CN116597350A - Flotation process fault early warning method based on BiLSTM predictive deviation - Google Patents

Flotation process fault early warning method based on BiLSTM predictive deviation Download PDF

Info

Publication number
CN116597350A
CN116597350A CN202310546858.7A CN202310546858A CN116597350A CN 116597350 A CN116597350 A CN 116597350A CN 202310546858 A CN202310546858 A CN 202310546858A CN 116597350 A CN116597350 A CN 116597350A
Authority
CN
China
Prior art keywords
model
data
early warning
fault
deviation
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
Application number
CN202310546858.7A
Other languages
Chinese (zh)
Inventor
廖一鹏
严欣
朱坤华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fuzhou University
Original Assignee
Fuzhou University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Fuzhou University filed Critical Fuzhou University
Priority to CN202310546858.7A priority Critical patent/CN116597350A/en
Publication of CN116597350A publication Critical patent/CN116597350A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/52—Scale-space analysis, e.g. wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/56—Extraction of image or video features relating to colour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/62—Extraction of image or video features relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application provides a flotation process fault early warning method based on BiLSTM prediction deviation, which comprises the following steps: firstly, acquiring foam video images, extracting color characteristics of a video sequence and constructing a sample data set; then, a time sequence prediction model based on the ResNet50 and the BiLSTM is constructed, and the space features extracted by the ResNet50 are input into the BiLSTM network for color time sequence prediction; secondly, predicting color characteristics of the time sequence by adopting a normal working data set, calculating the deviation degree according to the predicted value and the actual value, and determining an early warning threshold value; and finally, acquiring a foam video image in real time to predict color characteristics, and early warning production faults when the predicted deviation exceeds the limit. The method has the advantages of small prediction deviation and good fitting degree of the time sequence color characteristics, effectively advances the fault early warning time, can realize early warning of faults in the flotation process, and effectively reduces the waste of mineral resources and medicaments.

Description

Flotation process fault early warning method based on BiLSTM predictive deviation
Technical Field
The application relates to the technical field of froth flotation, in particular to a flotation process fault early warning method based on BiLSTM prediction deviation.
Background
Froth flotation is an important mineral separation technology, mineral resources and reagents are wasted due to faults in the flotation process, and an effective fault detection technology can help a flotation plant reduce reagent consumption and labor intensity, improve mineral recovery rate, promote optimal control of the flotation process and save production cost. Fault detection and diagnosis of flotation processes is often challenging, typically done by experienced operators by frequent visual inspection of the froth appearance, but such manual operations are always accompanied by serious response delays and measurement errors, and are time-consuming and laborious.
To overcome this problem, machine vision indicates flotation performance by extracting visual characteristics of the cell surface froth and transmitting the extracted characteristics to an operator or as input to a process control system to adjust the flotation process. Many machine vision-based flotation process fault detection methods are proposed successively, such as: describing a foam gray level image by wavelet transformation, and calculating a space gray level co-occurrence matrix to obtain texture features of the foam image, classifying the foam image by the features to realize fault state identification of a flotation process, wherein the fault detection precision is greatly influenced by the identification precision; the PDF with the size of the bubbles after being approximately segmented is established through a non-parameter kernel estimator, a dynamic weight model of the PDF with the size of the bubbles is established, stability analysis is carried out based on the weight model, a threshold criterion determined by stability conditions is obtained, fault detection is carried out, but the shapes of the bubbles are different and are unevenly distributed, the same segmentation algorithm cannot be applied to all conditions, and therefore the fault detection precision is low; after wavelet transformation and reconstruction are carried out on the foam gray level image, the white area of the binary image is calculated to design equivalent bubble size characteristics, the range of a normal foam image is determined to carry out flotation fault detection, but the wavelet transformation has limited direction selectivity and only can capture limited direction information, false information and serious appearance degradation are introduced when the image is processed, and the fault detection precision is influenced.
The effective fault detection technology can help the flotation plant reduce the consumption of medicaments, reduce the labor intensity and improve the recovery rate of minerals. Traditionally, flotation process fault detection is mostly done by experienced personnel by observing the appearance of the froth, which manual operation is accompanied by serious measurement errors and response delays and is time-consuming and laborious. Subsequent flotation process fault detection methods based on foam visual characteristics are sequentially proposed, but such methods are online real-time detection of flotation process faults, and when faults occur, the processing time left for operators is urgent.
Disclosure of Invention
Therefore, the application aims to provide the early-stage early-warning method for the faults in the flotation process based on the BiLSTM prediction deviation, so that the early-stage early warning of the faults in the flotation process is realized, and the waste of mineral resources and medicaments is effectively reduced.
In order to achieve the above purpose, the application adopts the following technical scheme: a flotation process fault early warning method based on BiLSTM prediction deviation degree comprises the following steps:
step 1, acquiring a historical flotation process data set, wherein the historical flotation process data set comprises a normal state data set and a fault state data set; preprocessing the data set such as data cleaning and selecting a model predicted amount;
the normal state data set therein is utilized with 6:2:2, dividing the proportion into a training set, a verification set and a test set, taking training set data as the input of a model, extracting 2 measurement point parameters related to faults, namely an image color a channel mean value and an a channel standard deviation of a training set video frame, taking time sequence data of the 2 measurement point parameters as the prediction output value of the model, training the fault prediction model, and testing the training effect of the model through the verification set;
step 3: firstly, saving a fault prediction model trained in the step 2, taking a normal test set as input data of the model, verifying generalization capability and prediction accuracy of the model, then taking Root Mean Square Error (RMSE), mean Absolute Error (MAE) and fitting coefficient R2 as indexes for evaluating a prediction effect, and finally calculating deviation degree and early warning threshold value under a normal flotation process state by combining color measurement point parameter values extracted from a corresponding foam video frame to obtain a fault early warning model;
step 4: and testing the fault early-warning model by the fault test set, inputting the fault test set into the fault early-warning model to obtain a predicted value, calculating the deviation according to the predicted value, early-warning according to whether the deviation exceeds an early-warning threshold value, and verifying the effectiveness of the early-warning model.
Step 5: and acquiring foam video images in real time, inputting the foam video images into a fault early warning model, obtaining a predicted value, calculating deviation according to the predicted value, and if the deviation exceeds an early warning threshold value, considering that the flotation process has a fault trend, and giving an early warning signal.
In a preferred embodiment, the data preprocessing is specifically preprocessing the collected flotation froth image dataset; the method specifically comprises the following steps:
(1) Data are cleaned by a Grabbs criterion method, and the specific steps of the criterion method are as follows:
step S1: let x be i (i=1, 2,3,., N) is a flotation process observation data sample, and an observation data model with μ as an observation object is established as follows:
x i ＝μ+p i ,p i N(0,σ 2 )
step S2: calculating a sample mean and variance of xi according to the following formula respectively;
step S3: calculating the construction statistic G such that the statistic G follows a transformation;
step S4: calculating statistics G according to α When G > G α When the sample is judged to be abnormal, the sample is directly removed;
step S5: step S1 to step S4 are circularly executed until the cleaning of the sample data set is completed;
(2) Normalizing the data, and compressing the data to be within a range of [0,1] by the following formula before model training;
x std ＝(x-x min )/(x max -x min )。
in a preferred embodiment, the construction of the fault early warning model of the flotation process is specifically as follows:
extracting characteristics of an input flotation foam video frame by adopting a ResNet50 network, inputting video frame data subjected to data preprocessing into the ResNet50 network, extracting spatial characteristics of data at each time point, and then transmitting the spatial characteristics to a BiLSTM network for time sequence prediction;
