A method and device for adaptive control of dosage based on multimodal recognition

Abstract

The present invention discloses a method and device for adaptive control of dosing amount based on multimodal recognition, which belongs to the field of pattern recognition and flotation dosing technology. The method includes using a trained coal quality recognition model to perform real-time recognition of the feed coal quality image, and after determining the multimodal working condition according to the recognition result, combining it with the preset rules to calculate the corresponding initial dosing amount. By collecting key features such as tailings ash content, tailings white foam volume, and tailings coarseness in real time, the triangle membership fuzzy method is used to perform interval division and weighted calculation to dynamically optimize the reagent compensation amount. The dosing decision value calculated based on the initial dosing amount and the comprehensive reagent compensation amount is transmitted to the intelligent dosing station in a timely manner to complete the closed-loop operation of dosing control. The present invention comprehensively considers the feed coal quality and tailings conditions, realizes the accurate calculation of dosing amount and real-time feedback of control decision under complex working conditions, and ensures the stable operation of the flotation process.

Description

Technical Field

[0001] The present invention belongs to the technical field of pattern recognition and flotation dosing, and in particular relates to a dosing amount adaptive control method and device based on multi-modal recognition. Background Art

[0002] Flotation is a mineral separation technology that utilizes differences in the physical and chemical properties of mineral surfaces to separate useful minerals from undesirable ones using reagents. Due to the widespread use of mechanized coal mining and heavy media separation processes, the proportion of fine coal particles has increased significantly. With the increasing demand for efficient utilization of fine coal slime resources, the importance of flotation technology has become increasingly prominent.

[0003] During the flotation process, chemical dosing plays a decisive role in flotation efficiency and product quality. However, chemical dosing is affected by multiple operating conditions, including coal quality changes, foam characteristics, equipment status, and operating parameters. This makes chemical dosing control a critical link and technical challenge in the flotation process.

[0004] Currently, flotation dosing in coal preparation plants primarily uses two methods: manual dosing and intelligent dosing. Manual dosing relies on experienced operators to make adjustments based on on-site observations. However, due to perceptual bias, manual dosing suffers from issues such as delayed response and difficulty adapting to complex and changing operating conditions, leading to significant fluctuations in flotation efficiency and product quality. In recent years, with the advancement of industrial automation, intelligent dosing technology has been widely researched and applied. Intelligent dosing methods utilize online detection data, combined with flotation models and control algorithms, to automatically adjust the reagents. However, existing intelligent dosing methods are often designed for a single operating condition. When operating conditions change significantly, these methods lack adaptability and struggle to meet the control requirements under multimodal conditions.

[0005] In summary, existing technologies are unable to effectively address the adaptability and stability issues of flotation dosing under multimodal conditions. Therefore, exploring an intelligent dosing method that can effectively cope with multimodal conditions is of great significance for improving flotation efficiency and product quality. Summary of the Invention

[0006] Purpose of the invention: In order to solve the problems of insufficient adaptability and stability of existing dosing methods under multimodal working conditions, the present invention provides a dosing adaptive control method based on multimodal identification. This method uses a coal quality identification model to perform real-time analysis of the quality of the incoming coal, sets the initial dosing amount, combines the dynamic detection of key indicators such as tailings ash content, foam volume, and coarseness, calculates the comprehensive dosing compensation amount, combines the compensation result with the initial dosing amount through a control unit, generates a final dosing decision value, and realizes real-time dynamic adjustment of the dosing process. This method can effectively cope with the real-time changes of multimodal working conditions such as coal quality, foam, and coarseness, overcomes the limitations of existing dosing technology under complex working conditions, and realizes stable control of flotation dosing.

[0007] Technical solution: The present invention provides a method for adaptively controlling dosage of a drug based on multimodal recognition, comprising the following steps:

[0008] Step 1: Set the initial dosage, build and train a coal quality recognition model, use the trained coal quality recognition model to perform real-time recognition of the incoming material video, determine the multimodal working conditions based on the recognition results, and combine them with the preset rules to output the initial dosage corresponding to different coal quality conditions;

[0009] Step 2: Calculate the dosage compensation amount. First, analyze and identify the tailings image using the tailings online ash analyzer, foam detection algorithm, and coarseness detection algorithm to obtain real-time tailings ash content, foam volume, and coarseness. Then, calculate the dosage compensation amount based on ash content, foam volume, and coarseness using the triangular membership fuzzy method. Finally, obtain the comprehensive dosage compensation amount through weighted summation.

[0010] Step 3, dosing closed-loop control, first determine the dosing amount decision value based on the comprehensive dosing compensation amount and the initial dosing amount, then establish a communication connection with the intelligent dosing station, and perform dosing control based on the dosing amount decision value.

[0011] Furthermore, step 1 specifically involves first collecting image data of the coal on the feeding belt during normal operation of the coal preparation plant. Then, based on the coal quality, the collected coal image data is divided into three coal quality categories: more lump coal, a balance of both, and more fine coal. Based on the experience of on-site workers, different initial dosages are set for the three coal quality categories to complete the dataset construction.

[0012] The collected feed image data is used as input and the initial dosage of the corresponding coal quality category is used as a label to train the coal quality identification model, determine the reliable parameters of the coal quality identification model, and obtain a trained coal quality identification model.

[0013] Furthermore, in step 2, the foam amount detection algorithm includes the following process:

[0014] By using formula (1), the detection area image in the original image is extracted;

[0015] I ROI (x,y)=Crop(I(x,y),x0,y0,w,h)(1)

[0016] The input image I(x,y) is cropped by Crop(·) to limit the image to the rectangular range (x0,y0,w,h) to obtain the detection area image I ROI (x,y);

[0017] By using formula (2), the obtained detection area image I ROI (x,y) converted to grayscale image Ig (x, y), and perform threshold segmentation to generate a binary image I b (x,y);

[0018]

[0019] Where, Respectively represent the red, green, and blue channel values ​​at the image pixel (x, y) of the detection area, 0.299, 0.587, and 0.114 represent the weights corresponding to the three channels, 1 represents the foam area, 0 represents the water surface area, and T represents the brightness threshold;

[0020] By formula (3), the binary image I b (x, y) is segmented into connected regions to obtain the foam region, and the total area A of the foam region is calculated by pixels. foam And the total area of ​​the detection area A ROI , the foam volume P is calculated based on the area values ​​of the two foam ;

[0021]

[0022] Where R i Represents the set of pixels contained in the i-th connected region, A i Represents the area of ​​the i-th connected region, N represents the number of connected regions after segmentation, w represents the width of the detection area, and h represents the height of the detection area.

[0023] Furthermore, in step 2, the roughness detection algorithm includes the following process:

[0024] By using formula (1) and formula (2), the original image L(x,y) of the tailings coarseness is extracted for detection area and converted into a grayscale image L g (x,y);

[0025] By using formula (4), the grayscale image L g (x,y) is Gaussian blurred to obtain the blurred image L b (x,y), and then the blurred image L b (x,y) is binarized using the adaptive threshold method to generate a binary image L t (x,y);

[0026]

[0027] Where G(i,j) is the Gaussian kernel, which indicates the influence of the pixel offset (i,j) on the center point, k is the radius of the Gaussian kernel, D(x,y) is the weighted average of the pixel grayscale values ​​in the local window, and C is the adjustment parameter used to refine the threshold.

[0028] By formula (5), the obtained binary image L t (x, y) performs morphological operations to obtain the morphologically processed image L m (x, y), detect all connected contours, calculate the area of ​​each contour, set a threshold to exclude contours with too large an area, and the number of retained contour elements is the number of detected roughness N d ;

[0029]

[0030] In the formula, MO(·) represents the morphological operation including corrosion and expansion, K is the structural element of the morphological operation, and C i represents the pixel set of the i-th connected contour, B i represents the area of ​​the i-th contour, B max represents the maximum area threshold, n represents the number of total contours, C f represents the set of retained contours, and |·| represents the operation of extracting the number of elements in the set.

[0031] Furthermore, in step 2, the triangular membership fuzzy method is used to calculate the dosage compensation amount based on the ash content. Specifically, according to the detection results of the online ash content meter, the ash content value is fuzzy processed using the triangular membership function, the ash content value is divided into three fuzzy intervals, and the corresponding dosage compensation amount is calculated according to the membership. The specific method is as follows:

[0032] Too low interval: The membership degree of the ash value to the "too low" fuzzy set is:

[0033]

[0034] Lower interval: The membership degree of the ash value to the “lower” fuzzy set is:

[0035]

[0036] Normal interval: The membership degree of the ash value to the “normal” fuzzy set is:

[0037]

[0038] Among them, x 灰 represents the gray value of real-time detection, a1, a2, a3 are the upper bound, median and lower bound of the triangular fuzzy based on gray, M 灰1 , M 灰2 , M 灰3 Indicates the benchmark dosing compensation amount corresponding to the too low, low and normal ash content ranges set by experts;

[0039] According to the maximum membership principle, the membership interval of the current ash value is determined, and the ash-based dosing in the corresponding interval is obtained. Furthermore, in step 2, the triangular membership fuzzy method is used to calculate the foam-based dosing compensation amount. Specifically, according to the foam detection result, the foam amount is fuzzy processed using the triangular membership function, the foam amount is divided into three fuzzy intervals, and the corresponding dosing compensation amount is calculated according to the membership. The specific method is as follows:

[0040] Excessive interval: The degree of membership of the foam quantity to the “excessive” fuzzy set is:

[0041]

[0042] More interval: The membership degree of the foam quantity to the “more” fuzzy set is:

[0043]

[0044] Normal interval: The membership degree of the foam quantity to the “normal” fuzzy set is:

[0045]

[0046] Among them, x 泡 represents the amount of foam detected in real time, b1, b2, b3 are the upper bound, median and lower bound of the triangular fuzzy based on foam, M 泡1 , M 泡2 , M 泡3 It indicates the benchmark dosing compensation amount corresponding to the excessive, high and normal foam volume ranges set by experts;

[0047] According to the maximum membership principle, the current foam amount membership interval is determined, and the foam-based dosing compensation amount M of the corresponding interval is obtained. 泡 .

[0048] Furthermore, in step 2, the triangular membership fuzzy method is used to calculate the dosage compensation amount based on the roughing. Specifically, according to the roughing detection result, the roughing amount is fuzzified using the triangular membership function, the roughing amount is divided into three fuzzy intervals, and the corresponding dosage compensation amount is calculated according to the membership. The specific method is as follows:

[0049] Too many intervals: The membership degree of the running quantity belonging to the "too many" fuzzy set is:

[0050]

[0051] More interval: The membership degree of the running quantity belonging to the "more" fuzzy set is:

[0052]

[0053] Normal interval: The membership degree of the running rough quantity to the "normal" fuzzy set is:

[0054]

[0055] Among them, x 粗 represents the amount of roughness detected in real time, c1, c2, c3 are the upper bound, median and lower bound of the triangular fuzzy based on roughness, M 粗1 , M 粗2 , M 粗3 It indicates the benchmark dosage compensation amount corresponding to the excessive, relatively high and normal running amount intervals set by experts;

[0056] According to the maximum membership principle, determine the membership interval of the current roughing amount and obtain the roughing-based dosing compensation amount M in the corresponding interval. 粗 .

[0057] Furthermore, in step 2, the comprehensive dosing compensation amount is calculated by the following formula:

[0058] M 综 =ω1·M 灰 +ω2·M 泡 +ω3·M 粗 (15)

[0059] Where M 灰 、M 泡 、M 粗 They represent the compensation amount of dosing based on ash, foam and coarseness calculated in the preset interval, ω1, ω2, ω3 represent the weights of the compensation amount of dosing based on ash, foam and coarseness, M 综 Indicates the comprehensive dosing compensation amount, and the value can be negative.

[0060] Furthermore, step 3 is specifically as follows: in the dosing control process, the initial dosing amount M is firstly calculated by formula (16). 初 and comprehensive dosing compensation amount M 综 The dosage decision value M is obtained by superposition 决 Then, a communication connection is established with the intelligent dosing station through the PLC, and the dosing decision value is converted into an actual dosing control signal to control the amount of reagent added during the flotation process in real time;

[0061] M 决 =M 初 +M 综 (16)

[0062] The present invention also discloses a drug dosage adaptive control device based on multimodal recognition, the device comprising an initial drug dosage setting unit, a drug dosage compensation unit and a drug dosage control unit;

[0063] The initial dosing setting unit uses a trained coal quality recognition model to perform real-time recognition of the feed video, determines the multimodal working conditions based on the recognition results, and combines them with preset rules to output the initial dosing amounts corresponding to different coal quality conditions. By matching the real-time recognition results with the rules, the initial dosing amount can be dynamically adjusted under variable feed conditions.

[0064] The dosing compensation unit first analyzes and identifies the tailings image using an online tailings ash meter, a foam detection algorithm, and a coarse run detection algorithm to obtain real-time tailings ash content, foam content, and coarse run content. It then calculates the dosing compensation amount based on the ash content, foam content, and coarse run content using a triangular membership fuzzy method, and then obtains a comprehensive dosing compensation amount through weighted summation. The dosing compensation amount is dynamically adjusted by comprehensively analyzing multiple key tailings parameters.

[0065] The dosing control unit first determines the dosing amount decision value by combining the comprehensive dosing compensation amount with the initial dosing amount, then establishes a communication connection with the intelligent dosing station through the PLC, converts the dosing amount decision value into an actual dosing control signal, and controls the amount of reagent added during the flotation process in real time; through closed-loop control, real-time optimization of the dosing amount is achieved.

[0066] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0067] (1) The present invention provides a flotation intelligent dosing control method under complex working conditions. The method uses an image recognition model, an online ash analyzer, and a detection algorithm to perform real-time detection of feed coal quality, tailings ash content, tailings foam content, and tailings coarseness. Based on multimodal working condition modeling, it achieves precise adaptation to complex working conditions. Through an adaptive dosing strategy, the initial dosing amount is dynamically adjusted. In combination with the triangular membership fuzzy control method, the dosing compensation amount based on ash content, foam content, and coarseness is comprehensively calculated to optimize the dosing decision.

[0068] (2) The present invention uses a PLC-intelligent dosing station closed-loop control system to convert the calculated dosing amount into an actual dosing control signal, achieving full-process automated control. Compared with traditional dosing methods, this method not only achieves adaptive, precise, and stable dosing optimization under multimodal and complex working conditions, improving flotation recovery, but also reduces reagent consumption and manual intervention through closed-loop control and intelligent feedback mechanisms, ensuring the stability and economy of the flotation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic diagram of the process of the present invention.

[0070] Figure 2 Flowchart for training and testing coal quality identification model.

[0071] Figure 3 This is a diagram of the foam amount detection process.

[0072] Figure 4 This is a diagram of the rough inspection process. DETAILED DESCRIPTION

[0073] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0074] The present invention provides a flotation intelligent dosing control method under complex working conditions, the flow chart is as follows Figure 1 As shown, it includes the following units:

[0075] The initial dosing setting unit uses a trained coal quality recognition model to perform real-time recognition of incoming material video. Based on the recognition results, it determines the multimodal operating conditions and combines them with pre-set rules to output the corresponding initial dosing amount for different coal quality conditions. By matching the real-time recognition results with the rules, the initial dosing amount can be dynamically adjusted under varying incoming material conditions, improving the adaptability of the dosing process.

[0076] The dosing compensation unit first analyzes and identifies tailings images using an online tailings ash analyzer, foam detection algorithm, and coarse run detection algorithm to obtain real-time tailings ash content, foam volume, and coarse run. A triangular membership fuzzy method is then used to calculate the dosing compensation amount based on ash, foam, and coarse run. A weighted summation is then performed to obtain the comprehensive dosing compensation amount. By comprehensively analyzing multiple key tailings parameters, the dosing compensation amount is dynamically adjusted to improve the stability of flotation dosing.

[0077] The dosing control unit first determines the dosage decision value by combining the comprehensive dosage compensation amount with the initial dosage. It then establishes a communication link with the intelligent dosing station via the PLC, converting the dosage decision value into an actual dosing control signal to control the dosage of reagents added during the flotation process in real time. This closed-loop control enables real-time optimization of the dosage, effectively responding to dynamic changes in complex operating conditions.

[0078] In the initial dosing setting unit, the training and testing process of the trained coal quality identification model is as follows: Figure 2 As shown:

[0079] Step A1: real-time collection of input coal quality image data;

[0080] Step A2: performing cropping, denoising and normalization preprocessing on the collected input coal quality image data;

[0081] Step A3: Based on manual experience, the pre-processed image data is divided into three categories: more lump coal, a balance of the two, and more fine coal, and then different initial dosages are calibrated for the three categories;

[0082] Step A4: Use the preprocessed image as the input of the coal quality recognition model and the manually calibrated image category as the output to perform model training;

[0083] Step A5: The trained coal quality recognition model performs real-time recognition on the pre-processed feed coal quality image, matches the manually calibrated initial dosing amount according to the judgment category, and obtains the final initial dosing amount value M 初 .

[0084] The coal quality identification model can combine manual experience and real-time feeding conditions to output the appropriate initial dosage in a timely manner.

[0085] In the dosing compensation unit, the dosing compensation amount based on ash content is calculated using the triangular membership fuzzy method: according to the detection results of the online ash content meter, the ash content value is fuzzy processed using the triangular membership function, the ash content value is divided into three fuzzy intervals, and the corresponding dosing compensation amount is calculated according to the membership. The specific method is as follows:

[0086] Too low interval: The membership degree of the ash value to the "too low" fuzzy set is:

[0087]

[0088] Lower interval: The membership degree of the ash value to the “lower” fuzzy set is:

[0089]

[0090] Normal interval: The membership degree of the ash value to the “normal” fuzzy set is:

[0091]

[0092] Among them, x 灰 represents the gray value of real-time detection, a1, a2, a3 are the upper bound, median and lower bound of the triangular fuzzy based on gray, M 灰1 , M 灰2 , M 灰3 It indicates the benchmark dosage compensation amount corresponding to the too low, low and normal ash content ranges set by experts.

[0093] According to the maximum membership principle, determine the membership interval of the current ash value and calculate the dosage compensation amount M based on the ash content. 灰 .

[0094] In the dosing compensation unit, the foam amount detection process is as follows Figure 3 As shown, according to the detection algorithm, the process of calculating the foam-based dosing compensation amount using the triangular membership fuzzy method includes the following steps:

[0095] Step B1: Extract the detection area image from the original image using formula (1) to improve the detection accuracy.

[0096] I ROI(x,y)=Crop(I(x,y),x0,y0,w,h)(1)

[0097] The input image I(x,y) is cropped by Crop(·) to limit the image to the rectangular range (x0,y0,w,h) to obtain the detection area image I ROI (x,y).

[0098] Step B2: Using formula (2), the obtained detection area image I ROI (x,y) converted to grayscale image I g (x, y), and perform threshold segmentation to generate a binary image I b (x,y).

[0099]

[0100] Where, They represent the red, green, and blue channel values ​​at the image pixel (x, y) in the detection area, respectively. 0.299, 0.587, and 0.114 represent the weights corresponding to the three channels, respectively. 1 represents the foam area, 0 represents the water surface area, and T represents the brightness threshold.

[0101] Step B3: Using formula (3), the binary image I b (x, y) is segmented into connected regions to obtain the foam region, and the total area A of the foam region is calculated by pixels. foam And the total area of ​​the detection area A ROI , the foam volume P is calculated based on the area values ​​of the two foam .

[0102]

[0103] Where R i Represents the set of pixels contained in the i-th connected region, A i Represents the area of ​​the i-th connected region, N represents the number of connected regions after segmentation, w represents the width of the detection area, and h represents the height of the detection area.

[0104] Step B4: Based on the detected foam volume P foam , combined with the triangular membership fuzzy method to calculate the foam-based dosing compensation amount, the process is as follows:

[0105] Excessive interval: The degree of membership of the foam quantity to the “excessive” fuzzy set is:

[0106]

[0107] More interval: The membership degree of the foam quantity to the “more” fuzzy set is:

[0108]

[0109] Normal interval: The membership degree of the foam quantity to the “normal” fuzzy set is:

[0110]

[0111] Among them, x 泡 represents the amount of foam detected in real time, b1, b2, b3 are the upper bound, median and lower bound of the triangular fuzzy based on foam, M 泡1 , M 泡2 , M 泡3 It indicates the benchmark dosing compensation amount corresponding to the excessive, high and normal foam volume ranges set by experts.

[0112] According to the maximum membership principle, the current foam amount membership interval is determined, and the calculation of the corresponding interval is based on the foam dosing compensation amount M 泡 .

[0113] In the dosing compensation unit, the rough running detection process is as follows Figure 4 As shown, according to the detection algorithm, the process of calculating the dosing compensation amount based on running roughness includes the following steps:

[0114] Step C1: Using formula (1) and formula (2), the original image L(x, y) of the tailings coarseness is extracted for detection area and converted into a grayscale image L g (x,y).

[0115] Step C2: Use formula (4) to obtain the grayscale image L g (x,y) is Gaussian blurred to obtain the blurred image L b (x,y), and then the blurred image L b (x,y) is binarized using the adaptive threshold method to generate a binary image L t (x,y).

[0116]

[0117] Where G(i,j) is the Gaussian kernel, which indicates the influence of the pixel offset (i,j) on the center point, k is the radius of the Gaussian kernel, D(x,y) is the weighted average of the pixel grayscale values ​​in the local window, and C is the adjustment parameter used to refine the threshold.

[0118] Step C3: Use formula (5) to obtain the binary image L t (x, y) performs morphological operations to obtain the morphologically processed image L m(x, y), detect all connected contours, calculate the area of ​​each contour, set a threshold to exclude contours with too large an area, and the number of retained contour elements is the number of detected roughness N d .

[0119]

[0120] In the formula, MO(·) represents the morphological operation including corrosion and expansion, K is the structural element of the morphological operation, and C i represents the pixel set of the i-th connected contour, B i represents the area of ​​the i-th contour, B max represents the maximum area threshold, n represents the number of total contours, C f represents the set of retained contours, and |·| represents the operation of extracting the number of elements in the set.

[0121] Step C4: Based on the detected running amount N d , combined with the triangular membership fuzzy method to calculate the foam-based dosing compensation amount, the process is as follows:

[0122] Too many intervals: The membership degree of the running quantity belonging to the "too many" fuzzy set is:

[0123]

[0124] More interval: The membership degree of the running quantity belonging to the "more" fuzzy set is:

[0125]

[0126] Normal interval: The membership degree of the running rough quantity to the "normal" fuzzy set is:

[0127]

[0128] Among them, x 粗 represents the amount of roughness detected in real time, c1, c2, c3 are the upper bound, median and lower bound of the triangular fuzzy based on roughness, M 粗1 , M 粗2 , M 粗3 It indicates the benchmark dosing compensation amount corresponding to the excessive, more and normal running amount ranges set by experts.

[0129] According to the maximum membership principle, determine the current roughing amount membership interval and obtain the corresponding roughing-based dosing compensation amount M 粗 .

[0130] In the dosing compensation unit, the comprehensive dosing compensation amount is calculated using the following formula:

[0131] M 综 =ω1·M灰 +ω2·M 泡 +ω3·M 粗 (15)

[0132] Where M 灰 、M 泡 、M 粗 They represent the compensation amount of dosing based on ash, foam and coarseness calculated in the preset interval, ω1, ω2, ω3 represent the weights of the compensation amount of dosing based on ash, foam and coarseness, M 综 Represents the comprehensive dosing compensation amount, which can be negative. This weighted calculation can flexibly adjust the dosing compensation weights based on ash content, foam, and coarseness under different working conditions, achieving coordinated optimization of multiple factors.

[0133] In the dosing control unit, the initial dosing amount M is first calculated using formula (16). 初 and comprehensive dosing compensation amount M 综 The dosage decision value M is obtained by superposition 决 Then, a communication connection is established with the intelligent dosing station through the PLC, and the dosing decision value is converted into an actual dosing control signal to control the dosing amount in real time during the flotation process.

[0134] M 决 =M 初 +M 综 (16)

[0135] Through the above process, the intelligent dosing control of coal flotation under complex working conditions is completed.

Claims

1. A method for adaptive control of dosage based on multimodal recognition, characterized in that: The following steps are involved: Step 1: Set the initial dosage, build and train a coal quality recognition model, use the trained coal quality recognition model to perform real-time recognition of the incoming material video, determine the multimodal working conditions based on the recognition results, and combine them with the preset rules to output the initial dosage corresponding to different coal quality conditions; Step 2: Calculate the dosage compensation amount. First, analyze and identify the tailings image using the tailings online ash analyzer, foam detection algorithm, and coarseness detection algorithm to obtain real-time tailings ash content, foam volume, and coarseness. Then, calculate the dosage compensation amount based on ash content, foam volume, and coarseness using the triangular membership fuzzy method. Finally, obtain the comprehensive dosage compensation amount through weighted summation. Among them, the calculation of the ash-based dosing compensation amount by the triangular membership fuzzy method is based on the detection results of the online ash analyzer, and the ash value is fuzzy processed using the triangular membership function, the ash value is divided into three fuzzy intervals, and the corresponding dosing compensation amount is calculated according to the membership; The calculation of the dosing compensation amount based on foam by the triangular membership fuzzy method is based on the foam detection results, the foam amount is fuzzy processed by using the triangular membership function, the foam amount is divided into three fuzzy intervals, and the corresponding dosing compensation amount is calculated according to the membership; The calculation of the dosing compensation amount based on roughing is based on the roughing detection results, and the triangular membership function is used to fuzzy process the roughing amount, divide the roughing amount into three fuzzy intervals, and calculate the corresponding dosing compensation amount according to the membership. Step 3, dosing closed-loop control, first determine the dosing amount decision value based on the comprehensive dosing compensation amount and the initial dosing amount, then establish a communication connection with the intelligent dosing station, and perform dosing control based on the dosing amount decision value.

2. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: Step 1 involves first collecting image data of the coal on the feed belt during normal operation of the coal preparation plant. Then, based on the coal quality, the collected coal image data is divided into three categories: lump coal, balanced coal and fine coal, and fine coal. Based on the experience of on-site workers, different initial dosages are set for each of the three coal quality categories to complete the dataset construction. The collected feed image data is used as input and the initial dosage of the corresponding coal quality category is used as a label to train the coal quality identification model, determine the reliable parameters of the coal quality identification model, and obtain a trained coal quality identification model.

3. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: In step 2, the foam amount detection algorithm includes the following process: By using formula (1), the detection area image in the original image is extracted; I ROI (x,y)=Crop(I(x,y),x0,y0,w,h) (1) The input image I(x,y) is cropped by Crop(·) to limit the image to the rectangular range (x0,y0,w,h) to obtain the detection area image I ROI (x,y); By using formula (2), the obtained detection area image I ROI (x,y) converted to grayscale image I g (x, y), and perform threshold segmentation to generate a binary image I b (x,y); Where, Respectively represent the red, green, and blue channel values ​​at the image pixel (x, y) of the detection area, 0.299, 0.587, and 0.114 represent the weights corresponding to the three channels, 1 represents the foam area, 0 represents the water surface area, and T represents the brightness threshold; By formula (3), the binary image I b (x, y) is segmented into connected regions to obtain the foam region, and the total area A of the foam region is calculated by pixels. foam And the total area of ​​the detection area A ROI , the foam volume P is calculated based on the area values ​​of the two foam ; Where R i Represents the set of pixels contained in the i-th connected region, A i Represents the area of ​​the i-th connected region, N represents the number of connected regions after segmentation, w represents the width of the detection area, and h represents the height of the detection area.

4. The method for adaptive control of dosage based on multimodal recognition according to claim 3, characterized in that: In step 2, the rough detection algorithm includes the following process: By using formula (1) and formula (2), the original image L(x,y) of the tailings coarseness is extracted for detection area and converted into a grayscale image L g (x,y); By using formula (4), the grayscale image L g (x,y) is Gaussian blurred to obtain the blurred image L b (x,y), and then the blurred image L b (x,y) is binarized using the adaptive threshold method to generate a binary image L t (x,y); Where G(i,j) is the Gaussian kernel, which indicates the influence of the pixel offset (i,j) on the center point, k is the radius of the Gaussian kernel, D(x,y) is the weighted average of the pixel grayscale values ​​in the local window, and C is the adjustment parameter used to refine the threshold. By formula (5), the obtained binary image L t (x, y) performs morphological operations to obtain the morphologically processed image L m (x, y), detect all connected contours, calculate the area of ​​each contour, set a threshold to exclude contours with too large an area, and the number of retained contour elements is the number of detected roughness N d ; In the formula, MO(·) represents the morphological operation including corrosion and expansion, K is the structural element of the morphological operation, and C i represents the pixel set of the i-th connected contour, B i represents the area of ​​the i-th contour, B max represents the maximum area threshold, n represents the number of total contours, C f represents the set of retained contours, and |·| represents the operation of extracting the number of elements in the set.

5. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: In step 2, the calculation of the dosage compensation amount based on the ash content by the triangular membership fuzzy method is based on the detection results of the online ash content meter, and the ash content value is fuzzy processed using the triangular membership function, the ash content value is divided into three fuzzy intervals, and the corresponding dosage compensation amount is calculated according to the membership. The specific method is as follows: Too low interval: The membership degree of the ash value to the "too low" fuzzy set is: Lower interval: The membership degree of the ash value to the "lower" fuzzy set is: Normal interval: The membership degree of the ash value to the "normal" fuzzy set is: Among them, x 灰 represents the gray value of real-time detection, a1, a2, a3 are the upper bound, median and lower bound of the triangular fuzzy based on gray, M 灰1 , M 灰2 , M 灰3 Indicates the benchmark dosing compensation amount corresponding to the too low, low and normal ash content ranges set by experts; According to the maximum membership principle, determine the membership interval of the current ash value and obtain the ash-based dosing compensation amount M for the corresponding interval. 灰 .

6. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: In step 2, the calculation of the foam-based dosing compensation amount by the triangular membership fuzzy method is based on the foam detection result, the foam amount is fuzzy processed by the triangular membership function, the foam amount is divided into three fuzzy intervals, and the corresponding dosing compensation amount is calculated according to the membership. The specific method is as follows: Excessive interval: The degree of membership of the foam quantity to the "excessive" fuzzy set is: More interval: The degree of membership of the foam quantity to the "more" fuzzy set is: Normal interval: The degree of membership of the foam quantity to the "normal" fuzzy set is: Among them, x 泡 represents the amount of foam detected in real time, b1, b2, b3 are the upper bound, median and lower bound of the triangular fuzzy based on foam, M 泡1 , M 泡2 , M 泡3 It indicates the benchmark dosing compensation amount corresponding to the excessive, high and normal foam volume ranges set by experts; According to the maximum membership principle, the current foam amount membership interval is determined, and the foam-based dosing compensation amount M of the corresponding interval is obtained. 泡 .

7. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: In step 2, the calculation of the dosing compensation amount based on the roughing is performed by the triangular membership fuzzy method according to the roughing detection result, and the roughing amount is fuzzified by the triangular membership function, and the roughing amount is divided into three fuzzy intervals. The corresponding dosing compensation amount is calculated according to the membership. The specific method is as follows: Too many intervals: The membership degree of the running quantity belonging to the "too many" fuzzy set is: More interval: The membership degree of the running quantity belonging to the "more" fuzzy set is: Normal interval: The membership degree of the running quantity to the "normal" fuzzy set is: Among them, x 粗 represents the amount of roughness detected in real time, c1, c2, c3 are the upper bound, median and lower bound of the triangular fuzzy based on roughness, M 粗1 , M 粗2 , M 粗3 It indicates the benchmark dosage compensation amount corresponding to the excessive, relatively high and normal running amount intervals set by experts; According to the maximum membership principle, determine the membership interval of the current roughing amount and obtain the roughing-based dosing compensation amount M in the corresponding interval. 粗 .

8. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: In step 2, the comprehensive dosing compensation amount is calculated by the following formula: M 综 =ω1·M 灰 +ω2·M 泡 +ω3·M 粗 (15) Where M 灰 、M 泡 、M 粗 They represent the compensation amount of dosing based on ash, foam and coarseness calculated in the preset interval, ω1, ω2, ω3 represent the weights of the compensation amount of dosing based on ash, foam and coarseness, M 综 Indicates the comprehensive dosing compensation amount, and the value can be negative.

9. The method for adaptive control of dosage based on multimodal recognition according to claim 1, characterized in that: Step 3 is as follows: During the dosing control process, the initial dosing amount M is first calculated using formula (16). 初 and comprehensive dosing compensation amount M 综 The dosage decision value M is obtained by superposition 决 , then establish a communication connection with the intelligent dosing station through PLC, convert the dosing amount decision value into the actual dosing control signal, and control the amount of reagent added in the flotation process in real time; M 决 =M 初 +M 综 (16)。 10. A dosing adaptive control device based on multimodal recognition, used to implement the method according to claim 1, characterized in that: The device includes an initial dosing setting unit, a dosing compensation unit and a dosing control unit; The initial dosing setting unit uses a trained coal quality recognition model to perform real-time recognition of the feed video, determines the multimodal working conditions based on the recognition results, and combines them with preset rules to output the initial dosing amounts corresponding to different coal quality conditions. By matching the real-time recognition results with the rules, the initial dosing amount can be dynamically adjusted under variable feed conditions. The dosing compensation unit first analyzes and identifies the tailings image using an online tailings ash meter, a foam detection algorithm, and a coarse run detection algorithm to obtain real-time tailings ash content, foam content, and coarse run content. It then calculates the dosing compensation amount based on the ash content, foam content, and coarse run content using a triangular membership fuzzy method, and then obtains a comprehensive dosing compensation amount through weighted summation. The dosing compensation amount is dynamically adjusted by comprehensively analyzing multiple key tailings parameters. The dosing control unit first determines the dosing amount decision value by combining the comprehensive dosing compensation amount with the initial dosing amount, then establishes a communication connection with the intelligent dosing station through the PLC, converts the dosing amount decision value into an actual dosing control signal, and controls the amount of reagent added during the flotation process in real time; through closed-loop control, real-time optimization of the dosing amount is achieved.

Images