CN115452824A - Method for predicting ash content of coal flotation tailings on dynamic overflow surface through multiple models - Google Patents
Method for predicting ash content of coal flotation tailings on dynamic overflow surface through multiple models Download PDFInfo
- Publication number
- CN115452824A CN115452824A CN202211045618.0A CN202211045618A CN115452824A CN 115452824 A CN115452824 A CN 115452824A CN 202211045618 A CN202211045618 A CN 202211045618A CN 115452824 A CN115452824 A CN 115452824A
- Authority
- CN
- China
- Prior art keywords
- tailing
- model
- tailings
- image
- flotation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000005188 flotation Methods 0.000 title claims abstract description 68
- 238000000034 method Methods 0.000 title claims abstract description 59
- 239000003245 coal Substances 0.000 title claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 33
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 17
- 238000011084 recovery Methods 0.000 claims abstract description 13
- 238000012360 testing method Methods 0.000 claims abstract description 13
- 238000013528 artificial neural network Methods 0.000 claims abstract description 4
- 238000012216 screening Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 13
- 238000012544 monitoring process Methods 0.000 claims description 8
- 239000007788 liquid Substances 0.000 claims description 6
- 239000000463 material Substances 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000004913 activation Effects 0.000 claims description 2
- 238000005286 illumination Methods 0.000 claims description 2
- 238000002372 labelling Methods 0.000 claims description 2
- 238000002360 preparation method Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 238000007637 random forest analysis Methods 0.000 claims description 2
- 239000002002 slurry Substances 0.000 claims 1
- 238000004519 manufacturing process Methods 0.000 abstract description 6
- 230000008901 benefit Effects 0.000 abstract description 5
- 230000001502 supplementing effect Effects 0.000 abstract 1
- 238000005516 engineering process Methods 0.000 description 7
- 238000001514 detection method Methods 0.000 description 6
- 238000004364 calculation method Methods 0.000 description 4
- 239000013589 supplement Substances 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000005251 gamma ray Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000003556 assay Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 230000001276 controlling effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000007599 discharging Methods 0.000 description 1
- 238000001035 drying Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000011049 filling Methods 0.000 description 1
- 239000010419 fine particle Substances 0.000 description 1
- 238000007667 floating Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000000227 grinding Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000013486 operation strategy Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 238000005498 polishing Methods 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000003892 spreading Methods 0.000 description 1
- 230000007480 spreading Effects 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000000967 suction filtration Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/01—Arrangements or apparatus for facilitating the optical investigation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N2021/8405—Application to two-phase or mixed materials, e.g. gas dissolved in liquids
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Geometry (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Quality & Reliability (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention discloses a method for predicting ash content of coal flotation tailings by multiple models on a dynamic overflow surface, which comprises the steps of constructing a dynamic overflow mode and a machine vision system for horizontally supplementing light around, and capturing flotation tailings images; classifying the tailing images under different working conditions based on a convolutional neural network to realize the screening of stable target tailing images; traditional characteristics and depth abstract characteristics are fused, and a regression predictor is connected to the outer layer of the depth convolution neural network for training, so that the ash content of the flotation tailings can be predicted in real time. The method is environment-friendly, efficient and high in prediction accuracy, effectively replaces a tailing testing mode with a complex original process and time and labor consumption, avoids high delay of feedback adjustment in a flotation production process, has important significance in ensuring efficient recovery of coal resources, upgrading intelligent industries and improving factory selection benefits, and has great application potential.
Description
Technical Field
The invention relates to the technical field of coal flotation, in particular to a method for predicting coal flotation tailing ash content on a dynamic overflow surface through multiple models.
Background
Parameters of the coal flotation process are redundant and complicated, variables are nonlinear, and the detection of product indexes is seriously lagged, so that the flotation process is difficult to realize the processes of automatic operation and closed-loop control. With the iterative update of calculation and detection technologies, sensors enable the accurate capture of flotation process variables to be possible, but the crucial flotation indexes still lack reliable detection technologies in the industry, especially the ash value of flotation tailings, which relates to the calculation of multiple performance parameters such as clean coal yield, combustible recovery rate, flotation perfection indexes and the like, and are very important for the evaluation and process control of the coal flotation process.
At present, researches on quick detection of ash content in flotation tailings mainly focus on three directions, namely gamma ray, photoelectric quick detection and machine vision. CN112486228A discloses a control system of a flotation machine using a gamma-ray ash detector, but the gamma-ray is always limited due to the radiation risk, high cost and the like; the photoelectric rapid detection is to predict the ash value of flotation tailings by detecting the white light passing rate of the tailings pulp through an element, and the method is limited by the fact that a light-transmitting device is easy to wear and affects the prediction performance; the machine vision technology reflects the working condition of tailings through the visual characteristics of the surface of the tailing pulp, and the process simulates the manual operation strategy of a flotation driver for regulating and controlling flotation production, so that the machine vision technology is more environment-friendly and reliable and has larger application potential.
In the existing research, most of the tailings are directly simulated by adopting the gray value of a tailing image, and although the gray value has stronger correlation with tailing ash, many detailed characteristics are ignored, so that the obtained error is larger; in addition, the air bubbles and oil particles existing in the flotation tailings can seriously interfere with the result of a tailing ash prediction model, and the characteristics have no benefit on the prediction of the ash content of the ore pulp tailings, so that an effective solution is not provided in the existing research. The coal flotation tailing pulp image is a typical background-free image, fine particle spots, oil film spreading textures and other detail features exist on the surface except for a single gray level brightness change feature, however, the detail features are more in need of pixel level expression, and the features are usually ignored by a traditional algorithm, so that the prediction potential of the model is reduced. The deep learning method can be used for learning through a large amount of labeled data and extracting higher-order strong robustness characteristics, so that the combination of the flotation data set and the deep learning is also a hot problem in the field.
Although the related technology provides technical support for the automatic operation and closed-loop control of the coal flotation process, due to the limitation of the application effect and the overall intelligent upgrading process of the plant, the coal flotation industrial site is still detected by a fast ash method, the method is complex in process, time-consuming and labor-consuming, has high delay for guiding flotation production, and indexes such as corresponding combustible body recovery rate and clean coal yield cannot be regulated and controlled in real time, so that a mature and reliable tailing monitoring method is urgently needed by the coal flotation industry as a power point for ensuring the efficient recovery of coal resources, the intelligent industrial upgrading and the benefit improvement of the plant.
Disclosure of Invention
Aiming at the technical defects, the invention aims to provide a method for predicting the ash content of coal flotation tailings by multiple models on a dynamic overflow surface, which is used for building a machine vision system for a dynamic overflow mode and horizontal peripheral light supplement and capturing flotation tailings images; classifying the tailing images under different working conditions based on a convolutional neural network to realize the screening of stable target tailing images; traditional characteristics and depth abstract characteristics are fused, and a regression predictor is connected to the outer layer of the depth convolution neural network for training, so that the ash content of the flotation tailings is predicted in real time; the method is environment-friendly, efficient and high in prediction precision, effectively replaces a tailing assay mode with a complex process, time and labor consumption, avoids high delay of feedback adjustment in the flotation production process, has important significance in ensuring efficient recovery of coal resources, intelligent industrial upgrading and improving plant selection benefits, and has great application potential.