after the ResNet50 network performs feature extraction on input data, accumulating for a period of time to form a feature value sequence, wherein the waveform formed by time fluctuation reflects the change of the state of the flotation process; inputting the data characteristics learned from the ResNet50 network into the BiLSTM network, performing time sequence coding on the data characteristics by the data characteristics, acquiring time sequence characteristic vectors, and then sending the time sequence characteristic vectors into a full connection layer to complete the prediction of time sequences;
constructing a flotation process fault prediction model based on ResNet50 and BiLSTM; the fault prediction model comprises a training model and a test model, and the two partial models mainly comprise three parts, namely data input, a depth network and prediction output; in the data input part, firstly, preprocessing a flotation froth visible light video frame through data, and then taking the flotation froth visible light video frame as the input of a prediction model, wherein a training set and a verification set are taken as the input of the training model, and a normal test set is taken as the input of a test model; the depth network comprises a ResNet50 and a BiLSTM network, the ResNet50 is used for extracting the spatial characteristics of input data at each time point, then transmitting the spatial characteristics to the correlation among BiLSTM network learning data, carrying out time sequence coding on the data characteristics, and extracting time sequence characteristics from the forward direction and the reverse direction; in the prediction output part, the extracted time sequence features are sent into a full-connection layer, a predicted value is output through the full-connection layer, wherein in a training model, a foam video image is converted to CIElab space to extract 2 measuring points in the aspect of the mean value and the variance of an a channel, time sequence data of 2 measuring point parameters are used as the predicted output value of the model, and the fault prediction model is trained, so that the fault prediction model is obtained; in the test model, the full-connection layer outputs the predicted value of the color measurement point parameter at the next moment to complete the prediction of the time sequence, and then the normal deviation degree and the early warning threshold value are calculated by combining the color measurement point parameter values extracted from the corresponding foam video frames to further obtain the fault early warning model.
In a preferred embodiment, the deviation definition and early warning strategy is specifically:
the constructed ResNet50-BiLSTM flotation process fault prediction model is trained by a normal flotation process data set, and after the model receives new time series data, the data of the next moment can be predicted according to a learning result; when the flotation process has a fault trend, the related monitoring variable has a certain deviation from the normal state data, and when the deviation exceeds a set safety threshold value, early faults of the flotation process are judged;
calculating residual error r between predicted value and actual value output by ResNet50-BiLSTM network model of this section ij ；
Wherein: r is (r) ij Is the residual of variable i at time j; y is ij Andthe actual value of the variable i at the moment j and the predicted value output by the model of the section are respectively; according to the selection of the model pre-measurement, 4 measurement point parameters are provided, so that residual data at each moment form a 4-dimensional vector, and the deviation degree of the flotation process at the moment from a normal state is calculated according to the following formula;
calculating the deviation degree of each moment so as to form a deviation degree sequence;
the deviation degree sequence has a plurality of extreme points and non-stationarity, the generalized extreme value theory is adopted to calculate the early warning threshold, and the specific steps of solving are as follows:
decomposing the deviation sequence into a plurality of minimum intervals with the same data points, and selecting 5 points as the minimum intervals; the method of calculating the maximum value M in each section is as follows:
M＝max{x 1 ,...,x n }
wherein: x is x i Is a value within a minimum interval; { x 1 ,...,x n -random sequences distributed and independent;
assuming that the distribution function of the random sequence is F, the sum { x } of the distribution of M 1 ,...,x n The relationship between the distribution functions F is:
P r {M≤z}＝P r {x 1 ≤z,……,x n ≤z}＝P r {x 1 ≤z}×…×P r {x n ≤z}＝{F(z)} n
since the distribution function F is unknown, it is assumed that μ and σ satisfy:
P r {(M-μ)/σ}→G(z)
wherein: μ is a position parameter; sigma is a scale parameter; g () is a generalized extremum distribution function, calculated according to the following equation:
G(z)＝exp{-[1+((z-μ)/σ)] -1/ξ }
wherein: the generalized extremum distribution function is defined by the set { z:1+ζ (z- μ)/σ > 0}, where the parameters μ and ζ satisfy respectively: - +_μ < +_σ0, - +_ζ < +_infinity;
obtaining a position parameter mu, a scale parameter sigma and a shape parameter xi of generalized extremum distribution by a maximum likelihood estimation method, and calculating an early warning threshold by the following formula;
T h ＝μ-σ[1-{-ln(1-α)} -ξ ]/ξ
in summary, in the flotation process, if the deviation degree sequence is maintained within the early warning threshold, the flotation process is judged to be normal, and if the deviation degree exceeds the early warning threshold, the flotation process is judged to have a fault trend, so that fault early warning is realized.
In a preferred embodiment, biLSTM builds both forward and reverse LSTM networks to extract the feature information; the same input sequence is respectively connected into a forward LSTM network and a backward LSTM network, and then the internal structures of the two LSTM networks are changed;
the operation process of BiLSTM is as follows:
the forward propagation update formula is as follows:
the backward propagation update formula is as follows:
the formula of the output after the superposition of the forward and backward network layers is as follows:
wherein: t represents a time series;hidden layer vectors at time t are represented, and arrows represent directions; x is x t And y t Respectively representing the input and the output at the time t; w (W) xh 、W hh And W is hy Weight matrices representing input-hidden layer, hidden layer-hidden layer, and hidden layer-output layer, respectively; b h And b y Offset vectors respectively representing the hidden layer and the output layer; h represents a hidden layer activation function.
In a preferred embodiment, the foam image is converted to CIELab color space, the digital features on the a-channel that characterize the red level are extracted, and the 2 statistics for that channel are calculated according to the following equation, namely, the mean μ and standard deviation σ, resulting in a total of 2-dimensional statistics for each frame of image: a channel mean and a channel variance, which form color feature vectors of the corresponding images;
wherein: p is p ij Representing pixel values of the image at (i, j); m and N represent the width and height of the image.
Compared with the prior art, the application has the following beneficial effects: the application provides a flotation process fault early warning method based on BiLSTM prediction deviation. And constructing a time sequence prediction model based on the ResNet50 and the BiLSTM, inputting the spatial features extracted by the ResNet50 into the BiLSTM network for color time sequence prediction, calculating the deviation degree according to the predicted value and the actual value, determining an early warning threshold value, and carrying out early warning on production faults when the predicted deviation degree exceeds the limit. The method has the advantages of small prediction deviation and good fitting degree of the time sequence color characteristics, effectively advances the fault early warning time, can realize early warning of faults in the flotation process, strives more time for operators to adjust, and effectively reduces the waste of mineral resources and medicaments.
Drawings
FIG. 1 is a block diagram of a BiLSTM network in accordance with a preferred embodiment of the present application;
FIG. 2 is a diagram of a predictive model framework in accordance with a preferred embodiment of the application;
FIG. 3 is a flow chart of a fault detection process for flotation in accordance with a preferred embodiment of the present application;
FIG. 4 is a diagram showing the prediction results of the ResNet50-BiLSTM model on the test set according to the early-stage fault early-warning effect of the preferred embodiment of the present application;
fig. 5 is a diagram of early-stage fault early-warning effect according to a preferred embodiment of the present application, (a) is a CNN-BiLSTM model early-warning effect, (b) is a res net50-LSTM model early-warning effect, and (c) is a res net50-BiLSTM model early-warning effect according to the present application.
Detailed Description
The application will be further described with reference to the accompanying drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the application. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present application; as used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
The application relates to a flotation process fault early warning method based on BiLSTM prediction deviation, which comprises the following steps: firstly, acquiring foam video images, extracting color characteristics of a video sequence and constructing a sample data set; then, a time sequence prediction model based on the ResNet50 and the BiLSTM is constructed, and the space features extracted by the ResNet50 are input into the BiLSTM network for color time sequence prediction; secondly, predicting color characteristics of the time sequence by adopting a normal working data set, calculating the deviation degree according to the predicted value and the actual value, and determining an early warning threshold value; and finally, acquiring a foam video image in real time to predict color characteristics, and early warning production faults when the predicted deviation exceeds the limit. The method has the advantages of small prediction deviation and good fitting degree of the time sequence color characteristics, effectively advances the fault early warning time, can realize early warning of faults in the flotation process, and effectively reduces the waste of mineral resources and medicaments.
The detailed technical scheme is as follows:
BiLSTM network model
The BiLSTM network overcomes the limitation of the RNN network and the LSTM network, and FIG. 1 shows the structure of the BiLSTM network, and from the figure, the BiLSTM builds two forward and reverse LSTM networks to extract characteristic information, so that the time sequence characteristics of data are more fully learned. The main idea is to access the same input sequence to the forward and backward LSTM networks respectively, then connect the hidden layers of the two networks together, and access the hidden layers to the output layer together for prediction. The BiLSTM does not change the internal structure of the LSTM network, and the LSTM is only used for data modeling from the forward direction and the reverse direction, and then the information is spliced together, but the network also solves the problem that the accuracy is reduced because the importance of the front and rear data information is changed due to the model structure of the LSTM on the basis of retaining the advantages of the LSTM.
As can be seen from fig. 1, the operation of the BiLSTM is as follows:
the forward propagation update formula is:
the backward propagation update formula is:
the formula of the output after the superposition of the forward and backward network layers is as follows:
wherein: t represents a time series;hidden layer vectors at time t are represented, and arrows represent directions; x is x t And y t Respectively representing the input and the output at the time t; w (W) xh 、W hh And W is hy Weight matrices representing input-hidden layer, hidden layer-hidden layer, and hidden layer-output layer, respectively; b h And b y Offset vectors respectively representing the hidden layer and the output layer; h represents a hidden layer activation function.
2. Flotation froth color feature extraction
Before the prediction model is built, model prediction is selected, and researchers find that the color of the surface of the flotation froth is closely related to the flotation production process index because the flotation production process is affected by a plurality of physical factors and chemical factors. The surface color of the flotation froth is a direct indicator of the mineral separation production index, and experienced operators of the flotation plant judge the production condition by observing the surface color of the froth and timely adjust the production process. Therefore, the fault early warning model constructed by the application carries out early warning on faults from the change of the color characteristics of the surface of the flotation froth.
The red degree in the foam image can more intuitively reflect the flotation production working condition. Because the CIELab color space can better separate the color information of the image and better accords with the visual perception characteristic of human beings, the foam image is converted into the CIELab color space, the digital characteristics on the channel a which can represent the red degree are extracted, and 2 statistics of the channel are calculated according to formulas (4) and (5), namely the mean value mu and the standard deviation sigma. Thus, a total of 2-dimensional statistics are obtained for each frame of image: an a-channel mean and an a-channel variance, which constitute color feature vectors of the corresponding images.
Wherein: pij represents the pixel value of the image at (i, j); m and N represent the width and height of the image.
3. Flotation process fault early warning based on BiLSTM and predictive deviation
3.1 data Pre-processing
In the flotation process, because the environment of a mineral flotation site is poor, the light of the surface of a flotation tank is weak, dust and fog are disturbed, so that phenomena such as shadow and blurring appear in a foam image shot by a site shooting system, and the problems that personnel recording deviation and production equipment are unstable sometimes appear, so that an abnormal sample can be acquired in the flotation site and cannot be directly used for inputting a prediction model, and the acquired flotation foam image dataset is preprocessed.
(1) Data cleansing
In order to avoid the interference of irrelevant data in the flotation process, the data are cleaned by a Grabbs criterion method according to the working condition of a flotation site, and the specific steps of the criterion method are as follows:
step 1: let x be i (i=1, 2,3,., N) is a flotation process observation data sample, and an observation data model with μ as an observation object is established as follows:
x i ＝μ+p i ,p i N(0,σ 2 ) (6)
step 2: calculating x according to the formula (7) and the formula (8), respectively i Is a sample mean and variance of (c).
Step 3: the statistic G is constructed according to the formula (9) such that the statistic G follows the transformation shown in the formula (10).
Step 4: calculating statistics G according to equation (11) α When G > G α And when the sample is judged to be abnormal, the sample is directly removed.
