with Co as the origin and distance index value R;

[0078] G pin is the average grayscale value of all pixels in the feature block.

[0079] S3: Gabor filter has good multi-scale characteristics and directionality. When used, it often only focuses on amplitude information and lacks full use of phase information, making it difficult to characterize the complex flow pattern boundary of horizontal oil-gas-water three-phase flow. The present invention combines the amplitude and phase information after Gabor processing with the improved local texture synthesis feature and adopts the feature addition method to fuse them, thus constructing an improved frequency domain synthesis feature KX.

[0080] The improved frequency domain synthesis feature KX is expressed as:

[0081] KX={kx 1 ,kx 2 ,kx 3 ,…,kx evkx ,…,kx evKmax}

[0082] Where evKmax is the maximum feature level in the improved frequency domain synthetic feature KX sorted from small to large by feature value;

[0083] kx evkx is the number of feature blocks with eigenvalue level evkx in the improved frequency domain synthetic feature KX.

[0084] kx evkx It is expressed as:

[0085]

[0086] Where v is the scale parameter of the Gabor filter; u is the direction parameter of the Gabor filter;

[0087] kc is the feature block number of the image LXp after being transformed by a filter with a scale parameter of v and a direction parameter of u; It is the improved frequency domain eigenvalue of the feature block numbered kc.

[0088] It is expressed as:

[0089]

[0090] In the formula, is the improved local texture synthesis feature value of the feature block kc;

[0091] PF(pfx,pfy) is the frequency domain amplitude of feature block kc;

[0092] PFθ(pfx, pfy) is the frequency domain phase of the feature block kc.

[0093] PF(pfx, pfy) and PFθ(pfx, pfy) are expressed as:

[0094]

[0095] Where, (pfx, pfy) is the center coordinate of feature block kc;

[0096] L(pfx, pfy) is the grayscale value of the center coordinates of the feature block kc.

[0097] Step three: Normalize the dimensions of the improved aggregated color features, improved local texture synthesis features and improved frequency domain synthesis features, use the improved multi-feature fusion method to construct the multi-level fusion feature MLFKF, and use the constructed multi-level fusion feature as the criterion for horizontal oil-gas-water three-phase flow pattern identification.

[0098] The multi-level fusion feature MLFKF is expressed as:

[0099] MLFKF = {mlf 1 ,mlf 2 ,mlf 3,…,mlf eml ,…,mlf emmax}

[0100] Where emmax is the maximum eigenvalue number in the hierarchical fusion feature MLFKF;

[0101] mlf eml It is the feature value numbered eml in the hierarchical fusion feature MLFKF.

[0102] mlf eml It is expressed as:

[0103]

[0104] Where mlk is the number of points on the cross section formed by the surface and the three-dimensional tensor SWZL;

[0105] MLK is the maximum number of points on the cross section formed by the surface and the three-dimensional tensor SWZL;

[0106] g(mx,my,mz) is the eigenvalue of a point on the cross section formed by the surface and the three-dimensional tensor SWZL.

[0107] The three-dimensional tensor SWZL is represented as:

[0108] SWZL=YX⊙HX⊙KX

[0109] The surface is expressed as

[0110] Mian:mx+my+mz-evml=0

[0111] Step 4: Based on the constructed multi-level fusion features, a horizontal oil-gas-water three-phase flow pattern feature database is established, GBDT is used to train the horizontal oil-gas-water three-phase flow pattern, and a horizontal oil-gas-water three-phase flow pattern identifier is generated. The horizontal oil-gas-water three-phase flow pattern identifier is used to identify the horizontal oil-gas-water three-phase flow pattern.

[0112] The beneficial effects of the present invention are as follows: an improved aggregate color feature is proposed, which integrates RGB and HSV spatial information to improve the expression of color features of horizontal oil-gas-water three-phase flow patterns; an improved local texture synthesis feature is proposed to improve the expression of texture features of horizontal oil-gas-water three-phase flow patterns; an improved frequency domain synthesis feature is proposed to improve the utilization rate of frequency domain information; an improved multi-feature fusion method is proposed to fuse the extracted horizontal oil-gas-water three-phase flow pattern features, construct a multi-level fusion feature, improve the existing horizontal oil-gas-water three-phase flow pattern recognition accuracy, and improve the existing horizontal oil-gas-water three-phase flow pattern recognition efficiency. Description of the drawings:

[0113] Figure 1It is a schematic diagram of a flow pattern image acquisition device for a horizontal oil-gas-water three-phase flow;