In order to solve the technical problem, the invention provides a method for predicting coal flotation tailing ash content on a dynamic overflow surface by multiple models, which comprises the following steps:
s1: a flotation tailing monitoring machine vision system of a dynamic overflow surface for capturing flotation tailing images is constructed and used for capturing tailing data sets for training flotation tailing models and data sources for model prediction, and the specific process is as follows:
s11: the machine vision system in the method comprises an industrial CCD camera (RGB), an LED surface light source, a tailing overflow weir body and an ore pulp recovery tank;
s12: capturing images by adopting self-developed image acquisition software, adjusting camera parameters, illumination intensity, tailing flow and setting acquisition time intervals before acquisition;
s13: preprocessing the acquired images of the tailing pulp, labeling each image correspondingly, wherein the labels are divided into working conditions and ash values, the working conditions are observed and labeled manually, and the ash values are obtained in a laboratory of a coal preparation plant by a fast ash method through the steps of suction filtration, drying, grinding, ash burning, weighing and the like;
s14: dividing a tailing data set into a training set and a testing set by adopting a random _ split random function in a Pythrch;
s2: design convolution neural network model training tailing operating mode (normal ore pulp image, oil drop pollution, bubble pollution, tailing disconnected material image) data set, the network structure includes that convolution layer and downsampling layer draw the image characteristic under the different operating modes, and the full articulamentum gathers the feature fusion, connects the classifier, discerns the operating mode image, and concrete process is:
s21: constructing an initial convolutional neural network, which comprises a convolutional layer, a downsampling layer and a full-connection layer, and obtaining network output through forward propagation, wherein the characteristics of the convolutional layer are calculated as follows:
wherein, output (i) Is the output of the convolution layer, f is the activation function, m is all the input flotation tailing data sets, w is the convolution weight value, b is the bias, x is the input characteristic;
s22: and externally connecting a softmax classifier to classify the tailing images under the four working conditions, wherein the classification result is calculated as follows:
wherein P represents the probability that the target is the jth class image, and K represents the dimension of the feature vector;
s23: the Loss function NLL _ Loss is determined and calculated as follows:
NLLLoss＝-∑log(P(y＝j|x))
and P is a classification probability value calculated by softmax, an error between the convolutional network prediction and a real label is calculated through NLL _ Loss, and a weight and a bias parameter are updated through back propagation circulation until the model reaches the optimal prediction precision, so that a target function of the convolutional neural network can be expressed as follows:
where m is the number of samples, L is the loss function, y t For the output of operating conditions, y p Is the desired output. And continuously repeating the convolutional network training in the training process until the model error is less than or equal to the expected value, and finishing one training.
S24: continuously iterating the training network according to the iteration times Epoch value set by the model until all Epoch times are trained, and finishing the model training;
s25: further, in step S21, dropout regularization parameters need to be connected to reduce redundancy of the model parameters, which can effectively avoid the over-fitting phenomenon;
s26: further verifying the model accuracy of the trained network model on a test set, and outputting the identification result of the tailing image, so as to finish a tailing working condition image prediction module in the method;
s3: on the basis of the step S2, tailings working condition images of irrelevant working conditions are eliminated through model prediction, and non-interference tailings overflow surface images are screened out to be used as tailings ash content prediction of the step, and the method is consistent with the method in the step S2, a convolutional neural network is adopted in the step, except that the traditional characteristics of gray scale and the like of the tailings ash content are fused in a full connection layer, and an RFR regression predictor is used for ash content prediction, and the method sets a prediction time interval as follows in the specific execution process:
s31: calculating the gray level mean value and the texture characteristics of the tailing image by adopting an OpenCV image processing library;
s32: similarly, a convolutional neural network consistent with the structure in step S21 is constructed, including convolutional layers, downsampling layers and full-connection layers, and network output is obtained through forward propagation.
S33: further, the input of the full connection layer in step S32 includes the gray level mean value, the texture feature in step S31 and the convolution feature in step S32;
s34: and (3) taking the output of the full-link layer in the step (2) as the input of a Random Forest Regression (RFR) model, selecting a loss function Root Mean Square Error (RMSE), and calculating according to the following formula:
wherein Y is obj Is the true ash value, Y model The grey value predicted by the model is obtained, and n is the number of tailing image samples;
s35: similar to the step S24, setting the number of training iterations, and optimizing the hyper-parameters to be optimized in the RFR by using Grid search (Grid SearchCV), including the number n _ estimators of weak learning machines, the maximum tree depth max _ depth, and the minimum cotyledon number min _ samples _ split;
s36: completing model training, and calculating model performance indexes on a test set;
s4: embedding the models in the step S2 and the step S3 into the same system, firstly predicting whether the tailing working condition image is qualified by using sequential logic, then predicting the ash value of the flotation tailings, predicting the ash value of the tailings once every 3 minutes, and representing the current real-time tailing ash by adopting a mean value of 3 minutes in order to ensure the reliability of the model result;
in conclusion, the invention discloses a method for predicting ash content of coal flotation tailings on a dynamic overflow surface by multiple models, and compared with the existing method and technology, the method has the following beneficial effects:
based on the powerful feature capturing capability of the convolutional neural network, the invention adopts a diffuse reflection LED light source surrounding horizontal surrounding light supplement method which is more suitable for light supplement on the surface of tailing pulp, and by means of the powerful feature extraction capability of the convolutional neural network, abnormal working condition images of an overflow liquid level are intelligently eliminated, so that stable target tailing images are screened; for the screened tailing pulp image, the traditional characteristics and the depth abstract characteristics are fused, a traditional gray scale and texture characteristics and a high-order abstract convolution characteristic fusion method is adopted, a regression predictor is arranged outside a depth convolution neural network to train a tailing ash prediction model, and the RFR plays an excellent ash prediction performance, so that the prediction precision of the model is finally improved, and the real-time prediction of the flotation tailing ash is realized;
the method effectively replaces the tailing assay mode with complex process, time and labor consumption, avoids high delay of feedback adjustment in the flotation production process, and has important significance in ensuring high-efficiency recovery of coal resources, intelligent industrial upgrading and improving the benefit of plant selection.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic structural diagram of a flotation tailing monitoring machine vision system in example 1 of the invention;
FIG. 2 is a sample diagram of flotation tailings of example 1 of the present invention under different working conditions; wherein 1 is a material-free working condition, 2 is an oil drop working condition, 3 is a froth working condition, and 4 is a normal working condition;
FIG. 3 is a flow chart of modeling and predicting ash content of coal flotation tailings by multiple models on a dynamic overflow surface in example 1 of the present invention;
FIG. 4 is a training curve of the working condition prediction network in embodiment 1 of the present invention.
Description of the reference numerals:
1. the tailing overflows the liquid level; 2. an overflow recovery tank; 3. a tailing feeding port; 4. a CCD industrial camera; 5. an LED area light source; 6. an overflow weir body; 7. a tailing discharge port.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
The invention provides a method for multi-model prediction of coal flotation tailing ash on a dynamic overflow surface, which comprises the following steps:
s1: the flotation tailing monitoring machine vision system shown in the figure 1 is set up, flotation tailing images are captured, and the application in the implementation is characterized in that a dynamic tailing overflow surface and a set of horizontal diffuse reflection LED polishing mode surrounding the periphery are adopted;
specifically, flotation tailing monitoring machine vision system includes overflow weir 6, machine vision system and overflow accumulator 2, and overflow weir 6 is the open circular cone structure in upper portion, and the ore pulp forms dynamic tailing overflow liquid level 1 at uncovered surface, and tailing pan feeding mouth 3 is installed to 6 bottoms of overflow weir, and 6 outside covers of overflow weir have overflow accumulator 2, tailing bin outlet 7 is installed to 2 bottoms of overflow accumulator. The ore pulp enters the overflow weir body 6 from the tailing feeding port 3, a dynamic tailing overflow liquid level 1 is formed on the surface of the overflow weir body 6, and the ore pulp is collected and recovered by the overflow recovery tank 2 and then discharged from the tailing discharging port 7.
A machine vision system is arranged above the overflow weir body 6, the machine vision system comprises a CCD industrial camera 4 vertically arranged above the overflow weir body 6 and LED area light sources 5 arranged on the periphery of the upper edge of the overflow recovery tank 2, and the installation mode of the annular horizontal light supplement of the LED area light sources 5 can reduce light pollution caused by liquid level reflection to the maximum extent.
Shooting a tailing image by adopting shooting software, and setting the original size to be 2048 × 2048 pixels and the format to be ". Png";
continuously photographing and collecting samples after the flotation machine starts to produce, detecting ash content labels, manually screening and marking working condition labels, and displaying sample diagrams of four flotation tailings under different working conditions in fig. 2;
the sorted flotation images are collected into two sets of data sets of flotation tailing working conditions and tailing ash, and the image clipping values in the data sets are 224 pixels by 224 pixels, so that the reading of a convolutional neural network is facilitated, the calculation power in the training process can be saved, the data sets are divided into training sets and testing sets according to the proportion of 80%, and the information of the sorted data sets is shown in the following table:
| working condition label | Number of samples | Ash label (%) | Number of samples |
| Is normal | 500 | ≥80 | 85 |
| Oil droplets | 500 | [70,80) | 136 |
| Floating bubble | 500 | [60,70) | 114 |
| Cutting material | 500 | <60 | 63 |
| Training set | 1600 | Training set | 310 |
| Test set | 400 | Test set | 88 |
S2: before training a working condition prediction model, training set data needs to be read in firstly, images are subjected to normalization preprocessing, then a pitorch deep learning framework is adopted to build a convolutional neural network, the optimized convolutional neural network in the embodiment is of a 5-layer structure, and the following table shows that:
wherein parameters in Conv2d are respectively expressed (input matrix depth, convolution kernel size, convolution kernel step length and all-zero filling), then three full-connection layers are connected through a network, the number of neurons is respectively 512, 256 and 4, a dropout structure with a coefficient of 0.2 is respectively connected behind the first two full-connection layers to prevent overfitting, and finally a softmax layer is connected to a model to serve as a classifier.
The optimal learning rate of the model during training is 0.01, the classification performance of the model on the test set is shown in fig. 4, and the classification accuracy on the test set reaches 98.9%, so that the method obtains better classification performance on the example working condition image classification.
S3: after the working condition image recognition is completed in the step S2, characteristic parameters of the tailings images screened in the step S2 are calculated, a gray level co-occurrence matrix of the tailings images is calculated by adopting OpenCV assistance, and six corresponding traditional characteristics of a gray level mean value, a standard deviation, a contrast ratio, dissimilarity, an inverse difference and an angular second moment are calculated.
Meanwhile, similar to the step S2, a convolutional neural network is used to extract the high-order abstract features of the tailing image, and the optimal network structure in this example is shown in the following table:
the extracted features and 6 traditional texture features are combined to be used as input of a full connection layer, the step passes through two full connection layers, the number of neurons is 1024 and 512 respectively, and a dropout structure with a coefficient of 0.2 is arranged behind the first full connection layer.
Then, the full connection layer is accessed into an RFR regressor for training, RMSE is adopted as a loss function, and the calculation method is as follows:
wherein Y is obj Is the true ash value, Y model And n is the number of tailing image samples.
The combination of the optimized RFR super parameters in this example is shown in the following table:
| hyper-parameter | Value of |
| n_estimators | 80 |
| |
7 |
| min_samples_split | 0.6 |
After model training is completed, the prediction accuracy RMSE =1.06 of the model on the tailing ash test set, and the result can effectively guide the production of the flotation process.
S4: and (3) sequentially connecting the models in the step (S2) and the step (S3) according to the flow in the figure 3, firstly predicting the working condition represented by the flotation tailing image, and if the image represents the normal tailing working condition, entering an ash content prediction model to predict the ash value represented by the image. If not, inputting a new industrial production image.
By adopting the mode, the real-time monitoring of the ash content of the coal flotation tailings can be realized. In specific implementation, the method provided in the present invention can be autonomously implemented by a person skilled in the art using a computer programming technology, and a computer program, a hardware device, a storage instruction, and the like corresponding to a computer algorithm for implementing the method also should be within the protection scope of the present invention.
The described embodiments of the present invention are intended to be illustrative rather than restrictive, and it should be apparent that various changes and modifications can be made herein by one skilled in the art without departing from the spirit and scope of the invention, and it is intended that all other embodiments of the invention including those skilled in the art that fairly fall within the scope of the appended claims are to be interpreted as broadly as the invention is entitled.