Step 5: steps 1 to 4 are cyclically performed until the sample dataset is washed.
(2) Data normalization
Before model training, the data is normalized by the formula (12), and the data is compressed to be within the range of [0,1] to eliminate the influence of different dimensions. Normalization can not only accelerate the calculation speed and the convergence speed of the model, but also improve the model precision to a certain extent.
x std ＝(x-x min )/(x max -x min ) (12)
3.2 construction of fault early warning model in flotation process
(1) ResNet50 extraction of spatial features
The deep learning can accurately extract hidden features from input data through a training model, and the ResNet50 network shows a good effect on a flotation froth image, so that the ResNet50 network is adopted to extract features of an input flotation froth video frame, video frame data subjected to data preprocessing is input into the ResNet50 network, spatial features of data at each time point are extracted, and then the spatial features are transmitted to the BiLSTM network for time sequence prediction.
(2) BiLSTM extraction of time series features
After the ResNet50 network performs feature extraction on the input data, a feature value sequence is formed after a period of accumulation, and the waveform formed by time fluctuation can reflect the change of the flotation process state. However, the ResNet50 network can only recognize the magnitude of local variations and cannot detect the sequence information of the input data. Therefore, the data features learned from the ResNet50 network can be input into the BiLSTM network, the data features are subjected to time sequence coding by the BiLSTM network, the time sequence feature vectors are obtained, and then the time sequence feature vectors are sent into the full connection layer to complete the prediction of the time sequence.
(3) Predictive model
In order to realize early fault early warning and accurate prediction of a future time sequence, the application constructs a flotation process fault prediction model based on ResNet50 and BiLSTM. The structure of the fault prediction model is shown in fig. 2, and the model comprises a training model and a test model, wherein the two models mainly comprise three parts of data input, a depth network and prediction output. In the data input part, firstly, preprocessing a flotation froth visible light video frame through data, and then taking the flotation froth visible light video frame as the input of a prediction model, wherein a training set and a verification set are taken as the input of the training model, and a normal test set is taken as the input of a test model; the deep network comprises a ResNet50 and a BiLSTM network, the ResNet50 is used for extracting the spatial characteristics of input data at each time point, then transmitting the spatial characteristics to the correlation among BiLSTM network learning data, carrying out time sequence coding on the data characteristics, and extracting time sequence characteristics from the forward direction and the reverse direction; in the prediction output part, the extracted time sequence features are sent into a full-connection layer, a predicted value is output through the full-connection layer, wherein in a training model, a foam video image is converted to CIElab space to extract 2 measuring points in the aspect of the mean value and the variance of an a channel, time sequence data of 2 measuring point parameters are used as the predicted output value of the model, and the fault prediction model is trained, so that the fault prediction model is obtained; in the test model, the full-connection layer outputs the predicted value of the color measurement point parameter at the next moment to complete the prediction of the time sequence, and then the normal deviation degree and the early warning threshold value are calculated by combining the color measurement point parameter values extracted from the corresponding foam video frames to further obtain the fault early warning model.
3.3 deviance definition and Pre-alarm strategy
The constructed ResNet50-BiLSTM flotation process fault prediction model is trained by a normal flotation process data set, so that the relation among variables in the normal flotation process state can be learned, and after the model receives new time series data, the next moment of data can be predicted according to the learned result. When the flotation process has a fault trend, a certain deviation of the related monitoring variable from the normal state data can occur, and when the deviation exceeds a set safety threshold value, early faults of the flotation process are judged.
(1) Deviation definition
The error between the deep learning network predicted value and the actual value is referred to as a residual, which may reflect the change in the target variable. Calculating residual error r between predicted value and actual value output by the ResNet50-BiLSTM network model of this section through formula (13) ij 。
Wherein: rij is the residual of variable i at time j; yij (y j)The actual value of the variable i at the moment j and the predicted value output by the model of the current section are respectively. According to the selection of the model pre-measurement, the section has 4 measurement point parameters, so the residual data of each momentA 4-dimensional vector is formed, and the deviation degree of the flotation process from the normal state at the moment is calculated according to the formula (14).
The degree of deviation at each moment is calculated to form a sequence of degrees of deviation.
(2) Early warning strategy
The deviation sequence has a plurality of extreme points and non-stationarity, and in order to further improve the expression capacity of the model, an early warning threshold value needs to be set for the model. The application adopts generalized extremum theory to calculate the early warning threshold value, and the specific steps of solving are as follows:
the deviation sequence is decomposed into a plurality of minimum intervals with the same data points, and 5 points are selected as the minimum intervals. The method of calculating the maximum value M in each section is as follows:
M＝max{x 1 ,...,x n } (15)
wherein: x is x i Is a value within a minimum interval; { x 1 ,...,x n And is a random sequence that is uniformly distributed and independent.
Let the distribution function of the random sequence in 1) be F, then the sum { x } of the distribution of M 1 ,...,x n The relationship between the distribution functions F is:
P r {M≤z}＝P r {x 1 ≤z,……,x n ≤z}＝P r {x 1 ≤z}×…×P r {x n ≤z}＝{F(z)} n (16)
since the distribution function F is unknown, it is assumed that μ and σ satisfy:
P r {(M-μ)/σ}→G(z) (17)
wherein: μ is a position parameter; sigma is a scale parameter; g () is a generalized extremum distribution function, calculated according to the following equation:
G(z)＝exp{-[1+((z-μ)/σ)] -1/ξ } (18)
wherein: the generalized extremum distribution function is defined by the set { z:1+ζ (z- μ)/σ > 0}, where the parameters μ and ζ satisfy respectively: - +_μ < +_σ0, - +_ζ < +_infinity.
4) And (3) obtaining a position parameter mu, a scale parameter sigma and a shape parameter xi of the generalized extremum distribution by a maximum likelihood estimation method, and calculating an early warning threshold by a formula (19).
T h ＝μ-σ[1-{-ln(1-α)} -ξ ]/ξ(19)
In summary, in the flotation process, if the deviation degree sequence is maintained within the early warning threshold, the flotation process is judged to be normal, and if the deviation degree exceeds the early warning threshold, the flotation process is judged to have a fault tendency, so that fault early warning is realized.
3.4 concrete implementation flow and step
