CN117078641A - Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes - Google Patents

Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes Download PDF

Info

Publication number
CN117078641A
CN117078641A CN202311069200.8A CN202311069200A CN117078641A CN 117078641 A CN117078641 A CN 117078641A CN 202311069200 A CN202311069200 A CN 202311069200A CN 117078641 A CN117078641 A CN 117078641A
Authority
CN
China
Prior art keywords
data
electric smelting
condition
training
smelting magnesium
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
CN202311069200.8A
Other languages
Chinese (zh)
Inventor
李帷韬
黄鑫兴
张心茹
吕顺辰
刘威
李奇越
孙伟
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.)
Hefei University of Technology
Original Assignee
Hefei University of Technology
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 Hefei University of Technology filed Critical Hefei University of Technology
Priority to CN202311069200.8A priority Critical patent/CN117078641A/en
Publication of CN117078641A publication Critical patent/CN117078641A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • 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
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • 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
    • G06N3/047—Probabilistic or stochastic networks
    • 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
    • G06N3/048—Activation functions
    • 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
    • G06N3/091—Active learning
    • 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
    • G06N3/092—Reinforcement learning
    • 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
    • G06V10/77—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
    • G06V10/778—Active pattern-learning, e.g. online learning of image or video features
    • 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
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
    • 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
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/70—Labelling scene content, e.g. deriving syntactic or semantic representations
    • 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
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • 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
    • 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
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30136—Metal
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30232—Surveillance
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/06—Recognition of objects for industrial automation

Abstract

The utility model provides a based on initiative reinforcement learning and multimode electric smelting magnesium stove under fire condition identification method, belong to industry magnesium smelting intelligent control technical field, solve the problem how to design a quick and accurate electric smelting magnesium stove under fire condition judgement method, through the historical data of gathering electric smelting magnesium stove, select most valuable image data sample and mark based on initiative reinforcement learning, based on time sequence construction image-current sample set, build neural network and initialize based on unlabeled multimode data, train multimode study neural network based on labeled multimode data, discern electric smelting magnesium stove under fire condition based on multimode neural network, compare with conventional detection method introducing initiative reinforcement learning algorithm, through selecting most valuable data sample to mark, and then reduce required mark data when the dataset is built, the required cost of labor has been reduced under the circumstances of guaranteeing algorithm identification accuracy, safe production has been ensured.