Claims (9)
1. A method for multi-model prediction of coal flotation tailing ash on a dynamic overflow surface is characterized by comprising the following steps:
s1: constructing a flotation tailing monitoring machine vision system for capturing a dynamic overflow surface of a flotation tailing image, and capturing a tailing data set for training a flotation tailing model and a data source for predicting the model;
s2: designing a convolutional neural network model to train different tailing working condition data sets, wherein the network structure comprises a convolutional layer and a downsampling layer to extract image characteristics under different working conditions, a full connecting layer integrates the characteristics, and a classifier is connected to recognize working condition images;
s3: predicting the tailing working condition image without the irrelevant working condition through the model in the step S2, screening out an interference-free tailing overflow surface image to be used for tailing ash prediction, training a tailing ash prediction model by adopting a method of fusing the traditional gray scale, texture characteristics and high-order abstract convolution characteristics, and training an RFR regression predictor outside a deep convolution neural network;
s4: and (3) embedding the models in the step (S2) and the step (S3) into the same system, firstly predicting whether the tailings working condition image is qualified by using sequential logic, and then predicting the ash content of the flotation tailings.
2. The method according to claim 1, wherein the flotation tailings monitoring machine vision system in step S1 comprises an overflow weir, a machine vision system and an overflow recovery tank, wherein the ore slurry enters the overflow weir from a tailings inlet, forms a dynamic tailings overflow liquid level on the open surface of the overflow weir, and is discharged from a tailings outlet after being collected and recovered by the overflow recovery tank;
the machine vision system is arranged above the overflow weir body and comprises a CCD industrial camera vertically arranged above the overflow weir body and a horizontal diffuse reflection LED surface light source surrounding the periphery of the overflow recovery groove.
3. The method for multi-model prediction of ash content in coal flotation tailings on a dynamic overflow surface as claimed in claim 2, wherein the step S1 comprises the following steps:
s11: capturing images by adopting image acquisition software, adjusting camera parameters, illumination intensity, tailing flow and setting acquisition time intervals before acquisition;
s12: preprocessing the acquired tailing pulp images, and labeling each image correspondingly, wherein the labels are divided into working conditions and ash values, the working conditions are observed and labeled manually, and the ash values are acquired by a fast ash method in a laboratory of a coal preparation plant;
s13: and dividing the tailing data set into a training set and a testing set by adopting a random _ split random function in the Pythrch.
4. The method for multi-model prediction of ash content in coal flotation tailings on a dynamic overflow surface as claimed in claim 1, wherein the step S2 comprises:
s21: constructing an initial convolutional neural network, comprising a convolutional layer, a downsampling layer and a full-connection layer, obtaining network output through forward propagation, wherein the characteristics of the convolutional layer are calculated as follows:
wherein, output (i) Is the output of the convolutional layer, f is the activation function, m is the flotation tailing data set of all inputs, w is the convolution weightThe value, b is the bias, x is the input characteristic;
s22: the external softmax classifier classifies the tailing images under different working conditions, and the classification result is calculated as follows:
wherein P represents the probability that the target is the jth class image, and K represents the dimensionality of the feature vector;
s23: the Loss function NLL _ Loss is determined and calculated as follows:
NLLLoss＝-∑log(P(y＝j|x))
and P is a classification probability value calculated by softmax, an error between the convolutional network prediction and a real label is calculated through NLL _ Loss, and a weight and a bias parameter are updated through back propagation circulation until the model reaches the optimal prediction precision, so that the target function of the convolutional neural network can be expressed as follows:
where m is the number of samples, L is the loss function, y t For the output of operating conditions, y p Is the desired output; continuously repeating the convolutional network training in the training process until the model error is less than or equal to the expected value, and finishing one training;
s24: continuously iterating the training network according to the iteration times Epoch value set by the model until all Epoch times are trained, and finishing model training;
s25: and further verifying the model precision of the trained network model on a test set, and outputting the identification result of the tailing image, so as to complete the prediction module of the tailing condition image in the method.