In order to realize early fault identification in the flotation process, firstly, an input data set is preprocessed, and then the preprocessed data set is input into a ResNet50 network to extract the spatial characteristics of data of each time point; then, the extracted spatial features are transmitted to a BiLSTM network to extract time sequence features, the time sequence features are transmitted to a full-connection layer, and a predicted value of the next moment is output through the full-connection layer; and finally, calculating the deviation degree through the obtained predicted value and the actual value, determining an alarm threshold value, and realizing early fault early warning in the flotation process. The flow of the fault early warning method in the flotation production process is shown in figure 3, and the specific implementation steps are as follows:
step 1, acquiring a historical flotation process data set, wherein the historical flotation process data set comprises a normal state data set and a fault state data set; preprocessing the data set such as data cleaning and selecting a model predicted amount;
the normal state data set therein is utilized with 6:2:2, dividing the proportion into a training set, a verification set and a test set, taking training set data as the input of a model, extracting 2 measurement point parameters related to faults, namely an image color a channel mean value and an a channel standard deviation of a training set video frame, taking time sequence data of the 2 measurement point parameters as the prediction output value of the model, training the fault prediction model, and testing the training effect of the model through the verification set;
step 3: firstly, saving a fault prediction model trained in the step 2, taking a normal test set as input data of the model, verifying generalization capability and prediction accuracy of the model, then taking Root Mean Square Error (RMSE), mean Absolute Error (MAE) and fitting coefficient R2 as indexes for evaluating a prediction effect, and finally calculating deviation degree and early warning threshold value under a normal flotation process state by combining color measurement point parameter values extracted from a corresponding foam video frame to obtain a fault early warning model;
step 4: and testing the fault early-warning model by the fault test set, inputting the fault test set into the fault early-warning model to obtain a predicted value, calculating the deviation according to the predicted value, early-warning according to whether the deviation exceeds an early-warning threshold value, and verifying the effectiveness of the early-warning model.
Step 5: and acquiring foam video images in real time, inputting the foam video images into a fault early warning model, obtaining a predicted value, calculating deviation according to the predicted value, and if the deviation exceeds an early warning threshold value, considering that the flotation process has a fault trend, and giving an early warning signal.
4 specific examples and illustrations
In order to verify the effectiveness of the method, foam images collected in a lead zinc ore floatation plant of the Ministry of mining industry, fujian, are taken as experimental samples, the hardware platforms of the experiments are Intel (R) Core (TM) i7-9800X CPU@3.80GHz, NVIDIA GeForce RTX 3080Ti and 128GB RAM, and the software running environments are Windows 10Matlab 2019a, python3.7 and Pytorch1.7, and the method provided by the application is verified through the experiments.
And selecting operation data of 12 hours in a normal flotation process state of a certain day of a flotation plant as experimental data, wherein the data sampling interval is 3s. After the ResNet50-BiLSTM model is built, the training set and the verification set training model are used for carrying out the test on the model by taking the training set and the verification set training model into the test set, and then the prediction result of the ResNet50-BiLSTM fault prediction model on the test set is shown in figure 4.
To further verify the predictive performance of the proposed ResNet50-BiLSTM model, the predictive effect of this model was compared with that of the CNN-BiLSTM, resNet50-LSTM model. Selection of RMSE, MAE and R 2 As evaluation indexes, taking the average value of each parameter as an evaluation result, and calculating each evaluation index as followsFormulas (20) - (22) show, wherein the RMSE reflects the accuracy of the prediction result, and lower RMSE indicates higher prediction accuracy; the MAE reflects the consistency of the predicted result, and the lower the MAE is, the smaller the error of the predicted result is; r is R 2 Reflecting the fitting degree of the prediction curve, R 2 The closer to 1 indicates the better fitting effect. The performance statistics of each model are shown in table 1. Compared with other two models, the CNN-BiLSTM model has larger deviation of the predicted result compared with the actual result, lower prediction accuracy and poorer fitting degree; compared with the prediction result of the CNN-BiLSTM model, the prediction result of the ResNet50-LSTM model is improved in various indexes, but a certain lifting space is provided; compared with the ResNet50-BiLSTM model, the ResNet50-BiLSTM model provided by the application has the advantages that the root mean square error is reduced by 20.6%, the average absolute error is reduced by 23.5%, the fitting coefficient is improved by 17.1%, and the ResNet50-BiLSTM model combines the advantages of the two models, so that the characteristics of the multi-element time sequence data are better integrated, the prediction accuracy is higher, the deviation is small, and the fitting degree is good.
Wherein: y is ij 、And->The actual value, the predicted value and the average value of the j-th moment variable i are respectively; j=1, 2, …, N; n is the length of the time series of data sets.
Table 1 performance statistics of predictive models
And then performing early warning test of faults: because the change rule of 2 measuring point variables in the flotation process is stable, whether the flotation process has faults or not cannot be effectively reflected, the ResNet50-BiLSTM model is utilized to predict the sequence of 2 measuring point variables for a period of time in the future, then the prediction deviation degree is calculated, and the early warning threshold value is determined. The fault data collected in a certain lead-zinc ore flotation plant are used as experimental data, CNN-BiLSTM, resNet50-LSTM and ResNet50-BiLSTM models are respectively adopted for fault early warning test, the early warning effect is shown in figure 5, the known flotation process generates faults at the time point of 2.8 hours, the CNN-BiLSTM early warning model fluctuates greatly, the deviation degree has no obvious rising trend before the faults occur, the deviation degree still has the condition of being lower than the threshold value a few days before the faults, the early warning threshold value of the model is 2.8687, the deviation degree is higher than the threshold value at the time point of 154 minutes, and the alarm is sent out 14 minutes in advance; the deviation degree of the rest two early warning models has a remarkable rising trend before faults occur, the early warning threshold value of the ResNet50-LSTM model is 2.3217, the deviation degree of the time point running to 140 minutes is higher than the threshold value, the alarm is sent out 28 minutes earlier, the early warning threshold value of the ResNet50-BiLSTM model is 1.9471, the deviation degree of the time point running to 106 minutes is higher than the threshold value, the alarm is sent out 62 minutes earlier, the deviation degree is above the threshold value after the alarm is sent out, and the rising trend of the deviation degree fluctuation is remarkable along with the deepening of the faults.

Claims (6)

1. The early-stage fault early-warning method for the flotation process based on BiLSTM prediction deviation degree is characterized by comprising the following steps of:
step 1, acquiring a historical flotation process data set, wherein the historical flotation process data set comprises a normal state data set and a fault state data set; preprocessing the data set such as data cleaning and selecting a model predicted amount;
the normal state data set therein is utilized with 6:2:2, dividing the proportion into a training set, a verification set and a test set, taking training set data as the input of a model, extracting 2 measurement point parameters related to faults, namely an image color a channel mean value and an a channel standard deviation of a training set video frame, taking time sequence data of the 2 measurement point parameters as the prediction output value of the model, training the fault prediction model, and testing the training effect of the model through the verification set;
step 3: firstly, saving a fault prediction model trained in the step 2, taking a normal test set as input data of the model, verifying generalization capability and prediction accuracy of the model, then taking Root Mean Square Error (RMSE), mean Absolute Error (MAE) and fitting coefficient R2 as indexes for evaluating a prediction effect, and finally calculating deviation degree and early warning threshold value under a normal flotation process state by combining color measurement point parameter values extracted from a corresponding foam video frame to obtain a fault early warning model;
step 4: testing a fault early warning model through a fault test set, inputting the fault test set into the fault early warning model to obtain a predicted value, calculating a deviation degree according to the predicted value, early warning according to whether the deviation degree exceeds an early warning threshold value, and verifying the effectiveness of the early warning model;
step 5: and acquiring foam video images in real time, inputting the foam video images into a fault early warning model, obtaining a predicted value, calculating deviation according to the predicted value, and if the deviation exceeds an early warning threshold value, considering that the flotation process has a fault trend, and giving an early warning signal.
2. The early warning method for flotation process faults based on BiLSTM predictive deviation as claimed in claim 1, wherein the data preprocessing is specifically preprocessing the collected flotation froth image data set; the method specifically comprises the following steps:
(1) Data are cleaned by a Grabbs criterion method, and the specific steps of the criterion method are as follows:
step S1: let x be i (i=1, 2,3,., N) is a flotation process observation data sample, and an observation data model with μ as an observation object is established as follows:
x i ＝μ+p i ,p i N(0,σ 2 )
step S2: respectively calculate x according to the following i Is the mean and variance of the samples;
step S3: calculating the construction statistic G such that the statistic G follows a transformation;
step S4: calculating statistics G according to α When G > G α When the sample is judged to be abnormal, the sample is directly removed;
step S5: step S1 to step S4 are circularly executed until the cleaning of the sample data set is completed;
(2) Normalizing the data, and compressing the data to be within a range of [0,1] by the following formula before model training;
x std ＝(x-x min )/(x max -x min )。
3. the early-stage fault early-warning method for the flotation process based on BiLSTM predictive deviation according to claim 1, wherein the construction of the early-warning model for the flotation process is specifically as follows:
extracting characteristics of an input flotation foam video frame by adopting a ResNet50 network, inputting video frame data subjected to data preprocessing into the ResNet50 network, extracting spatial characteristics of data at each time point, and then transmitting the spatial characteristics to a BiLSTM network for time sequence prediction;
after the ResNet50 network performs feature extraction on input data, accumulating for a period of time to form a feature value sequence, wherein the waveform formed by time fluctuation reflects the change of the state of the flotation process; inputting the data characteristics learned from the ResNet50 network into the BiLSTM network, performing time sequence coding on the data characteristics by the data characteristics, acquiring time sequence characteristic vectors, and then sending the time sequence characteristic vectors into a full connection layer to complete the prediction of time sequences;
constructing a flotation process fault prediction model based on ResNet50 and BiLSTM; the fault prediction model comprises a training model and a test model, and the two partial models mainly comprise three parts, namely data input, a depth network and prediction output; in the data input part, firstly, preprocessing a flotation froth visible light video frame through data, and then taking the flotation froth visible light video frame as the input of a prediction model, wherein a training set and a verification set are taken as the input of the training model, and a normal test set is taken as the input of a test model; the depth network comprises a ResNet50 and a BiLSTM network, the ResNet50 is used for extracting the spatial characteristics of input data at each time point, then transmitting the spatial characteristics to the correlation among BiLSTM network learning data, carrying out time sequence coding on the data characteristics, and extracting time sequence characteristics from the forward direction and the reverse direction; in the prediction output part, the extracted time sequence features are sent into a full-connection layer, a predicted value is output through the full-connection layer, wherein in a training model, a foam video image is converted to CIElab space to extract 2 measuring points in the aspect of the mean value and the variance of an a channel, time sequence data of 2 measuring point parameters are used as the predicted output value of the model, and the fault prediction model is trained, so that the fault prediction model is obtained; in the test model, the full-connection layer outputs the predicted value of the color measurement point parameter at the next moment to complete the prediction of the time sequence, and then the normal deviation degree and the early warning threshold value are calculated by combining the color measurement point parameter values extracted from the corresponding foam video frames to further obtain the fault early warning model.
4. The early-stage fault early-warning method for a flotation process based on BiLSTM predictive deviation as set forth in claim 1, wherein the deviation definition and early-warning strategy is specifically as follows:
the constructed ResNet50-BiLSTM flotation process fault prediction model is trained by a normal flotation process data set, and after the model receives new time series data, the data of the next moment can be predicted according to a learning result; when the flotation process has a fault trend, the related monitoring variable has a certain deviation from the normal state data, and when the deviation exceeds a set safety threshold value, early faults of the flotation process are judged;
calculating residual error r between predicted value and actual value output by ResNet50-BiLSTM network model of this section ij ；
Wherein: r is (r) ij Is the residual of variable i at time j; y is ij Andthe actual value of the variable i at the moment j and the predicted value output by the model of the section are respectively; according to the selection of the model pre-measurement, 4 measurement point parameters are provided, so that residual data at each moment form a 4-dimensional vector, and the deviation degree of the flotation process at the moment from a normal state is calculated according to the following formula;
calculating the deviation degree of each moment so as to form a deviation degree sequence;
the deviation degree sequence has a plurality of extreme points and non-stationarity, the generalized extreme value theory is adopted to calculate the early warning threshold, and the specific steps of solving are as follows:
decomposing the deviation sequence into a plurality of minimum intervals with the same data points, and selecting 5 points as the minimum intervals; the method of calculating the maximum value M in each section is as follows:
M＝max{x 1 ,...,x n }
wherein: x is x i Is a value within a minimum interval; { x 1 ,...,x n -random sequences distributed and independent;
assuming that the distribution function of the random sequence is F, the sum { x } of the distribution of M 1 ,...,x n The relationship between the distribution functions F is:
P r {M≤z}＝P r {x 1 ≤z,……,x n ≤z}＝P r {x 1 ≤z}×…×P r {x n ≤z}＝{F(z)} n
since the distribution function F is unknown, it is assumed that μ and σ satisfy:
P r {(M-μ)/σ}→G(z)
wherein: μ is a position parameter; sigma is a scale parameter; g () is a generalized extremum distribution function, calculated according to the following equation:
G(z)＝exp{-[1+((z-μ)/σ)] -1/ξ }
wherein: the generalized extremum distribution function is defined by the set { z:1+ζ (z- μ)/σ > 0}, where the parameters μ and ζ satisfy respectively: - +_μ < +_σ0, - +_ζ < +_infinity;
obtaining a position parameter mu, a scale parameter sigma and a shape parameter xi of generalized extremum distribution by a maximum likelihood estimation method, and calculating an early warning threshold by the following formula;
T h ＝μ-σ[1-{-ln(1-α)} -ξ ]/ξ
in summary, in the flotation process, if the deviation degree sequence is maintained within the early warning threshold, the flotation process is judged to be normal, and if the deviation degree exceeds the early warning threshold, the flotation process is judged to have a fault tendency, so that fault early warning is realized.
5. The early-stage fault early-warning method for a flotation process based on BiLSTM predictive deviation according to claim 1, wherein BiLSTM constructs a forward LSTM network and a reverse LSTM network to extract characteristic information; the same input sequence is respectively connected into a forward LSTM network and a backward LSTM network, and then the internal structures of the two LSTM networks are changed;
the operation process of BiLSTM is as follows:
the forward propagation update formula is as follows:
the backward propagation update formula is as follows:
the formula of the output after the superposition of the forward and backward network layers is as follows:
wherein: t represents a time series;hidden layer vectors at time t are represented, and arrows represent directions; x is x t And y t Respectively representing the input and the output at the time t; w (W) xh 、W hh And->Weight matrices representing input-hidden layer, hidden layer-hidden layer, and hidden layer-output layer, respectively; b h And b y Offset vectors respectively representing the hidden layer and the output layer; h represents a hidden layer activation function.
6. The early-stage fault early-warning method for a flotation process based on a BiLSTM predictive deviation as claimed in claim 3, wherein the foam image is converted into a CIELab color space, digital features on an a channel capable of representing the red degree are extracted, 2 statistics of the channel are calculated according to the following formula, namely a mean value mu and a standard deviation sigma, and each frame of image obtains 2-dimensional statistics in total: a channel mean and a channel variance, which form color feature vectors of the corresponding images;
wherein: p is p ij Representing pixel values of the image at (i, j); m and N represent the width and height of the image.
CN202310546858.7A 2023-05-16 2023-05-16 Flotation process fault early warning method based on BiLSTM predictive deviation Pending CN116597350A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310546858.7A CN116597350A (en) 2023-05-16 2023-05-16 Flotation process fault early warning method based on BiLSTM predictive deviation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310546858.7A CN116597350A (en) 2023-05-16 2023-05-16 Flotation process fault early warning method based on BiLSTM predictive deviation

Publications (1)

Publication Number Publication Date
CN116597350A true CN116597350A (en) 2023-08-15

Family

ID=87593140

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310546858.7A Pending CN116597350A (en) 2023-05-16 2023-05-16 Flotation process fault early warning method based on BiLSTM predictive deviation

Country Status (1)

Country Link
CN (1) CN116597350A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117078232A (en) * 2023-10-17 2023-11-17 山东沪金精工科技股份有限公司 Processing equipment fault prevention system and method based on big data
CN117792933A (en) * 2024-02-27 2024-03-29 南京市微驰数字科技有限公司 Network flow optimization method and system based on deep learning

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117078232A (en) * 2023-10-17 2023-11-17 山东沪金精工科技股份有限公司 Processing equipment fault prevention system and method based on big data
CN117078232B (en) * 2023-10-17 2024-01-09 山东沪金精工科技股份有限公司 Processing equipment fault prevention system and method based on big data
CN117792933A (en) * 2024-02-27 2024-03-29 南京市微驰数字科技有限公司 Network flow optimization method and system based on deep learning
CN117792933B (en) * 2024-02-27 2024-05-03 南京市微驰数字科技有限公司 Network flow optimization method and system based on deep learning

Similar Documents

Publication Publication Date Title
CN110310261B (en) Contact net dropper defect detection model training method and defect detection method
CN116597350A (en) Flotation process fault early warning method based on BiLSTM predictive deviation
CN110598736A (en) Power equipment infrared image fault positioning, identifying and predicting method
CN113240199B (en) Port ship track prediction method based on DILATE _ TLSTM
CN112669305A (en) Metal surface rust resistance test bench and rust resistance evaluation method
CN116757097A (en) Digital twin hydraulic engineering operation and maintenance monitoring system and method
CN108711148A (en) A kind of wheel tyre defect intelligent detecting method based on deep learning
CN111141653B (en) Tunnel leakage rate prediction method based on neural network
CN113065702B (en) Landslide displacement multi-linear prediction method based on ST-SEEP segmentation method and space-time ARMA model
CN112116198A (en) Data-driven process industrial state perception network key node screening method
CN116486305B (en) Image recognition-based deep sea suspended particulate matter concentration prediction method
CN110335294A (en) Mine water pump house leakage detection method based on frame difference method Yu 3D convolutional neural networks
CN113901927B (en) Underwater object shape recognition method based on flow field pressure time course
CN111783616B (en) Nondestructive testing method based on data-driven self-learning
CN111753776B (en) Structural damage identification method based on echo state and multi-scale convolution combined model
CN115451803B (en) Water gate real-time monitoring and early warning method based on fusion of GNSS and measuring robot
CN117521512A (en) Bearing residual service life prediction method based on multi-scale Bayesian convolution transducer model
CN107437112B (en) A kind of mixing RVM model prediction methods based on the multiple dimensioned kernel function of improvement
CN112365093A (en) GRU deep learning-based multi-feature factor red tide prediction model
CN113468823B (en) Optical module damage detection method and system based on machine learning
CN112069621B (en) Method for predicting residual service life of rolling bearing based on linear reliability index
CN111855946B (en) Water quality analysis method based on fish behavior analysis
CN114581729A (en) High-resolution remote sensing impervious surface extraction method based on weak supervised learning
CN113378365A (en) Method, system and device for optimizing variable point detection
CN112884205A (en) Electromechanical equipment fault prediction method

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