Description

Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes
Technical Field
The invention belongs to the technical field of intelligent control of industrial magnesium smelting, and relates to an active reinforcement learning and multi-mode electric smelting magnesium furnace underburn working condition identification method.
Background
The fused magnesia has the characteristics of fire resistance, high temperature resistance, corrosion resistance, oxidation resistance and the like, is an important strategic raw material in China, and is widely applied to various fields of aviation, military industry and the like. The preparation of the electric smelting magnesia in China mostly takes magnesite as a raw material, a three-phase alternating current electric smelting magnesia furnace is utilized to heat the magnesite to be more than 2800 ℃ for smelting, and then the magnesium oxide obtained by smelting is subjected to cooling crystallization, and impurities are removed, so that the high-quality electric smelting magnesia is obtained.
The smelting process of the electric smelting magnesium furnace comprises the working conditions of furnace starting, feeding, normal smelting, underburning and the like. The underburn condition is an underburn condition, and the occurrence of the underburn condition is usually caused by incomplete melting of raw materials due to impurities in the raw materials, so that the generated bubbles cause local overhigh temperature in the furnace. If the underburn working condition is not found and treated in time, the quality of the product can be greatly reduced, and serious accidents such as furnace wall burning leakage, molten raw material leakage and the like can be caused, so that the safety of personnel is threatened. Therefore, the timely judgment of the underburn working condition is very important for the preparation of the fused magnesia. In actual production, the underburn working condition is judged by mainly manually inspecting and observing the furnace wall and flame state, but the method relies on the experience of operators, so that misjudgment or missed judgment is easy to occur, and meanwhile, the production field environment is bad, so that personal safety risks exist.
Currently, identification studies of the underburn condition of an electric smelting magnesium furnace can be roughly divided into two types. The method utilizes the change mode of three-phase current to identify working conditions, and a rule reasoning algorithm based on three-phase current values is used for summarizing a set of working condition judgment expert rule base by analyzing historical current statistical characteristics under different working conditions, so that the underburn working conditions are judged according to the current values acquired in real time on site. However, because the current value has a large amount of noise, the working condition judgment is not ideal only by the current characteristics, and the method is only suitable for being used as an auxiliary method. The other type of method mainly utilizes monitoring images produced by the electric smelting magnesium furnace to identify working conditions. The method utilizes visual information contained in flame images of the furnace wall and the furnace mouth to establish a sensing model of the working condition, but the thermal imaging device adopted by the technology has higher cost and is difficult to realize large-scale industrial application. Tests on the two types of methods in the industrial field show that the satisfactory degree is difficult to achieve in real-time performance and accuracy by adopting the current or image under-burning condition identification technology singly. On the one hand, although the change mode of the current can reflect the production working condition to a certain extent, the characteristics are difficult to identify by people, so that the current working condition marks for the current data still need to be indirectly determined through the monitoring video at the corresponding moment. On the other hand, the cost of manually marking the production image of the electric smelting magnesium furnace is high, and the image is difficult to accurately mark in the initial stage of underburn or in the transitional state. Therefore, how to combine the sensitivity and rapidness of the current feature with the accuracy of the image feature so as to accurately warn when the underburn condition is not yet completely molded is a urgent problem to be solved.
The conventional research method has a supervised learning method, and can train a model with better performance based on massive labeled data and by combining images and current characteristics under the same working condition. However, as heavy industrial equipment, the electric smelting magnesium furnace is used for marking data by workers with certain working ages for manufacturing working condition data sets, so that the manufacturing cost is high, most marked data possibly have the problem of serious characteristic coupling, the help to the data sets is not large, and even the over fitting of a model is possibly caused, so that the robustness of the model is reduced. Therefore, how to select the samples with value in the mass data for marking and use the rest of the samples without marking so as to reduce the corresponding labor cost in the model training process, ensure the recognition accuracy and performance of the model and realize cost reduction and efficiency improvement is a problem to be solved.
Disclosure of Invention
The invention aims to solve the technical problem of how to design a rapid and accurate method for judging the underburn condition of an electric smelting magnesium furnace, which can liberate workers from dangerous and high-strength work and realize safe production.
The invention solves the technical problems through the following technical scheme:
the method for identifying the underburn condition of the electric smelting magnesium furnace based on the active reinforcement learning and the multiple modes comprises the following steps:
step 1, collecting historical working condition data of an electric smelting magnesium furnace;
step 2, selecting the most valuable image data sample for marking based on active reinforcement learning, wherein the method comprises the following steps:
step 2.1, inputting unlabeled training data
Step 2.2, randomly selecting some samples to label when training is started, and taking the samples as an initial labeling training setThen use the labeling sample set +.>Performing classification training to obtain an initial model theta c 。
Step 2.3, defining the (i, j) th element of the data calculation state S, S according to formula (1) as:
wherein x is u As an unlabeled training sample,is the corresponding unknown tag, σ (·) is the sigmoid function;
step 2.4, making action=pi (S; θ) based on the Actor network a ) Selecting K unlabeled training samples;
step 2.5 updating the labeled training dataClassifier based on this training
Step 2.6, calculating a state S' and a reward r based on the updated training data;
step 2.7, repeating the steps 2.3, 2.4, 2.5 and 2.6 until the marking cost reaches the budget;
step 3, sampling based on the same time sequence of the image data set and the three-phase current data set, and obtaining a three-phase current data set D C Is selected outThree-phase current data with matched images are used for constructing an image-current data pair data set;
step 4, constructing and initializing a neural network based on unlabeled multi-mode data, wherein the specific method is as follows:
step 4.1, extracting the characteristics of unlabeled image data of the electric smelting magnesium furnace by adopting a CNN convolution layer;
step 4.2, extracting three-phase current data characteristics corresponding to the image data by adopting a self-encoder, and encoding output dimensions of the three-phase current data characteristics into dimensions corresponding to the image characteristics extracted by CNN;
step 4.3, carrying out feature fusion on the current features and the image features based on a transducer multi-mode encoder;
step 4.4, the fused characteristics are transmitted to a neural network for model pre-training, and an initial fused magnesium underburn working condition identification model is obtained;
and step 5, training the multi-modal learning neural network based on the marked multi-modal data, wherein the specific method comprises the following steps of:
step 5.1, carrying out feature extraction and feature fusion on the marked data;
step 5.2, the fused marked data features are transmitted to an initial working condition recognition model to recognize working conditions, and a Softmax layer calculates classification results of the working conditions;
step 5.3, introducing fault labels of corresponding working conditions, and calculating cross entropy loss based on model classification results and real labels;
and 6, processing the operation condition video of the electric smelting magnesium furnace to be identified and the three-phase current at the corresponding moment, and outputting the condition judging result, so as to judge whether the electric smelting magnesium furnace at the current moment is in an abnormal state according to the judging result, and if so, carrying out condition abnormality alarm prompt.
Further, the collecting the historical operating condition data of the electric smelting magnesium furnace in the step 1 includes: video data and three-phase current data in the running process of the electric smelting magnesium furnace, and constructing a data set D= (D) based on the video data and the three-phase current data I ；D C ) Wherein D is I ＝(I 1 ，I 2 ，…I k ，…，I n )，D C ＝(C 1 ，C 2 ，…，C k ，…，C n )，D I For image dataset extracted based on video data, D C For constructing a three-phase current data set, n is the total time of the data set, C k For D C The kth element, I k For D I K=1, 2,3 … … n.
Further, the calculation formula of the Softmax layer in step 5.2 is as follows:
wherein f i For inputting the corresponding feature vector of the Softmax layer.
Further, the calculation formula of the cross entropy loss in step 5.3 is:
wherein N is the total number of samples, y represents the real label of the samples, and p represents the probability that the network prediction samples belong to the class, namely the output of each sample through the Softmax layer.
The invention has the advantages that:
according to the method, the historical data of the electric magnesium melting furnace is collected, the most valuable image data samples are selected for marking based on active reinforcement learning, an image-current sample set is constructed based on time sequence, a neural network is constructed and initialized based on unlabeled multi-mode data, the multi-mode learning neural network is trained based on the labeled multi-mode data, the underburn condition of the electric magnesium melting furnace is identified based on the multi-mode neural network, compared with the conventional detection method, the active reinforcement learning algorithm is introduced, the most valuable data samples are selected for marking, the labeled data required during the construction of the data set are further reduced, the required labor cost is reduced under the condition that the identification accuracy of the algorithm is guaranteed, and the safe production is guaranteed.
Drawings
FIG. 1 is a flow chart of the method for identifying the underburn condition of the electric smelting magnesium furnace based on active reinforcement learning and multiple modes.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The technical scheme of the invention is further described below with reference to the attached drawings and specific embodiments:
example 1
As shown in fig. 1, the method for identifying the underburn condition of the electric smelting magnesium furnace based on active reinforcement learning and multiple modes in the embodiment of the invention comprises the following steps:
step 1, collecting historical working condition data of an electric smelting magnesium furnace, comprising the following steps: video data and three-phase current data in the running process of the electric smelting magnesium furnace, and a data set D= (D) is constructed based on the video data and the three-phase current data I ，D C ) Wherein D is I ＝(I 1 ，I 2 ，…I k ，…，I n )，D C ＝(C 1 ，C 2 ，…，C k ，…，C n )，D I For image dataset extracted based on video data, D C For constructing a three-phase current data set, n is the total time of the data set, C k For D C The kth element, I k For D I K=1, 2,3 … … n; since the image dataset is sampled at the same time as the three-phase current dataset, I k And C k Is the image-current sample pair at the same time.
Step 2, selecting the most valuable image data sample based on active reinforcement learning (Active Reinforcement Learning, ARL) and marking the image data sample by a human expert, wherein each basic concept of the active reinforcement learning is explained as follows:
state (State): in order to select samples that are advantageous for improving classification performance, the prediction results of the samples to be selected by the current classifier need to be considered. Based on this, the state isDesigned as a matrix comprising unlabeled training samples X U Is>Is the number of unlabeled training samples and C is the number of categories. Mathematically, the (i, j) th element of S is defined as:
wherein x is u As an unlabeled training sample,is the corresponding unknown label, σ (·) is the sigmoid function.
Action (Action): will be theta a Defined as the parameters of the Actor network. Since the goal of the Actor network is to select samples from unlabeled training data sets for labeling, we define action as a vectorWhere each element corresponds to an unlabeled training sample. A sigmoid function is used as an activation function for each element, and a value between 0 and 1 is obtained. Learning strategy pi (S; theta) a ) Action a is generated from state S. After the motion vector is obtained, all candidate samples except the already selected samples are arranged in descending order, and the first K samples with the highest values are selected for annotation. The label provided by the selected sample and the marker is marked +.>The training data of the enhanced annotation is recorded asWhere i represents the current training round number.
State Transition (State Transition): after the selected samples are annotated and added to the labeled training data, the enhanced training data set (X L ，Y L ) Updating the classifier. Thereafter, a new state matrix S' may be obtained according to equation (1) using a new classifier. The minimization equation is:
wherein y is c As a real tag it is possible to provide a real tag,is the predicted tag.
Rewards (Reward): in order to improve the performance of the classifier, ARL is expected to have the Actor network pay more attention to those samples that are highly likely to be misclassified by the classifier. To achieve this, the ARL designs a novel bonus function by taking into account the use of predictions obtained from the classifier and the real labels given by the expert. Specifically, for selected samplesDefinition k ij True tag given for labeling expert for category j, +.>As a predictive label for category j for this sample obtained from the classifier. For a classification network theta c The definition of rewards (Reward) is:
if a sample isAnd calculating the difference value of the real label and the predicted probability value for each category in turn, and summing the difference values of all the categories. If the difference value of the final summation is smaller, the predicted probability value is closer to the real label. Thus, in expression (3), the higher the prize r, the greater the gap between the selected sample predictions and the true labels, meaning that these samples with erroneous predictions should be more focused by the classifier. The Actor network will therefore be encouraged to select those samples of the current model that are poorly classified during the learning process.
Based on the above definition, the training procedure is as follows:
step 2.1, inputting unlabeled training data
Step 2.2, randomly selecting some samples to label when training is started, and taking the samples as an initial labeling training setThen use the labeling sample set +.>Performing classification training to obtain an initial model theta c 。
Step 2.3, calculating the state S of the data according to the formula (1);
step 2.4, making action=pi (S; θ) based on the Actor network a ) Selecting K unlabeled training samples;
step 2.5 updating the labeled training dataClassifier based on this training
Step 2.6, calculating a state S' and a reward r based on the updated training data;
step 2.7, repeating the steps 2.3, 2.4, 2.5 and 2.6 until the marking cost reaches the budget.
Step 3, based on a time sequence alignment method, a three-phase current data set D is obtained C Three-phase current data matched with the image are selected, and an image-current data pair data set is constructed.
And 4, constructing a neural network based on unlabeled multi-mode data and initializing. Aiming at the multi-mode data of the electric smelting magnesium furnace, the multi-mode training method designed by the method comprises the following modules:
CNN convolutional layer, AE self-encoding layer, transducer multi-mode encoder, softmax classification layer:
wherein the CNN convolution layer comprises A convolution blocks, which are respectively marked as Conv 1 K、Conv a K、Conv A K, wherein Conv A K represents a level A convolution block;
the AE coding layer includes B self-encoders respectively named AutoEncoder 1 K、AutoEncoder b K、AutoEncoder B K, wherein AutoEncoder B K represents a B-th level self-encoder;
the transducer multimode Encoder includes C encoders, respectively denoted as Encoders 1 K、Encoder c K、Encoder C K, wherein Encoder C K represents the C-th encoder. Each encoder includes a Self attribute layer, a Cross attribute layer and a forward propagation layer.
Step 4.1, extracting the characteristics of unlabeled image data of the electric smelting magnesium furnace by adopting a CNN convolution layer;
step 4.2, extracting three-phase current data characteristics corresponding to the image data by adopting a self-encoder, and encoding output dimensions of the three-phase current data characteristics into dimensions corresponding to the image characteristics extracted by CNN;
step 4.3, carrying out feature fusion on the current features and the image features based on a transducer multi-mode encoder;
and 4.4, delivering the fused characteristics to a neural network for model pre-training to obtain an initial fused magnesium underburn condition identification model.
And step 5, training the multi-modal learning neural network based on the marked multi-modal data. The specific process is as follows:
step 5.1, carrying out feature extraction and feature fusion on the marked data based on the steps 4.1, 4.2 and 4.3;
and 5.2, the fused marked data features are transmitted to an initial working condition recognition model for working condition recognition, and a classification result of the working condition is calculated by a Softmax layer, wherein the calculation formula of Softmax is as follows:
wherein f i For inputting the corresponding feature vector of the Softmax layer.
Step 5.3, introducing fault labels of corresponding working conditions, and calculating the cross entropy loss based on the model classification result and the real labels, wherein a calculation formula of the cross entropy loss is as follows:
where N is the total number of samples, y represents the true label of the samples, and p represents the probability that the network predicted samples belong to the class, i.e. the output of each sample through the Softmax layer. And 5, continuously iterating and repeating training, and finally obtaining the neural network model capable of accurately identifying the underfiring condition of the electric smelting magnesium furnace.
Step 6, identifying the underburn working condition of the electric smelting magnesium furnace based on the multi-modal neural network: processing the operation condition video of the electric smelting magnesium furnace to be identified and the three-phase current at the corresponding moment through the steps 4.1, 4.2 and 4.3, outputting the condition judging result through the model obtained in the step 5, judging whether the electric smelting magnesium furnace at the current moment is in an abnormal state according to the judging result, and if so, carrying out condition abnormality alarm prompt.
The above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (4)

1. The method for identifying the underburn condition of the electric smelting magnesium furnace based on the active reinforcement learning and the multiple modes is characterized by comprising the following steps:
step 1, collecting historical working condition data of an electric smelting magnesium furnace;
step 2, selecting the most valuable image data sample for marking based on active reinforcement learning, wherein the method comprises the following steps:
step 2.1, inputting unlabeled training data
Step 2.2, randomly selecting some samples to label when training is started, and taking the samples as an initial labeling training setThen use the labeling sample set +.>Performing classification training to obtain an initial model theta c ；
Step 2.3, defining the (i, j) th element of the data calculation state S, S according to formula (1) as:
wherein x is u As an unlabeled training sample,is the corresponding unknown tag, σ (·) is the sigmoid function;
step 2.4, making action=pi (S; θ) based on the Actor network a ) Selecting K unlabeled training samples;
step 2.5 updating the labeled training dataTraining classifier based on this->
Step 2.6, calculating a state S' and a reward r based on the updated training data;
step 2.7, repeating the steps 2.3, 2.4, 2.5 and 2.6 until the marking cost reaches the budget;
step 3, sampling based on the same time sequence of the image data set and the three-phase current data set, and obtaining a three-phase current data set D C Three-phase current data matched with the image are selected, and an image-current data pair data set is constructed;
step 4, constructing and initializing a neural network based on unlabeled multi-mode data, wherein the specific method is as follows:
step 4.1, extracting the characteristics of unlabeled image data of the electric smelting magnesium furnace by adopting a CNN convolution layer;
step 4.2, extracting three-phase current data characteristics corresponding to the image data by adopting a self-encoder, and encoding output dimensions of the three-phase current data characteristics into dimensions corresponding to the image characteristics extracted by CNN;
step 4.3, carrying out feature fusion on the current features and the image features based on a transducer multi-mode encoder;
step 4.4, the fused characteristics are transmitted to a neural network for model pre-training, and an initial fused magnesium underburn working condition identification model is obtained;
and step 5, training the multi-modal learning neural network based on the marked multi-modal data, wherein the specific method comprises the following steps of:
step 5.1, carrying out feature extraction and feature fusion on the marked data;
step 5.2, the fused marked data features are transmitted to an initial working condition recognition model to recognize working conditions, and a Softmax layer calculates classification results of the working conditions;
step 5.3, introducing fault labels of corresponding working conditions, and calculating cross entropy loss based on model classification results and real labels;
and 6, processing the operation condition video of the electric smelting magnesium furnace to be identified and the three-phase current at the corresponding moment, and outputting the condition judging result, so as to judge whether the electric smelting magnesium furnace at the current moment is in an abnormal state according to the judging result, and if so, carrying out condition abnormality alarm prompt.
2. The method for identifying the underburn condition of the electric smelting magnesium furnace based on the active reinforcement learning and the multiple modes according to claim 1, wherein the step 1 of collecting the historical condition data of the electric smelting magnesium furnace comprises the following steps: video data and three-phase current data in the running process of the electric smelting magnesium furnace, and constructing a data set D= (D) based on the video data and the three-phase current data I ，D C ) Wherein D is I ＝(I 1 ，I 2 ，…，I k ，…，I n )，D C ＝(C 1 ，C 2 ，…，C k ，…，C n )，D I For image dataset extracted based on video data, D C For constructing a three-phase current data set, n is the total time of the data set, C k For D C The kth element, I k For D I K=1, 2,3 … … n.
3. The method for identifying the underburn condition of the electric smelting magnesium furnace based on the active reinforcement learning and the multiple modes according to claim 2, wherein the calculation formula of the Softmax layer in the step 5.2 is as follows:
wherein f i For inputting the corresponding feature vector of the Softmax layer.
4. The method for identifying the underburn condition of the electric smelting magnesium furnace based on the active reinforcement learning and the multiple modes according to claim 3, wherein the calculation formula of the cross entropy loss in the step 5.3 is as follows:
wherein N is the total number of samples, y represents the real label of the samples, and p represents the probability that the network prediction samples belong to the class, namely the output of each sample through the Softmax layer.
CN202311069200.8A 2023-08-23 2023-08-23 Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes Pending CN117078641A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311069200.8A CN117078641A (en) 2023-08-23 2023-08-23 Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311069200.8A CN117078641A (en) 2023-08-23 2023-08-23 Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes

Publications (1)

Publication Number Publication Date
CN117078641A true CN117078641A (en) 2023-11-17

Family

ID=88716721

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311069200.8A Pending CN117078641A (en) 2023-08-23 2023-08-23 Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes

Country Status (1)

Country Link
CN (1) CN117078641A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117496133A (en) * 2024-01-03 2024-02-02 山东工商学院 Closed bus R-CNN temperature fault monitoring method based on multi-mode data

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117496133A (en) * 2024-01-03 2024-02-02 山东工商学院 Closed bus R-CNN temperature fault monitoring method based on multi-mode data
CN117496133B (en) * 2024-01-03 2024-03-22 山东工商学院 Closed bus R-CNN temperature fault monitoring method based on multi-mode data

Similar Documents

Publication Publication Date Title
CN116625438B (en) Gas pipe network safety on-line monitoring system and method thereof
CN112001527B (en) Industrial production process target data prediction method of multi-feature fusion depth neural network
CN105956618B (en) Converter steelmaking blowing state identification system and method based on image dynamic and static characteristics
CN104777830B (en) A kind of multiple operating modes process monitoring method based on KPCA mixed models
CN117078641A (en) Method for identifying underburn condition of electric smelting magnesium furnace based on active reinforcement learning and multiple modes
Duan et al. Product technical life prediction based on multi-modes and fractional Lévy stable motion
EP2776815A1 (en) System and method for improving automated defect detection user interface
CN104793606B (en) Industrial method for diagnosing faults based on improved KPCA and HMM
CN107861481B (en) Fault diagnosis method based on semi-supervised new-class mining of industrial big data graph
CN111126206B (en) Smelting state detection system and method based on deep learning
Liu et al. Model fusion and multiscale feature learning for fault diagnosis of industrial processes
CN106599450A (en) Priori knowledge-based method for monitoring fault of magnesite electric melting furnace by kernel flexible manifold embedding
Zhou et al. Application of time series data anomaly detection based on deep learning in continuous casting process
CN111222798B (en) Complex industrial process key index soft measurement method
CN117422935B (en) Motorcycle fault non-contact diagnosis method and system
Liu et al. Disturbance robust abnormality diagnosis of fused magnesium furnaces using deep neural networks
WO2022058116A1 (en) Explaining machine learning output in industrial applications
Rayhana et al. Automated Defect-Detection System for Water Pipelines Based on CCTV Inspection Videos of Autonomous Robotic Platforms
Chang et al. Multimode process mode identification with coexistence of quantitative information and qualitative information
CN114842383B (en) DVRL-VST-based electric smelting magnesium furnace working condition video identification method
CN108182302B (en) Early fault detection method based on modified clustering semi-supervised kernel principal component analysis
Jiang et al. A new monitoring method for the blocking time of the taphole of blast furnace using molten iron flow images
Zhang et al. DuAK: Reinforcement Learning-Based Knowledge Graph Reasoning for Steel Surface Defect Detection
Li et al. A Deep Learning–Based Diagnosis Model Driven by Tuyere Images Big Data for Iron‐Making Blast Furnaces
Zhang et al. PredNet based sequence image disturbance processing of fused magnesium furnaces

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