[0114] Figure 2 It is a flow chart for identifying flow patterns of horizontal oil-gas-water three-phase flow based on multi-level fusion features;

[0115] Figure 3 Schematic diagram of pixel distribution in feature blocks for improved local texture synthesis features;

[0116] Figure 4 A schematic diagram of flow pattern identification of horizontal oil-gas-water three-phase flow based on multi-level fusion features;

[0117] Figure 5 This is a comparison chart of the flow pattern identification accuracy of horizontal oil-gas-water three-phase flow obtained by training the flow pattern identifier based on multi-level fusion features and the flow pattern identifier obtained by training based on existing features;

[0118] Figure 6 This is a comparison diagram of the flow pattern identification efficiency of horizontal oil-gas-water three-phase flow between the flow pattern identifier obtained based on the improved multi-feature fusion method and the type identifier obtained based on the existing multi-feature fusion algorithm. Specific implementation method:

[0119] Embodiment 1

[0120] Referring to the figures, a method for identifying flow patterns of a horizontal oil-gas-water three-phase flow based on multi-level fusion features is provided. The method for identifying flow patterns of a horizontal oil-gas-water three-phase flow includes the following steps:

[0121] Step 1: Reference Figure 1 Arrange an unmanned horizontal oil-gas-water three-phase flow image acquisition device. Set different working conditions of the horizontal pipe through oil phase, water phase, and gas phase flow control valves, and use a high-speed camera device to collect 5,600 flow pattern images of horizontal oil-gas-water three-phase flow under 30 working conditions, 700 for each flow pattern. Through manual flow pattern division, establish a flow pattern image database of horizontal oil-gas-water three-phase flow.

[0122] In the above steps, in the horizontal oil-gas-water three-phase flow pattern image database LX, the pixel point distribution of the pth image LXp is expressed as follows:

[0123]

[0124] Where m is the total number of pixel rows in the image LXp; j is the jth row of pixel rows in the image LXp;

[0125] n is the total number of columns of LXp pixels in the image; i is the i-th column of LXp pixels in the image;

[0126] a j,i is the pixel at the jth row and ith column of the image LXp.

[0127] Step 2: Reference Figure 2 , 3 , to construct the horizontal oil-gas-water three-phase flow pattern features. According to the horizontal oil-gas-water three-phase flow pattern image LXp collected by a high-speed camera, the improved aggregate color feature, improved local texture synthesis feature and improved frequency domain synthesis feature are used to construct the corresponding features.

[0128] The improved aggregate color feature, the improved local texture synthesis feature, and the improved frequency domain feature construction method are as follows:

[0129] S1: In order to further improve the ability of color features of horizontal oil-gas-water three-phase flow to express spatial information, the present invention proposes an improved method for constructing aggregate color features: histograms are established for the RGB color space and HSV color space channels of the horizontal oil-gas-water three-phase flow pattern image LXp, and feature fusion is performed using the feature vector inner product method to construct an improved aggregate color feature YX.

[0130] The improved aggregate color feature YX is expressed as:

[0131] YX={yx 1 ,yx 2 ,yx 3 ,…yx evyx ,…,yx evYmax}

[0132] Where evYmax is the maximum grayscale level of the pixels in the image LXp after the RGB and HSV channel dimensions are normalized and sorted from small to large;

[0133] yx evyx It is the feature value with gray level evyx in the improved aggregate color feature YX.

[0134] yx evyx It is expressed as:

[0135]

[0136] In the formula, C evyx Y RGB For the image LXp, the vector group with RGB channel grayscale level evyx;

[0137] C evyx Y HSV For the image LXp, the vector group whose HSV channel grayscale is evyx.

[0138] C evyx Y RGB It is expressed as:

[0139]

[0140] In the formula, R evyx is the number of pixels with gray level evyx in the R channel space of the image LXp;

[0141] G evyx is the number of pixels with gray level evyx in the G channel space of the image LXp;

[0142] B evyx is the number of pixels with gray level evyx in the B channel space of the image LXp;

[0143] R yr is the gray value of the pixel numbered yr in the image LXp, R channel space;

[0144] G yg is the gray value of the pixel numbered yg in the G channel space of the image LXp;

[0145] B yb It is the gray value of the pixel numbered yb in the B channel space of the image LXp.

[0146] C evyx Y HSV It is expressed as:

[0147]

[0148] In the formula, H evyx is the number of pixels with gray level evyx in the image LXp, H channel space;

[0149] S evyx is the number of pixels with gray level evyx in the image LXp, S channel space;

[0150] V evyx is the number of pixels with gray level evyx in the image LXp, V channel space;

[0151] H yh is the gray value of the pixel numbered yh in the image LXp, H channel space;

[0152] S ys is the gray value of the pixel numbered ys in the image LXp, S channel space;

[0153] V yv It is the gray value of the pixel numbered yv in the image LXp, V channel space.

[0154] S2: Local texture features have the advantages of rotation invariance, grayscale invariance, low computational complexity, and insensitivity to illumination changes. However, due to the complex flow pattern of horizontal oil-gas-water three-phase flow, traditional local texture features are difficult to characterize the flow pattern texture of oil-gas-water three-phase flow. To address this problem, the present invention proposes a local angle feature and a local distance feature, and uses a feature splicing method to perform feature fusion, thereby constructing an improved local texture synthesis feature.

[0155] The improved local texture synthesis feature construction method is as follows:

[0156] The improved local texture synthesis feature HX is expressed as:

[0157] HX={hx 1 ,hx 2 ,hx 3 ,…,hx evhx ,…,hx evHmax}

[0158] Where evHmax is the maximum eigenvalue of the eigenvalues ​​in the improved local texture synthesis feature HX sorted from small to large;

[0159] hx evhx is the number of feature blocks in the image LXp whose improved local texture synthesis feature value is evh.

[0160] The horizontal oil-gas-water three-phase flow pattern image LXp, after grayscale processing, the distribution of feature blocks in the image is expressed as:

[0161]

[0162] Where N is the total number of feature block columns in the image LXp; I is the Ith column of the feature block in the image LXp;

[0163] M is the total number of feature block rows in the image LXp; J is the Jth row of feature blocks in the image LXp;

[0164] A J,I It is the feature block of the Jth row and the Ith column of the image LXp.

[0165] hx evhx It is expressed as:

[0166]

[0167] Where hc is the feature block number in the image LXp;

[0168] It is the improved local texture synthesis feature value of the hc-th feature block in the image LXp.

[0169] Improved local texture synthesis feature value It is expressed as:

[0170]

[0171] In the formula, is the local angle eigenvalue of the hcth feature block in the image LXp;

[0172] is the local distance eigenvalue of the hcth feature block in the image LXp;

[0173] P A is the total number of local angle feature description points of the hc-th feature block in the image LXp;

[0174] P B is the total number of local distance feature description points of the hc-th feature block in the image LXp;

[0175] P quan for The total number of feature description points.

[0176] Local angle eigenvalue It is expressed as:

[0177]

[0178] In the formula, p a is the local distance of the hcth feature block in the image LXp to the pth a Characteristic description points;

[0179] Co is the gray value of the central pixel of the hcth feature block in the image LXp;

[0180] θ is the angle index value with Co as the origin; R is the distance index value with Co as the origin;

[0181] C θ,R is the gray value of the pixel with Co as the origin, angle index value θ, and distance index value R;

[0182] is the local distance pth with the angle index value θ of the hcth feature block in the image LXp a The feature values ​​of the feature description points.

[0183] It is expressed as:

[0184]

[0185] In the formula, w(G θ )for The dynamic weight of g(G θ,R )for Proportional group value.

[0186] w(G θ ) and g(G θ,R ) are respectively expressed as:

[0187]

[0188] LG B P PB,R It is expressed as:

[0189]

[0190] In the formula, p b is the local angle of the hcth feature block in the image LXp b Characteristic description points;

[0191] is the local angle p with Co as the origin and distance index value R b Characteristic values ​​of the description points;

[0192] G R is the average grayscale value of all pixels with Co as the origin and distance index value R;

[0193] G pin is the average grayscale value of all pixels in the feature block.

[0194] S3: Gabor filter has good multi-scale characteristics and directionality. When used, it often only focuses on amplitude information and lacks full use of phase information, making it difficult to characterize the complex flow pattern boundary of horizontal oil-gas-water three-phase flow. The present invention combines the amplitude and phase information after Gabor processing with the improved local texture synthesis feature and adopts the feature addition method to fuse them, thus constructing an improved frequency domain synthesis feature KX.

[0195] The improved frequency domain synthesis feature KX is expressed as:

[0196] KX={kx 1 ,kx 2 ,kx 3 ,…,kx evkx ,…,kx evKmax}

[0197] Where evKmax is the maximum feature level in the improved frequency domain synthetic feature KX sorted from small to large by feature value;

[0198] kx evkx is the number of feature blocks with eigenvalue level evkx in the improved frequency domain synthetic feature KX.

[0199] kx evkx It is expressed as:

[0200]

[0201] Where v is the scale parameter of the Gabor filter; u is the direction parameter of the Gabor filter;

[0202] kc is the feature block number of the image LXp after being transformed by a filter with a scale parameter of v and a direction parameter of u; It is the improved frequency domain eigenvalue of the feature block numbered kc.

[0203] It is expressed as:

[0204]

[0205] In the formula, is the improved local texture synthesis feature value of the feature block kc;

[0206] PF(pfx,pfy) is the frequency domain amplitude of feature block kc;

[0207] PFθ(pfx, pfy) is the frequency domain phase of the feature block kc.

[0208] PF(pfx, pfy) and PFθ(pfx, pfy) are expressed as:

[0209]

[0210] Where, (pfx, pfy) is the center coordinate of feature block kc;

[0211] L(pfx, pfy) is the grayscale value of the center coordinates of the feature block kc.

[0212] Step three: Normalize the dimensions of the improved aggregated color features, improved local texture synthesis features and improved frequency domain synthesis features, use the improved multi-feature fusion method to construct the multi-level fusion feature MLFKF, and use the constructed multi-level fusion feature as the criterion for horizontal oil-gas-water three-phase flow pattern identification.

[0213] The multi-level fusion feature MLFKF is expressed as:

[0214] MLFKF = {mlf 1 ,mlf 2 ,mlf 3 ,…,mlf eml ,…,mlf emmax}

[0215] Where emmax is the maximum eigenvalue number in the hierarchical fusion feature MLFKF;

[0216] mlf eml It is the feature value numbered eml in the hierarchical fusion feature MLFKF.

[0217] mlf eml It is expressed as:

[0218]

[0219] Where mlk is the number of points on the cross section formed by the surface and the three-dimensional tensor SWZL;

[0220] MLK is the maximum number of points on the cross section formed by the surface and the three-dimensional tensor SWZL;

[0221] g(mx,my,mz) is the eigenvalue of a point on the cross section formed by the surface and the three-dimensional tensor SWZL.

[0222] The three-dimensional tensor SWZL is represented as:

[0223] SWZL=YX⊙HX⊙KX

[0224] The surface is expressed as

[0225] Mian:mx+my+mz-evml=0

[0226] Step 4: Reference Figure 4 , to identify the horizontal oil-gas-water three-phase flow pattern. Based on the constructed multi-level fusion features, a horizontal oil-gas-water three-phase flow pattern feature database is established, GBDT is used to train the horizontal oil-gas-water three-phase flow pattern, and a horizontal oil-gas-water three-phase flow pattern identifier is generated. The horizontal oil-gas-water three-phase flow pattern identifier is used to identify the horizontal oil-gas-water three-phase flow pattern.

[0227] The beneficial effects of the present invention are as follows: an improved aggregate color feature is proposed, which integrates RGB and HSV spatial information to improve the expression of color features of horizontal oil-gas-water three-phase flow patterns; an improved local texture synthesis feature is proposed to improve the expression of texture features of horizontal oil-gas-water three-phase flow patterns; an improved frequency domain synthesis feature is proposed to improve the utilization rate of frequency domain information; an improved multi-feature fusion method is proposed to fuse the extracted horizontal oil-gas-water three-phase flow pattern features, construct a multi-level fusion feature, improve the existing horizontal oil-gas-water three-phase flow pattern recognition accuracy, and improve the existing horizontal oil-gas-water three-phase flow pattern recognition efficiency.

[0228] Embodiment 2

[0229] In order to verify the accuracy and efficiency of the horizontal oil-gas-water three-phase flow pattern identification based on multi-level fusion features in the present invention, the horizontal oil-gas-water three-phase flow pattern feature construction method applied by the present invention and the existing method were used for experimental verification. The experimental results are as follows Figure 5 As shown in Figure 1, the flow pattern identifier based on the multi-level fusion feature training constructed by the present invention has the highest Slug flow pattern identification accuracy, up to 98.9%, and the average oil-gas-water three-phase flow pattern identification accuracy is 97.43%, which is much higher than the flow pattern identifier obtained by training based on existing features. The improved multi-feature fusion method applied by the present invention and the existing multi-feature fusion method are used for experimental verification. The experimental results are shown in Figure 1. Figure 6 As shown, the horizontal oil-gas-water three-phase flow type identifier obtained based on the multi-feature fusion method proposed in the present invention has an average identification accuracy higher than the existing method. In terms of identification speed, the Slug flow type identification speed is the highest, up to 47.19ms, and the average flow type identification speed is 48.34ms, which is better than the horizontal oil-gas-water three-phase flow type identifier obtained based on the existing multi-feature fusion method.

[0230] The present invention adopts a horizontal oil-gas-water three-phase flow pattern identification method based on multi-level fusion features to improve the existing horizontal oil-gas-water three-phase flow pattern identification accuracy and enhance the existing horizontal oil-gas-water three-phase flow pattern identification efficiency.

Claims

1. A method for identifying flow patterns of horizontal oil-gas-water three-phase flow, characterized by: The horizontal oil-gas-water three-phase flow pattern identification method comprises the following steps: Step 1: Set different working conditions of horizontal oil-gas-water three-phase flow through oil phase, water phase and gas phase flow control valves, use high-speed camera device to collect flow pattern pictures of horizontal oil-gas-water three-phase flow under different working conditions, and establish flow pattern picture database LX of horizontal oil-gas-water three-phase flow by artificial flow pattern division; In the above steps, in the horizontal oil-gas-water three-phase flow pattern image database LX, the pixel point distribution of the pth image LXp is expressed as follows: Where m is the total number of pixel rows in the image LXp; j is the jth row of pixel rows in the image LXp; n is the total number of columns of LXp pixels in the image; i is the i-th column of LXp pixels in the image; a j,i is the pixel at the jth row and ith column of the image LXp; Step 2: Based on the horizontal oil-gas-water three-phase flow pattern image LXp collected by a high-speed camera, the improved aggregate color feature, the improved local texture synthesis feature and the improved frequency domain synthesis feature are used to construct corresponding features; The improved aggregate color feature, the improved local texture synthesis feature, and the improved frequency domain feature construction method are as follows: S1: A histogram is established for the RGB color space and HSV color space channels of the horizontal oil-gas-water three-phase flow pattern image LXp, and the feature vector inner product method is used for feature fusion to construct an improved aggregate color feature YX; The improved aggregate color feature YX is expressed as: YX={yx1,yx2,yx3,…yx evyx ,…,yx evYmax } Where evYmax is the maximum grayscale level of the pixels in the image LXp after the RGB and HSV channel dimensions are normalized and sorted from small to large; yx evyx is the feature value with gray level evyx in the improved aggregate color feature YX; yx evyx It is expressed as: In the formula, C evyx Y RGB For the image LXp, the vector group with RGB channel grayscale level evyx; C evyx Y HSV is the image LXp, the vector group whose HSV channel gray level is evyx; C evyx Y RGB It is expressed as: In the formula, R evyx is the number of pixels with gray level evyx in the R channel space of the image LXp; G evyx is the number of pixels with gray level evyx in the G channel space of the image LXp; B evyx is the number of pixels with gray level evyx in the B channel space of the image LXp; R yr is the gray value of the pixel numbered yr in the image LXp, R channel space; G yg is the gray value of the pixel numbered yg in the G channel space of the image LXp; B yb is the gray value of the pixel numbered yb in the B channel space of the image LXp; C evyx Y HSV It is expressed as: In the formula, H evyx is the number of pixels with gray level evyx in the image LXp, H channel space; S evyx is the number of pixels with gray level evyx in the image LXp, S channel space; V evyx is the number of pixels with gray level evyx in the image LXp, V channel space; H yh is the gray value of the pixel numbered yh in the image LXp, H channel space; S ys is the gray value of the pixel numbered ys in the image LXp, S channel space; V yv is the gray value of the pixel numbered yv in the image LXp, V channel space; S2: A local angle feature and a local distance feature are proposed, and feature splicing method is used for feature fusion to construct an improved local texture synthesis feature; The improved local texture synthesis feature construction method is as follows: The improved local texture synthesis feature HX is expressed as: HX={hx1,hx2,hx3,…,hx evhx ,…,hx evHmax } Where evHmax is the maximum eigenvalue of the eigenvalues ​​in the improved local texture synthesis feature HX sorted from small to large; hx evhx The number of feature blocks in the image LXp where the improved local texture synthesis feature value is evhx; The horizontal oil-gas-water three-phase flow pattern image LXp, after grayscale processing, the distribution of feature blocks in the image is expressed as: Where N is the total number of feature block columns in the image LXp; I is the Ith column of the feature block in the image LXp; M is the total number of feature block rows in the image LXp; J is the Jth row of feature blocks in the image LXp; A J,I is the feature block of the Jth row and the Ith column of the picture LXp; hx evhx It is expressed as: Where hc is the feature block number in the image LXp; is the improved local texture synthesis feature value of the hc-th feature block in the image LXp; Improved local texture synthesis feature value It is expressed as: In the formula, is the local angle eigenvalue of the hcth feature block in the image LXp; is the local distance eigenvalue of the hcth feature block in the image LXp; P A is the total number of local angle feature description points of the hc-th feature block in the image LXp; P B is the total number of local distance feature description points of the hc-th feature block in the image LXp; P quan for Total number of feature description points; Local angle eigenvalue It is expressed as: In the formula, p a is the local distance of the hcth feature block in the image LXp to the pth a Characteristic description points; Co is the gray value of the central pixel of the hcth feature block in the image LXp; θ is the angle index value with Co as the origin; R is the distance index value with Co as the origin; C θ,R is the gray value of the pixel with Co as the origin, angle index value θ, and distance index value R; is the local distance pth with the angle index value θ of the hcth feature block in the image LXp a The feature values ​​of the feature description points; It is expressed as: In the formula, w(G θ )for The dynamic weight of g(G θ,R )for Ratio group value; w(G θ ) and g(G θ,R ) are respectively expressed as: It is expressed as: In the formula, p b is the local angle of the hcth feature block in the image LXp b Characteristic description points; HXT(G pb,R ) is the local angle p with Co as the origin and distance index value R b Characteristic values ​​of the description points; G R is the average grayscale value of all pixels with Co as the origin and distance index value R; G pin is the average gray value of all pixels in the feature block; S3: By combining the amplitude and phase information after Gabor processing with the improved local texture synthesis feature, the feature addition method is used to fuse and construct the improved frequency domain synthesis feature KX; The improved frequency domain synthesis feature KX is expressed as: KX={kx1,kx2,kx3,…,kx evkx ,…,kx evKmax } Where evKmax is the maximum feature level in the improved frequency domain synthetic feature KX sorted from small to large by feature value; kx evkx is the number of feature blocks with eigenvalue level evkx in the improved frequency domain synthetic feature KX; kx evkx It is expressed as: Where v is the scale parameter of the Gabor filter; u is the direction parameter of the Gabor filter; kc is the feature block number of the image LXp after being transformed by a filter with a scale parameter of v and a direction parameter of u; is the improved frequency domain eigenvalue of the feature block numbered kc; It is expressed as: In the formula, is the improved local texture synthesis feature value of the feature block kc; PF(pfx,pfy) is the frequency domain amplitude of feature block kc; PFθ(pfx,pfy) is the frequency domain phase of feature block kc; PF(pfx, pfy) and PFθ(pfx, pfy) are expressed as: Where, (pfx, pfy) is the center coordinate of feature block kc; L(pfx, pfy) is the gray value of the center coordinate of the feature block kc; Step 3: normalize the dimensions of the improved aggregate color features, the improved local texture synthesis features and the improved frequency domain synthesis features, and use the improved multi-feature fusion method to construct the multi-level fusion feature MLFKF, and use the constructed multi-level fusion feature as the criterion for the identification of the horizontal oil-gas-water three-phase flow pattern; The multi-level fusion feature MLFKF is expressed as: MLFKF={mlf1,mlf2,mlf3,…,mlf eml ,…,mlf emmax } Where emmax is the maximum eigenvalue number in the hierarchical fusion feature MLFKF; mlf eml It is the feature value numbered eml in the hierarchical fusion feature MLFKF; mlf eml It is expressed as: Where mlk is the number of points on the cross section formed by the surface and the three-dimensional tensor SWZL; MLK is the maximum number of points on the cross section formed by the surface and the three-dimensional tensor SWZL; g(mx,my,mz) is the eigenvalue of the point on the cross section formed by the surface and the three-dimensional tensor SWZL; The three-dimensional tensor SWZL is represented as: SWZL=YX⊙HX⊙KX The surface is expressed as Mian:mx+my+mz-evml=0 Step 4: Based on the constructed multi-level fusion features, a horizontal oil-gas-water three-phase flow pattern feature database is established, GBDT is used to train the horizontal oil-gas-water three-phase flow pattern, and a horizontal oil-gas-water three-phase flow pattern identifier is generated. The horizontal oil-gas-water three-phase flow pattern identifier is used to identify the horizontal oil-gas-water three-phase flow pattern.

Images