5. The method for multi-model prediction of ash content in coal flotation tailings on a dynamic overflow surface as claimed in claim 4, wherein Dropout regularization parameters are connected in the step S21 to reduce redundancy of model parameters, so that an over-fitting phenomenon can be effectively avoided.
6. The method for multi-model prediction of ash content in coal flotation tailings on a dynamic overflow surface as claimed in claim 1, wherein the step S3 comprises:
s31: calculating the gray level mean value and the texture characteristics of the tailing image by adopting an OpenCV image processing library;
s32: constructing a convolutional neural network with a structure consistent with that in the step S21, wherein the convolutional neural network comprises a convolutional layer, a downsampling layer and a full-connection layer, and obtaining network output through forward propagation;
s33: and (4) taking the output of the full-link layer in the step (S32) as the input of a Random Forest Regression (RFR) model, selecting a loss function Root Mean Square Error (RMSE), and calculating according to the following formula:
wherein Y is obj Is the true ash value, Y model The grey value predicted by the model is obtained, and n is the number of tailing image samples;
s34: setting training iteration times, and optimizing hyper-parameters to be optimized in RFR by adopting Grid search (Grid SearchCV), wherein the hyper-parameters include the number n _ estimators of weak learning machines, the maximum tree depth max _ depth and the minimum cotyledon number min _ samples _ split;
s36: and finishing model training and calculating the performance index of the model on the test set.
7. The method for multi-model prediction of ash content in coal flotation tailings on a dynamic overflow surface as claimed in claim 5, wherein the input of the full-connected layer in step S32 comprises the gray-scale mean value, the texture feature and the convolution feature in step S31.
8. The method for multi-model prediction of ash content of coal flotation tailings on a dynamic overflow surface as claimed in claim 1, wherein in step S4, due to relatively smooth fluctuation of the ash content of the tailings, a prediction time interval is set, the ash content of the tailings is predicted every 3 minutes, and in order to ensure reliability of a model result, a mean value of 3 minutes is adopted to represent the current real-time ash content of the tailings.
9. The method for multi-model prediction of coal flotation tailing ash on a dynamic overflow surface of claim 1, wherein the different conditions comprise a no-material condition, an oil drop condition, a froth condition, and a normal condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211045618.0A CN115452824A (en) | 2022-08-30 | 2022-08-30 | Method for predicting ash content of coal flotation tailings on dynamic overflow surface through multiple models |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211045618.0A CN115452824A (en) | 2022-08-30 | 2022-08-30 | Method for predicting ash content of coal flotation tailings on dynamic overflow surface through multiple models |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115452824A true CN115452824A (en) | 2022-12-09 |
Family
ID=84300984
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211045618.0A Pending CN115452824A (en) | 2022-08-30 | 2022-08-30 | Method for predicting ash content of coal flotation tailings on dynamic overflow surface through multiple models |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115452824A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116738120A (en) * | 2023-08-11 | 2023-09-12 | 齐鲁工业大学(山东省科学院) | Copper grade SCN modeling algorithm for X fluorescence grade analyzer |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104084293A (en) * | 2014-07-29 | 2014-10-08 | 四川达竹煤电（集团）有限责任公司石板选煤发电厂 | Dense medium coarse slime sorting device and sorting process |
| CN109269951A (en) * | 2018-09-06 | 2019-01-25 | 山西智卓电气有限公司 | Floating tail-coal ash content, concentration, coarse granule detection method of content based on image |
| CN111753912A (en) * | 2020-06-28 | 2020-10-09 | 中国矿业大学 | Coal slime flotation clean coal ash content prediction method based on deep learning |
-
2022
- 2022-08-30 CN CN202211045618.0A patent/CN115452824A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104084293A (en) * | 2014-07-29 | 2014-10-08 | 四川达竹煤电（集团）有限责任公司石板选煤发电厂 | Dense medium coarse slime sorting device and sorting process |
| CN109269951A (en) * | 2018-09-06 | 2019-01-25 | 山西智卓电气有限公司 | Floating tail-coal ash content, concentration, coarse granule detection method of content based on image |
| CN111753912A (en) * | 2020-06-28 | 2020-10-09 | 中国矿业大学 | Coal slime flotation clean coal ash content prediction method based on deep learning |
Non-Patent Citations (5)
| Title |
|---|
| 于业达;高鹏飞;赵一舟;潘国庆;陈通;: "基于深度卷积神经网络的蚕蛹雌雄自动识别", 蚕业科学, no. 02, 15 April 2020 (2020-04-15) * |
| 吴庆祥, 严闪, 黄晞, 陈振荣: "可移动机器人在中心对称环境中的自定位算法", 计算机研究与发展, no. 01, 16 January 2004 (2004-01-16) * |
| 武林海;刘洋;: "煤泥浮选中I-ELM的应用", 煤炭技术, no. 03, 10 March 2019 (2019-03-10) * |
| 郭西进;魏凌敖;杨春宝;: "基于深度卷积网络的煤泥浮选尾矿灰分检测方法研究", 煤炭技术, no. 02, 10 February 2020 (2020-02-10) * |
| 陈兵;查宇飞;张胜杰;李运强;张园强;: "基于迁移学习和特征融合的航空器图像分类算法", 火力与指挥控制, no. 07, 15 July 2018 (2018-07-15) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116738120A (en) * | 2023-08-11 | 2023-09-12 | 齐鲁工业大学(山东省科学院) | Copper grade SCN modeling algorithm for X fluorescence grade analyzer |
| CN116738120B (en) * | 2023-08-11 | 2023-11-03 | 齐鲁工业大学(山东省科学院) | Copper grade SCN modeling algorithm for X fluorescence grade analyzer |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ren et al. | State of the art in defect detection based on machine vision | |
| Jia et al. | Detection and segmentation of overlapped fruits based on optimized mask R-CNN application in apple harvesting robot | |
| Lv et al. | Cascade network for detection of coal and gangue in the production context | |
| CN109460753B (en) | Method for detecting floating object on water | |
| Zhao et al. | A visual long-short-term memory based integrated CNN model for fabric defect image classification | |
| CN113469177B (en) | Deep learning-based drainage pipeline defect detection method and system | |
| CN111753912A (en) | Coal slime flotation clean coal ash content prediction method based on deep learning | |
| CN110689011A (en) | Solar cell panel defect detection method of multi-scale combined convolution neural network | |
| CN110852395A (en) | Ore granularity detection method and system based on autonomous learning and deep learning | |
| CN109063728A (en) | A kind of fire image deep learning mode identification method | |
| Li et al. | An image-based hierarchical deep learning framework for coal and gangue detection | |
| CN103344583B (en) | A kind of praseodymium-neodymium (Pr/Nd) component concentration detection system based on machine vision and method | |
| CN109035226B (en) | Mura defect detection method based on LSTM model | |
| CN106874929B (en) | Pearl classification method based on deep learning | |
| CN110599459A (en) | Underground pipe network risk assessment cloud system based on deep learning | |
| CN115452824A (en) | Method for predicting ash content of coal flotation tailings on dynamic overflow surface through multiple models | |
| CN115294033A (en) | Tire belt layer difference level and misalignment defect detection method based on semantic segmentation network | |
| CN112749675A (en) | Potato disease identification method based on convolutional neural network | |
| CN116843650A (en) | SMT welding defect detection method and system integrating AOI detection and deep learning | |
| CN114841453A (en) | Clean coal ash content prediction method in flotation process | |
| CN115205255A (en) | Stone automatic grading method and system based on deep learning | |
| Zhang et al. | Fabric defect detection based on visual saliency map and SVM | |
| CN112561885B (en) | YOLOv 4-tiny-based gate valve opening detection method | |
| Ma et al. | A hierarchical attention detector for bearing surface defect detection | |
| Fu et al. | Quantitative ore texture analysis with convolutional neural networks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |