CN111898669B - Abnormal event early warning system of direct-current submerged arc furnace based on machine learning - Google Patents
Abnormal event early warning system of direct-current submerged arc furnace based on machine learning Download PDFInfo
- Publication number
- CN111898669B CN111898669B CN202010725828.9A CN202010725828A CN111898669B CN 111898669 B CN111898669 B CN 111898669B CN 202010725828 A CN202010725828 A CN 202010725828A CN 111898669 B CN111898669 B CN 111898669B
- Authority
- CN
- China
- Prior art keywords
- data
- module
- machine learning
- direct
- current
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Strategic Management (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Fuzzy Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Vertical, Hearth, Or Arc Furnaces (AREA)
Abstract
The invention discloses a machine learning-based direct-current submerged arc furnace abnormal event early warning system which comprises a sensor module, a data acquisition module, a central processing module, a man-machine interaction module, a machine learning module, a numerical control output module and an execution mechanism module. The system is based on a machine learning technology which is at the front of the current technology and is mature day by day, and is mainly used for predicting and early warning abnormal events such as material collapse, electrode soft and hard breakage and the like which often occur in the production process of the direct-current submerged arc furnace.
Description
Technical Field
The invention relates to the technical field of submerged arc furnace control, in particular to a direct-current submerged arc furnace abnormal event early warning system based on machine learning.
Background
At present, the industrial ferroalloy direct current smelting technology with positive and negative electrodes arranged vertically to a hearth is less in industrial application, some related smelting technologies are still in the fumbling process, but the furnace type has the initial effect on the existing rare production line, has obvious energy-saving and environment-friendly advantages, has the advantages of high power factor, small short-net reactance, no skin effect and the like compared with an alternating current electric furnace, and has good powder adaptability; compared with a direct current furnace with bottom electrode technology, the direct current furnace has no bottom electrode consumption, so that regular bottom electrode overhaul work is not needed, and the continuity of smelting production can be ensured.
However, in the production smelting process of the vertical positive and negative electrode direct current electric furnace, the heating value of the positive electrode anode region is larger than that of the negative electrode cathode region according to the electron bombardment principle, so that the roasting characteristics of the vertical two electrodes are different, and the accidents of soft and hard breaking of the electrodes often occur. Meanwhile, the smelting of the submerged arc furnace carries out complex chemical reduction reaction in the hearth of the main body, the process is hidden and difficult to measure, larger ambiguity, randomness and uncertainty exist, if the operation is improper, the accident of material collapse and electrode soft and hard breakage often occurs, and the two abnormal events are the most frequent accidents of the direct current submerged arc furnace, and the normal production of the submerged arc furnace is seriously influenced. Once an accident occurs, a light person consumes a great deal of manpower, financial resources and time; serious persons can damage some equipment of the submerged arc furnace body, even generate high-temperature smoke or hot solution to spray out of the furnace, so that operators in a dangerous area burn and scald, furnace body explosion and the like are caused, and the consequences are serious.
The machine learning intelligence is put into use in industrial production, data are used as strategic resources, the centralized management of equipment related data is completed by means of the support of a big data processing system, an industrial data world is established through data representation and processing, an algorithm model is used for processing analysis, the core value of the industry is mined, the optimal production experience is settled, production targets are used as guiding, production process parameters are adjusted, and the production flow is controlled reversely, so that the preset production index is finally achieved.
The ore-smelting furnace has high equipment price, large maintenance engineering quantity and huge economic loss after shutdown and shutdown. At present, two abnormal events of material collapse and electrode hardness breakage which frequently occur in the production process of the submerged arc furnace are difficult to predict, corresponding treatment is generally carried out after an accident occurs, smooth production is difficult to ensure, the occurrence of the two abnormal events is avoided as much as possible only by means of artificial production experience, for example, a conservative capacity reduction mode is adopted, the two abnormal events still occur, the occurrence cannot be avoided at all, meanwhile, the understanding and the operation methods of the technological process of the direct current submerged arc furnace by different technicians are different, and the possibility of occurrence of the two abnormal events is increased. At present, there are control modes adopting fixed threshold values of electrode current to try to solve the problem, and the effect is improved compared with a simple manual experience, but the effect of the mode can not meet the production requirement due to the influence of complex furnace conditions, raw material components, environmental factors and the like in the production process, and abnormal events still occur at time, so that the problem is not solved fundamentally. With the evolution of science and technology, equipment maintenance means are gradually developed, and from passive maintenance to intelligent predictive maintenance, production loss and equipment maintenance cost caused by unplanned shutdown are reduced. The original manual maintenance and the fixed threshold maintenance depend on experience and cannot meet the increasingly complex working condition requirements, so that development of a direct-current submerged arc furnace abnormal event early warning system and method based on machine learning is urgently needed, abnormal information can be perceived before equipment fails by means of failure prediction combining big data analysis technology and mechanism analysis, and loss caused by unplanned shutdown is reduced.
Disclosure of Invention
According to the problems existing in the prior art, the invention discloses a machine learning-based abnormal event early warning system of a direct-current submerged arc furnace, which is a sensor module for performing sensing detection on a plurality of physical data parameters of the direct-current submerged arc furnace in the production process and converting sensed information into an electric signal according to a certain rule;
the data acquisition module is used for receiving the electric signals transmitted by the sensor module according to a certain period, and performing data processing on the received electric signals by adopting an anti-shake, filtering and passivating method and outputting the processed electric signals;
the central processing module is used for receiving the data information transmitted by the data acquisition module, and performing logic control on the received signals and adjusting a control strategy of a production process;
the man-machine interaction module is used for receiving the logic control information transmitted by the central processing module, displaying the received data information in real time, transmitting a control instruction of a process operator to the central processing module for logic control, and realizing bidirectional data communication with the central processing module;
the machine learning module is used for receiving logic control data transmitted by the central processing module, analyzing and extracting characteristics of massive historical data measuring points produced by smelting in a direct-current furnace, estimating output characteristics of current electrode current in real time, acquiring electrode current output characteristic residual errors according to actual output characteristics of the current electrode current and the electrode current output characteristics estimated by a direct-current submerged arc furnace model, setting a proper threshold in the electrode current output characteristic residual errors as a judgment basis for abnormal events of material collapse and electrode soft and hard breakage, and outputting an early warning signal when the characteristic residual errors are larger than the set threshold; the machine learning module analyzes and models massive historical data measuring points produced by direct current furnace smelting by means of the intelligent equipment fault prediction engine, specifically establishes prediction algorithms of two abnormal events of material collapse and electrode soft and hard breakage in direct current furnace smelting, and uses historical data to verify the algorithms in a reverse direction so as to verify the prediction algorithms of material collapse and electrode soft and hard breakage accidents, and performs early warning analysis on the abnormalities; the machine learning module is used for carrying out big data processing, feature extraction, analysis learning and business application on the received logic control data and feeding back learning result information to the central processing module;
the numerical control output module is used for receiving the control instruction sent by the central processing module and converting the information into various physical signals to be output;
and the executing mechanism module is used for receiving the physical signals transmitted by the numerical control output module, and executing actions corresponding to the physical signals are implemented on various devices in the production process of the direct-current submerged arc furnace.
Further, the machine learning module comprises a feature extraction module, an analysis learning module and a business application module;
the feature extraction module adopts an unsupervised machine learning method to analyze the principal components of the data: firstly, receiving original data of a direct-current submerged arc furnace, carrying out zero-valued treatment on the data, solving a covariance matrix of sampling characteristics, carrying out diagonalization treatment on the covariance matrix, constructing a diagonal matrix with new characteristic covariance of 0, and thus calculating a characteristic set matrix consisting of new standard orthogonal basis vectors, and retaining new characteristics with high contribution rate of main components;
the analysis learning module adopts a supervised machine learning method to train and test the direct-current submerged arc furnace model: firstly, receiving direct-current submerged arc furnace characteristic data transmitted by a characteristic extraction module, carrying out normalization processing on the characteristic data, training an XGBoost model by adopting a training data set, and carrying out parameter tuning on the XGBoost model by adopting a test data set so as to finish characteristic classification;
and the service application module receives the abnormal data transmitted by the analysis learning module to perform state monitoring and fault diagnosis on the direct-current submerged arc furnace.
And the service application module receives the abnormal data transmitted by the analysis learning module to perform state monitoring and fault diagnosis on the direct-current submerged arc furnace.
Furthermore, the machine learning module is based on mass multidimensional data collected and accumulated in the production process of the direct-current submerged arc furnace, historical data of each measuring point and accident record data are obtained from an existing system, an intelligent analysis engine is built in a server, the large data are subjected to machine learning and deep mining by utilizing elastic data storage, calculation processing capacity and artificial intelligence technology of a large data server, data exploration and feature modeling are performed, and model direct-current furnace algorithm results are generated, wherein the direct-current furnace algorithm results comprise an accident prediction model and an accident early warning model, the accident prediction model comprises a material collapse prediction model and an electrode soft and hard breakage prediction model, and the accident early warning model comprises a material collapse early warning module and an electrode soft and hard breakage early warning model.
Further, the machine learning module constructs measuring point data of the direct-current submerged arc furnace in a normal smelting state into a memory matrix, row vectors of the matrix represent the operation data of all measuring points at a certain moment, column vectors represent the operation data of the measuring points at different moments, and when the deviation between real-time output and expected output based on current input is larger than a set threshold value, abnormal state early warning is sent out.
Furthermore, the machine learning module adopts a mode of combining unsupervised learning and supervised learning according to the working principle that the input and output state information has a certain mapping relation in the healthy running state of the direct current furnace, and queries an abnormal mode which deviates from the normal state through a big data simulation method and queries abnormal data characteristics.
Further, when the machine learning module uses the historical data to perform reverse verification on the algorithm, the historical data is obtained in the following manner: historical data segments with fixed time length before the occurrence time of a typical abnormal event are intercepted for a plurality of times and used as direct current furnace input data.
Due to the adoption of the technical scheme, the abnormal event early warning system for the direct-current submerged arc furnace based on machine learning is provided, and because the positive electrode and the negative electrode in the direct-current submerged arc furnace are arranged in pairs, each pair of electrodes is basically not mutually restricted. Therefore, an abnormal event early warning system and method are developed for each positive electrode pair, early warning of abnormal events such as material collapse and electrode soft and hard breakage is achieved, the abnormal events are communicated to a main control system to adjust a control strategy, the whole intelligent level of the direct-current submerged arc furnace is improved, powerful guarantee is provided for continuous production, and huge economic losses caused by material collapse and electrode soft and hard breakage accidents to users can be avoided or reduced.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic diagram of the system of the present invention;
FIG. 2 is a schematic diagram of a machine learning pre-warning method of the present invention;
FIG. 3 is a schematic diagram of the operation of the feature extraction module of the present invention;
FIG. 4 is a schematic diagram of the operation of the analysis learning module of the present invention;
fig. 5 is a schematic diagram of a big data machine learning process of the direct current submerged arc furnace in the present invention.
Detailed Description
In order to make the technical scheme and advantages of the present invention more clear, the technical scheme in the embodiment of the present invention is clearly and completely described below with reference to the accompanying drawings in the embodiment of the present invention:
the system is based on a machine learning technology which is at the front of the current technology and is mature day by day, and mainly aims at predicting and early warning abnormal events such as material collapse, electrode soft and hard breakage and the like which often occur in the production process of the direct-current submerged arc furnace, and the basic principle of the system is as shown in the figure 1: various accidents of the submerged arc furnace in the smelting production process are generally accompanied by the change of a plurality of parameters, main change parameters of various accidents and other parameter combinations with higher relativity with the parameters are found, other parameter combinations can be established as input parameters, the main change parameters are taken as input and output models of output parameters, and the accurate control of the work of the direct-current submerged arc furnace is realized by controlling the models, so that the problem of early warning of the accident of material collapse and electrode soft and hard breakage in the prior art is solved, and huge economic losses caused to users by the occurrence of the accident of material collapse and electrode soft and hard breakage are avoided or reduced. The system specifically comprises a sensor module M1, a data acquisition module M2, a central processing module M3, a man-machine interaction module M4, a machine learning module M5, a numerical control output module M6 and an execution mechanism module M7. The machine learning module M5 further comprises three sub-modules, namely a feature extraction module M5-1, an analysis learning module M5-2 and a business application module M5-3, wherein each module adopts the following working mode:
sensor module M1: the sensing detection is carried out on various physical data parameters in the production process, and the sensed information is converted into an electric signal according to a certain rule and is output, so that the electric signal can be utilized by the data acquisition module M2.
Data acquisition module M2: the output signals of the sensor module M1 are collected according to a certain period, the collected signals are subjected to data processing by using anti-shake, filtering, passivation and other methods, the availability of the data is improved, and then the collected data are output to the central processing module M3.
The central processing module M3: the signals collected from the data acquisition module M2 are subjected to logic control processing, then the processed real-time information is transmitted to the man-machine interaction module M4 for process operators to check, and control instructions sent by the man-machine interaction module M4 are received to enter logic control of the central processing module M3, so that bidirectional data communication with the man-machine interaction module M4 is completed, and real man-machine friendly interaction is achieved. The central processing module M3 sends a large amount of data to the machine learning module M5, then the machine learning module M5 performs large data processing, feature extraction, analysis learning and business application, and then the machine learning module M5 returns learning result information to the central processing module M3; the central processing module M3 processes the received prediction information and adjusts the production process control strategy, a control signal is transmitted to the actuating mechanism module M7 through the numerical control output module M6, the final actual action of the field device is completed, then the production field process data parameters are changed along with the actual action, the production field process data parameters are perceived by the sensor module M1, and the whole production system is ensured to run reliably, normally and optimally.
Man-machine interaction module M4: friendly communication with process operators is realized, information received from the central processing module M3 is displayed in real time, and control instructions of the process operators are sent to the central processing module M3.
Machine learning module M5: and acquiring a large amount of data in the central processing module M3, sequentially performing the functions of big data processing, feature extraction, analysis and learning and business application of the early warning system, and reversely transmitting the predicted result information back to the central processing module M3 after the single machine learning task is completed.
Numerical control output module M6: and receiving control instructions sent by the central processing module M3 in real time, converting the information into various physical signals and transmitting the physical signals to the execution mechanism module M7.
The actuator module M7: and receiving various physical signals sent by the numerical control output module M6, and completing various equipment actions in the production process of the direct-current submerged arc furnace.
Further, as shown in fig. 2, the machine learning module M5 has the following pre-warning method: using data read in the central processing module M3 during production of the direct current submerged arc furnace such as: raw material proportion, electrode electrical parameters, furnace bottom temperature, furnace pressure, furnace gas components, electrode vibration and other parameters, an early warning model selects electrode current as output characteristics, a training set and a testing set are split, a material collapse and electrode soft and hard breakage early warning model is established according to steps in an algorithm principle, and then the model is used for prediction.
Wherein the history data 201 is obtained as follows: based on massive multidimensional data collected and accumulated in the production process of the direct-current submerged arc furnace, historical data 201 of each measuring point are obtained from the existing system.
The acquisition mode of the direct current furnace input data 202 is as follows: in the dc furnace history data 201, a history data segment having a fixed time length before the occurrence of a typical abnormal event is intercepted a plurality of times as a process of performing feature calculation 203 as dc furnace input data 202, thereby extracting corresponding features.
The electrode current output data 204 is obtained by: in the dc furnace history data 201, a history data segment having a fixed time length before the occurrence of a typical abnormal event is captured multiple times, and the data segment is in one-to-one correspondence with the dc furnace input data 202, and is used as electrode current output data 204 to perform feature calculation 205 to extract corresponding features.
The training process of the direct current furnace model comprises the following steps: based on the dc furnace input data characteristics and the electrode current output data characteristics, the dc furnace model is trained 206 by machine learning, and the model is stored in the dc furnace model prediction 209. Wherein the real-time data 207 is derived from the real-time data acquired by the direct current submerged arc furnace in the current actual production smelting process. The actual dc furnace input 208 is derived from real-time dc submerged arc furnace data 207, and the specific parameters are the same as the dc furnace input data 202 in the historical data 201.
Further, the process of predicting 209 the model of the direct current furnace is as follows: the unit accepts model parameters generated by the direct current furnace model training 206, saves and utilizes the model parameters, calls the model to predict according to the actual input 208 characteristics of the direct current furnace, and predicts the current output characteristics of the current electrode in real time, as shown in fig. 2.
Further, the estimated electrode current output 210 process is: the direct current furnace model prediction 209 unit calculates the estimated characteristics based on the data of the actual direct current furnace input 208.
Further, the electrode current actual output 211 process: the real-time data 207 from the direct current submerged arc furnace has specific parameters identical to the electrode current output data 204 in the historical data, and the corresponding characteristics are extracted by performing characteristic calculation.
Further, electrode current residual calculation 213 process: and (3) aiming at the electrode current actual output 211 characteristic calculated in real time and the model estimated electrode current output characteristic, making an absolute value difference value of the electrode current actual output 211 characteristic and the model estimated electrode current output characteristic, namely an electrode current output characteristic residual error.
Further, the abnormal event judgment 214 process: and selecting a proper threshold value from the electrode current output characteristic residual error as an early warning signal of abnormal events such as material collapse, electrode soft and hard breakage and the like.
The working principle of the abnormal event early warning system of the direct-current submerged arc furnace based on machine learning is explained, and the following explanation can be regarded as the explanation of two specific embodiments of abnormal events, namely material collapse and electrode soft and hard breakage.
The predictability of direct current submerged arc furnace smelting accidents, namely, certain abnormal conditions exist in a period of time before the smelting process accidents happen, and the aim of a machine learning algorithm is to find the abnormal conditions, specifically adopting the following modes:
the machine learning analysis of the data adopts a mode of deploying a local server, and a big data processing edge server is deployed locally, so that the method has obvious advantages in computing capacity and speed compared with a common PC (personal computer), is used for researching, modeling and developing a support algorithm, and is easier to realize a control algorithm. Therefore, based on mass multidimensional data collected and accumulated in the production process of the direct-current submerged arc furnace, historical data and accident record data of each measuring point are obtained from an existing system, data communication is transmitted to a big data server, an intelligent analysis engine is built in the server, and machine learning and deep mining are carried out on the big data by utilizing the elastic data storage, calculation processing capacity and artificial intelligence technology of the big data server, so that data exploration and feature modeling are carried out. By means of an intelligent equipment failure prediction engine, analysis and modeling are carried out on massive historical data measuring points produced by direct-current furnace smelting, prediction algorithms of two abnormal events of material collapse and electrode soft and hard breakage in direct-current furnace smelting are established in a targeted mode, the historical data are used for carrying out reverse verification on the algorithms, so that the prediction algorithms of material collapse and electrode soft and hard breakage accidents are verified, and early warning analysis is carried out on the abnormalities.
Further, the measuring point data of the direct-current submerged arc furnace in a normal smelting state is constructed into a memory matrix, row vectors of the matrix represent the operation data of all measuring points at a certain moment, and column vectors represent the operation data of the measuring points at different moments. And when the deviation between the real-time output and the expected output based on the current input is large, an abnormal state early warning is sent.
Under the healthy running state of the direct current furnace, certain mapping relation exists between the input state information and the output state information. When the relation deviates, the expected mapping relation among all parameters is destroyed, and the smelting process can be in an abnormal state. The machine learning algorithm research of the design combines the data characteristics of a sample set, adopts a mode of combining unsupervised learning and supervised learning, finds out an abnormal mode deviating from a normal state through big data simulation, and finds out abnormal data characteristics.
Unsupervised learning process embodiment:
in the invention, the feature extraction of data repeatedly mentioned in the description of the machine learning process adopts an unsupervised learning mode, and Principal Component Analysis (PCA) is carried out on the data by utilizing a feature value decomposition (EVD) method, so as to reduce the feature dimension of a study object, and meanwhile, the unavoidable information loss in the dimension reduction process is reduced as much as possible, the most effective way is to replace the original sampling feature by a group of new features which are independent in linearity and have smaller dimension, as shown in fig. 3, and the specific calculation steps are as follows:
s201: firstly, obtaining original data of a direct-current submerged arc furnace;
s202: carrying out data zero-mean processing, namely data centralization, on the original data;
s203: then, obtaining a covariance matrix of the sampling characteristics;
s204: then diagonalizing the covariance matrix;
s205: constructing a diagonal matrix with new feature covariance of 0;
s206: calculating to obtain a new feature set matrix composed of standard orthogonal basis vectors through the diagonal matrix;
s207: the new features which are irrelevant to each other are concentrated and discarded, and the new features with higher contribution rate of the main components are reserved;
s208: therefore, the main information of the research object is described, the operation amount of subsequent data processing is greatly reduced, the characteristic calculation process is completed, and the dimension reduction is completed.
Of course, we can also obtain the feature set matrix composed of the target standard orthogonal base vectors directly by the tool provided by the languages of Python or R and the like through a method of obtaining all component achievements of singular value decomposition at one time, and the two methods of feature value decomposition (EVD) and Singular Value Decomposition (SVD) are essentially based on a real symmetric square matrix as a result of multiplying the sampling matrix by its transposed matrix, and the feature set matrix composed of the new target standard orthogonal base vectors is obtained by utilizing the excellent characteristics of the square matrix, which is not described herein.
Supervised learning process embodiment:
after the data features are extracted and the dimension is reduced, the training and testing of the direct-current submerged arc furnace model are carried out in a supervised learning mode. And selecting a proper time length to split the data characteristics of abnormal events such as direct current furnace material collapse and electrode soft and hard breakage, removing data of a fault removal time period, marking a sample of a period of time before the accident occurrence time as an abnormal sample, and marking data of other times as a normal sample. After data splitting, two classification models can be respectively established for two accidents, wherein the XGboost classification algorithm is used for establishing the two classification models, and the specific steps are shown in fig. 4.
S211: firstly, acquiring characteristic data of a direct-current submerged arc furnace;
s212: then, carrying out feature value normalization processing to obtain dimensionless feature data, facilitating comparison and weighting of indexes of different units or orders, wherein the normalization is beneficial to better utilizing data by the model;
s213: the method comprises the steps of carrying out data grouping on samples, and dividing all acquired sample data into training samples and test samples according to a ratio of 4:1 after arranging and grouping;
s214: the training samples form a training data set;
s215: the test samples form a test data set;
s216: training an XGBoost model, wherein the idea of an XGBoost algorithm is to split continuously added trees according to characteristics, one tree added each time is actually used for learning a new function, the residual error of the last prediction is simulated, leaf nodes of the tree correspond to a score, and finally the score corresponding to each tree is added to obtain the predicted value of the sample; substituting training sample data into an XGBoost model for training, selecting a loss function, adding a regularization term on the basis, performing secondary Taylor expansion on an objective function, traversing all values of all features as splitting points according to the principle, selecting a point with the maximum gain for splitting, and iteratively constructing a fault splitting tree by continuously splitting the features until the regularization requirement of splitting is met, and not splitting the tree, so that the model is trained;
s217: performing parameter tuning on the XGBoost model by using the test set data, and selecting the parameter corresponding to the highest accuracy of model classification under each group of parameters as the optimal parameter;
s218: and (5) finishing the establishment of the feature classification model.
The XGBoost algorithm can still keep a good training effect under the condition of large data volume, and finally, data to be tested in real time are substituted into a trained model to detect the abnormal operation state of the current direct-current submerged arc furnace, whether the accident of material collapse and electrode soft and hard breaking is about to occur or not is predicted, and then the abnormal event in the direct-current submerged arc furnace smelting process is avoided by feeding back the abnormal operation state to a basic automatic control system to timely adjust the smelting strategy of the direct-current submerged arc furnace.
In summary, the method for early warning of abnormal events of the direct-current submerged arc furnace based on machine learning comprises the following specific and complete steps:
the method comprises the following steps of S1, a sensor module M1 consisting of a temperature sensor, a pressure sensor, a vibration sensor, a position transmitter, an electric energy detector, a limit switch and the like of a production site of a direct-current submerged arc furnace, performing sensing detection on various physical data parameters in the production process, converting sensed information into an electric signal according to a certain rule, and transmitting the electric signal to a data acquisition module M2 consisting of an ET200SP and a frequency converter through a data cable or a communication cable;
s2, the data acquisition module M2 acquires output signals of the sensor module M1 according to a certain period, and then the acquired data are transmitted to the central processing module M3 through an RJ45 Ethernet cable or an optical fiber via a factory bus switch;
s3, after the central processing module M3 carries out logic control processing on the signals collected from the data acquisition module M2, the signals are simultaneously transmitted to the man-machine interaction module M4 and the machine learning module M5 through an RJ45 Ethernet cable or an optical fiber via a terminal bus switch, the man-machine interaction module M4 executes the step 7, and the machine learning module M5 carries out big data machine learning of the direct-current submerged arc furnace according to the early warning method principles shown in the figures 3 and 5;
s4, in the machine learning process, a machine learning module M5 firstly performs feature extraction of each part of data according to the steps shown in the figure 2-1 by a feature extraction module M5-1, and an unsupervised machine learning method is adopted;
s5, the extracted features are further transmitted to an analysis learning module M5-2, training and testing of the direct-current submerged arc furnace model are carried out according to the steps shown in FIG. 4, and a supervised machine learning method is adopted;
s6, transmitting abnormal data appearing in the analysis learning module M5-2 to the business application module M5-3, wherein the business application module M5-3 realizes the functions of state monitoring, fault diagnosis and the like, and performs data communication and seamless butt joint with the central processing module M3;
s7: the man-machine interaction module M4 displays important information in the central processing module M3 in real time, receives a control instruction of a process operator and reversely transmits the control instruction to the central processing module M3;
s8: when the early warning information of the abnormal event is received, the central processing module M3 automatically switches the production process control strategy of the direct-current submerged arc furnace or receives an adjustment strategy instruction from a process operator of the man-machine interaction module M4;
s9: the central processing module M3 is transmitted to the numerical control output module M6 through an RJ45 Ethernet cable or an optical fiber via a factory bus switch;
s10: the control signal is transmitted to an actuating mechanism module M7 consisting of an electromagnetic valve, a motor and the like through a numerical control output module M6, so that the actual action of the field device of the direct-current submerged arc furnace is completed.
S11: the data parameters of the production field process are changed along with the change, and are perceived by the sensor module M1, so that the process is continuously circulated, and the reliable, stable and optimal operation of the whole production system is always ensured.
As shown in fig. 3, the dc furnace algorithm result 301 includes the generation of an accident prediction model 302 and an accident pre-warning model 303.
The accident prediction model comprises a collapse prediction model 304 and an electrode soft and hard break prediction model 305, and the recall rate on the verification effect is more than 95%;
the accident pre-warning model comprises a material collapse pre-warning model 306 and an electrode soft and hard breakage pre-warning model 307, and pre-warns the accident in advance through real-time residual errors existing in a period of time before the accident occurs and dynamic threshold analysis. Early warning is carried out 1 hour in advance on the collapse accident; the accident of the electrode soft and hard break is early warned for 4 hours.
The direct-current submerged arc furnace early warning system and the direct-current submerged arc furnace early warning method based on machine learning are designed and developed, wherein part of related technologies can be partially applied and expanded to an alternating-current submerged arc furnace control system, and the part of technologies are also in the protection scope of the scheme and are all constrained by the technical scheme.
The foregoing is only a preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art, who is within the scope of the present invention, should make equivalent substitutions or modifications according to the technical scheme of the present invention and the inventive concept thereof, and should be covered by the scope of the present invention.
Claims (6)
1. The utility model provides a direct current submerged arc furnace abnormal event early warning system based on machine learning which characterized in that includes:
the sensor module (M1) is used for sensing and detecting a plurality of physical data parameters of the direct-current submerged arc furnace in the production process and converting the sensed information into an electric signal according to a certain rule;
the data acquisition module (M2) is used for receiving the electric signals transmitted by the sensor module (M1) according to a certain period, and the data acquisition module (M2) is used for carrying out data processing and outputting on the received electric signals by adopting an anti-shake, filtering and passivating method;
the central processing module (M3) is used for receiving the data information transmitted by the data acquisition module (M2), and the central processing module (M3) is used for carrying out logic control on the received signals and adjusting the control strategy of the production process;
the man-machine interaction module (M4) is used for receiving logic control information transmitted by the central processing module (M3), the man-machine interaction module (M4) is used for displaying the received data information in real time, and sending a control instruction of a process operator to the central processing module (M3) for logic control, so that bidirectional data communication with the central processing module (M3) is realized;
the machine learning module (M5) is used for receiving logic control data transmitted by the central processing module (M3), the machine learning module (M5) is used for analyzing and extracting characteristics of massive historical data measuring points produced by smelting in the direct-current furnace, estimating output characteristics of current electrode current in real time, acquiring electrode current output characteristic residual errors according to actual output characteristics of the current electrode current and the electrode current output characteristics estimated by the direct-current submerged arc furnace model, setting a proper threshold in the electrode current output characteristic residual errors as a judging basis for abnormal events of material collapse and electrode soft and hard breakage, and outputting an early warning signal when the characteristic residual errors are larger than the set threshold; the machine learning module (M5) analyzes and models massive historical data measuring points produced by direct-current furnace smelting by means of an equipment fault intelligent prediction engine, specifically establishes prediction algorithms of two abnormal events of material collapse and electrode soft and hard breakage in direct-current furnace smelting, and uses historical data to carry out reverse verification on the algorithms so as to verify the prediction algorithms of material collapse and electrode soft and hard breakage accidents, and carries out early warning analysis on the abnormalities; the machine learning module (M5) also performs big data processing, feature extraction, analysis learning and business application on the received logic control data, and simultaneously feeds back learning result information to the central processing module (M3);
a numerical control output module (M6) for receiving the control instruction sent by the central processing module (M3), wherein the numerical control output module (M6) converts information into various physical signals and outputs the physical signals;
and the executing mechanism module (M7) is used for receiving the physical signals transmitted by the numerical control output module (M6), and the executing mechanism module (M7) is used for executing actions corresponding to the physical signals on various devices in the production process of the direct-current submerged arc furnace.
2. The machine learning based direct current submerged arc furnace abnormal event early warning system of claim 1, further characterized by: the machine learning module (M5) comprises a feature extraction module (M5-1), an analysis learning module (M5-2) and a business application module (M5-3);
the feature extraction module (M5-1) adopts an unsupervised machine learning method to analyze the principal components of the data: firstly, receiving original data of a direct-current submerged arc furnace, carrying out zero-valued treatment on the data, solving a covariance matrix of sampling characteristics, carrying out diagonalization treatment on the covariance matrix, constructing a diagonal matrix with new characteristic covariance of 0, and thus calculating a characteristic set matrix consisting of new standard orthogonal basis vectors, and retaining new characteristics with high contribution rate of main components;
the analysis learning module (M5-2) adopts a supervised machine learning method to train and test the direct-current submerged arc furnace model: firstly, receiving direct-current submerged arc furnace characteristic data transmitted by a characteristic extraction module (M5-1), carrying out normalization processing on the characteristic data, training an XGBoost model by adopting a training data set, and carrying out parameter tuning on the XGBoost model by adopting a test data set so as to finish characteristic classification;
the business application module (M5-3) receives the abnormal data transmitted by the analysis learning module (M5-2) to perform state monitoring and fault diagnosis on the direct-current submerged arc furnace.
3. The machine learning based direct current submerged arc furnace abnormal event early warning system of claim 2, further characterized by: the machine learning module (M5) is based on mass multidimensional data acquired and accumulated in the production process of the direct current submerged arc furnace, historical data of each measuring point and accident record data are obtained from an existing system, an intelligent analysis engine is built in a server, the large data are subjected to machine learning and deep mining by utilizing elastic data storage, calculation processing capacity and artificial intelligence technology of a large data server, data exploration and feature modeling are performed, a direct current furnace algorithm result (301) is generated, the direct current furnace algorithm result (301) comprises an accident prediction model (302) and an accident early warning model (303), the accident prediction model (302) comprises a collapse prediction model (304) and an electrode soft and hard breakage prediction model (305), and the accident early warning model (303) comprises a collapse early warning module (306) and an electrode soft and hard breakage early warning model (307).
4. The machine learning based direct current submerged arc furnace abnormal event warning system of claim 3, further characterized by: the machine learning module (M5) constructs measuring point data of the direct current submerged arc furnace in a normal smelting state into a memory matrix, row vectors of the matrix represent the operation data of all measuring points at a certain moment, column vectors represent the operation data of the measuring points at different moments, and when the deviation between real-time output and expected output based on current input is larger than a set threshold value, abnormal state early warning is sent out.
5. The machine learning based direct current submerged arc furnace abnormal event early warning system of claim 2, further characterized by: the machine learning module (M5) adopts a mode of combining unsupervised learning and supervised learning according to the working principle that the input and output state information has a certain mapping relation in the healthy running state of the direct current furnace, and queries an abnormal mode with deviation from the normal state through a big data simulation method and queries abnormal data characteristics.
6. The machine learning based abnormal event early warning system for direct current submerged arc furnaces according to any one of claims 1 to 5, wherein the machine learning module (M5) uses the history data to perform reverse verification on the algorithm, wherein the history data is obtained by: historical data segments with fixed time length before the occurrence time of a typical abnormal event are intercepted for a plurality of times and used as direct current furnace input data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010725828.9A CN111898669B (en) | 2020-07-24 | 2020-07-24 | Abnormal event early warning system of direct-current submerged arc furnace based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010725828.9A CN111898669B (en) | 2020-07-24 | 2020-07-24 | Abnormal event early warning system of direct-current submerged arc furnace based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111898669A CN111898669A (en) | 2020-11-06 |
| CN111898669B true CN111898669B (en) | 2023-07-14 |
Family
ID=73189945
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010725828.9A Active CN111898669B (en) | 2020-07-24 | 2020-07-24 | Abnormal event early warning system of direct-current submerged arc furnace based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111898669B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112348101B (en) * | 2020-11-16 | 2023-04-07 | 中冶赛迪信息技术(重庆)有限公司 | Steel rolling fuel consumption early warning method and system based on abnormal data analysis |
| CN112926749A (en) * | 2020-12-30 | 2021-06-08 | 国网宁夏电力有限公司信息通信公司 | Intelligent power grid information equipment monitoring system and method |
| CN113281581B (en) * | 2021-04-14 | 2022-04-01 | 中国重型机械研究院股份公司 | Test method of LF furnace electrode test system |
| CN113686922A (en) * | 2021-06-28 | 2021-11-23 | 深圳爱莫科技有限公司 | Online intelligent monitoring device and method for risks of reaction kettle |
| CN113962405A (en) * | 2021-09-02 | 2022-01-21 | 大连重工机电设备成套有限公司 | Intelligent submerged arc furnace remote operation and maintenance system |
| CN114202159A (en) * | 2021-11-12 | 2022-03-18 | 寰清能源科技(上海)有限公司 | Online monitoring method and system suitable for material collapse accident of closed submerged arc furnace |
| CN114518732A (en) * | 2022-02-10 | 2022-05-20 | 宁夏瑞资联实业有限公司 | Furnace control system of DCS submerged arc furnace |
| CN116379793B (en) * | 2023-06-02 | 2023-08-15 | 青岛智控菲特软件科技有限公司 | Submerged arc furnace short-net regulation and control data processing method |
| CN116647586A (en) * | 2023-07-24 | 2023-08-25 | 山东工泵电机有限公司 | Method for realizing remote control of water and fertilizer integrated intelligent pump house by cloud computing |
| CN117303510B (en) * | 2023-09-07 | 2024-06-25 | 兰州理工大学 | Self-adaptive control method for waveform characteristic analysis of wastewater treatment power supply |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104020672A (en) * | 2014-05-29 | 2014-09-03 | 西安交通大学 | Intelligent control method of DC hollow electrode calcium carbide furnace |
| CN108460144A (en) * | 2018-03-14 | 2018-08-28 | 西安华光信息技术有限责任公司 | A kind of coal equipment fault early-warning system and method based on machine learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1914638A1 (en) * | 2006-10-18 | 2008-04-23 | Bp Oil International Limited | Abnormal event detection using principal component analysis |
| CN109710636B (en) * | 2018-11-13 | 2022-10-21 | 广东工业大学 | Unsupervised industrial system anomaly detection method based on deep transfer learning |
-
2020
- 2020-07-24 CN CN202010725828.9A patent/CN111898669B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104020672A (en) * | 2014-05-29 | 2014-09-03 | 西安交通大学 | Intelligent control method of DC hollow electrode calcium carbide furnace |
| CN108460144A (en) * | 2018-03-14 | 2018-08-28 | 西安华光信息技术有限责任公司 | A kind of coal equipment fault early-warning system and method based on machine learning |
Non-Patent Citations (1)
| Title |
|---|
| 基于主元分析的高炉异常炉况检测;窦克勤;叶昊;张海峰;李宏玉;;上海交通大学学报(第12期);1862-1867 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111898669A (en) | 2020-11-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111898669B (en) | Abnormal event early warning system of direct-current submerged arc furnace based on machine learning | |
| CN111596604B (en) | Intelligent fault diagnosis and self-healing control system and method for engineering equipment based on digital twinning | |
| CN107807860B (en) | Power failure analysis method and system based on matrix decomposition | |
| CN113708350B (en) | Cloud edge cooperation-based power distribution area heavy overload abnormality judgment method and system | |
| CN106443433B (en) | Breaker state monitoring system and method based on opening and closing coil current | |
| CN114386537A (en) | Lithium battery fault diagnosis method and device based on Catboost and electronic equipment | |
| CN117612345A (en) | Power equipment state monitoring and alarming system and method | |
| CN117408162B (en) | Power grid fault control method based on digital twin | |
| CN116961215A (en) | Rapid fault response processing method for power system | |
| CN106296440A (en) | Based on transformer station's warning information analysis and decision system integrated for ANN and ES and method | |
| CN118213936A (en) | Voltage monitoring and voltage protecting system and method | |
| CN116938676A (en) | Communication risk combined early warning method based on data source analysis | |
| CN116418117A (en) | Equipment detection system for intelligent power grid | |
| CN117471346A (en) | Method and system for determining remaining life and health status of retired battery module | |
| CN109030976B (en) | Power electronic equipment monitoring and diagnosing system and method | |
| CN118193954A (en) | Power distribution network abnormal data detection method and system based on edge calculation | |
| CN116704729A (en) | Industrial kiln early warning system and method based on big data analysis | |
| CN116796261B (en) | Closed switch equipment mechanical characteristic prediction method based on artificial intelligence | |
| CN117390944A (en) | Substation operation condition simulation system | |
| CN117390546A (en) | Multimode database fusion calculation model for instant anti-electricity-theft detection | |
| CN111934903A (en) | Docker container fault intelligent prediction method based on time sequence evolution genes | |
| CN111391694A (en) | Multi-level rapid data monitoring method and system for operation and maintenance of charging station | |
| CN118094232B (en) | Dynamic monitoring production energy consumption control system and method | |
| CN117034157B (en) | Hydropower equipment fault identification method and system combining multimodal operation data | |
| CN116168518A (en) | Lime kiln early warning system and method based on big data analysis |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |