CN112001527B - Industrial production process target data prediction method of multi-feature fusion depth neural network - Google Patents

Industrial production process target data prediction method of multi-feature fusion depth neural network Download PDF

Info

Publication number
CN112001527B
CN112001527B CN202010744152.8A CN202010744152A CN112001527B CN 112001527 B CN112001527 B CN 112001527B CN 202010744152 A CN202010744152 A CN 202010744152A CN 112001527 B CN112001527 B CN 112001527B
Authority
CN
China
Prior art keywords
data
neural network
layer
gating
feature fusion
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
Application number
CN202010744152.8A
Other languages
Chinese (zh)
Other versions
CN112001527A (en
Inventor
曾九孙
欧阳航
丁克勤
蔡晋辉
姚燕
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.)
China Jiliang University
Original Assignee
China Jiliang University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China Jiliang University filed Critical China Jiliang University
Priority to CN202010744152.8A priority Critical patent/CN112001527B/en
Publication of CN112001527A publication Critical patent/CN112001527A/en
Application granted granted Critical
Publication of CN112001527B publication Critical patent/CN112001527B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Human Resources & Organizations (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Economics (AREA)
  • Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Molecular Biology (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Operations Research (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Feedback Control In General (AREA)
  • Image Analysis (AREA)

Abstract

The utility model discloses an industrial production process target data prediction method of a multi-feature fusion deep neural network. Collecting time sequence data of other variables related to the key variables in industrial equipment by using a sensor through equidistant sampling, and carrying out predictive analysis on the time sequence data of the key variables in the flow industry; inputting the training data into a depth convolution neural network designed and constructed in advance to train; the historical data of the key variables are input to a deep gating circulating neural network for learning after being segmented according to time steps; and the output characteristics obtained by the two networks are fused and input into the full-connection layer by utilizing a multi-characteristic fusion method, and network parameters are optimized through back propagation, so that the prediction precision is improved. The utility model provides reliable and effective target variable parameter prediction for process monitoring in industrial production, and relieves the hysteresis of measuring critical variables such as molten iron silicon content and the like in industrial production.

Description

Industrial production process target data prediction method of multi-feature fusion depth neural network
Technical Field
The utility model belongs to an industrial parameter prediction method in the field of flow industrial production, and particularly relates to an industrial production process key parameter prediction method based on multi-feature fusion.
Background
The current industrial system gradually goes to the intelligent, integrated and automatic degree, the functions of the whole industrial system are more and more perfect, and therefore, the correlation among variables in the system is more and more compact. Meanwhile, with the appearance of various sensors, more and more parameters are available in the process industry, and a data source is provided for industrial big data processing. However, there is still a great hysteresis in the measurement of critical parameters in the process industry, such as the silicon content of molten iron in blast furnace ironmaking, which is often obtained only by soft measurement, and in the conventional direct measurement method, there is a great hysteresis.
The key parameters in the process industry can be predicted in advance, so that the current and future states of the whole industrial system can be predicted, a practitioner can find out which equipment in the industrial system has problems in time, and the system is convenient to operate in advance, so that the system is stabilized, and industrial accidents are prevented. The realization of the prediction of key parameters of the process industry has great significance for stabilizing the whole industrial system and the industrial production safety.
There are two common solutions for industrial process critical parameter monitoring. One approach is to use new instruments for direct measurement with a high degree of real-time and accuracy by continually developing new instruments. The second method is to measure other variables related to the critical parameters to be detected, obtain nonlinear relations among the variables by a statistical or neural network method for calculating the critical variables, and indirectly obtain the target parameters. Compared with the second method, the first method has the problems of high equipment cost, long research and development period, long analysis period, serious hysteresis and the like, so that operators cannot know the real-time state of an industrial system in time, the production quality is reduced, and the energy waste is serious. The second method, also called soft measurement, directly measures easily monitored variables and builds a predictive model by learning the potential relationship between the target variable and the known variable [2] Thus, the real-time prediction and estimation of the target variable can be completedThe method has a short development cycle and low cost relative to the first method.
The Chinese patent with the publication number of CN110400007A discloses a molten iron quality forecasting method based on an improved gate control circulating neural network, which reduces the complexity of a model by combining a reset gate and an update gate of the gate control circulating neural network into a single disposal gate. However, the gating cyclic neural network used in the method is not good at capturing the characteristics of the data dimension, so that the information of the data dimension is easy to lose, and the prediction effect is not ideal for more complex process industry problems and working conditions. The method and the device can solve the problem that the data dimension information is lost by introducing the neural network structure which is good at learning the data dimension through the multi-feature fusion technology, and finally obtain the predicted value by using the time dimension feature and the data dimension feature at the same time, so that the prediction precision of the key parameters is further improved.
Disclosure of Invention
In order to solve the technical problems in the background art, the utility model provides a process industry key parameter prediction method based on a multi-feature fusion depth neural network, which can perform data mining on original time sequence signals in the process industry, learn the relation between target key parameters and easily-measured parameters and predict the target key parameters in real time.
The method is suitable for monitoring the state of the flow industry real-time system, particularly has strong sequence correlation of the target key parameters, has strong inertia of the measured state parameters of the industrial system, provides reliable and effective target variable parameter prediction for process monitoring in industrial production, and relieves the hysteresis of measuring key variables such as silicon content of molten iron in industrial production.
The technical scheme adopted by the utility model is that the method specifically comprises the following steps:
the method generally comprises the following steps of:
step 1, sampling industrial production processes under normal working conditions at different moments through equal time intervals to obtain production state data containing target variable parameters, and preprocessing the production state data to obtain a data set;
and 2, inputting the data set into a multi-feature fusion depth neural network to obtain a prediction result of the target variable parameter.
The method specifically comprises the following steps:
step 1, sampling industrial production processes under normal working conditions at different moments through equal time intervals to obtain production state data, wherein the production state data are multivariate parameters obtained through detection of different sensors, the acquired production state data are respectively segmented according to the moments and the sensors, and time sequence data and sensor classification data are respectively used as two data sets which are subsequently and simultaneously input into a multi-feature fusion depth neural network comprising two kinds of neural networks; specifically, the signal time sequence data of the target variable can be divided into sub-data with the same length according to time sequence, and each sub-data contains the same number of sampling points.
The production state data refer to variable parameters obtained through sensor detection, including data of target variable parameters and reference variable parameters, wherein the target variable parameters are variable parameters which cannot be obtained through the sensor direct real-time accurate detection and can only be obtained through sensor lag detection, and the reference variable parameters are variable parameters which can be obtained through the sensor direct real-time accurate detection;
the target data are target variable parameters which cannot be obtained through direct real-time accurate detection by the sensor and can only be obtained through sensor hysteresis detection, and are also key parameters in the industrial production process. The target variable parameters generally refer to key parameters such as silicon content of molten iron, butane concentration at the bottom of the well and the like which are difficult to directly measure in the process industry.
Each sensor detects and obtains a variable parameter, and data obtained by different sensors at the same time are different variable parameters.
Step 2, the multi-feature fusion depth neural network mainly comprises a depth convolution neural network, a depth gating circulation neural network and a multi-feature fusion layer, wherein the output ends of the depth convolution neural network and the depth gating circulation neural network are connected to the multi-feature fusion layer;
2.1, each sensor detecting the time series data as the variable parameter directly obtained at the same moment, taking the reference variable parameter in the time series data as input, taking the target variable parameter obtained by sensor hysteresis detection in the time series data as output, inputting the target variable parameter into a deep convolutional neural network for processing, learning to obtain the relation between different variable parameters at the same moment, and obtaining the relation between the target variable parameter and the variable parameter of the production state data;
2.2, setting time steps, sampling a part corresponding to the target variable parameters in the sensor classification data according to the time steps, inputting the sampled data into a single-step predicted depth gating cyclic neural network, learning the relation between different moments of the target variable parameters, and learning the periodicity and the trending of a time sequence to obtain the relation between a plurality of historical moment data of the target variable parameters and the current moment data; the target variable is the sensor measurement parameter measured with actual hysteresis.
2.3, inputting the output prediction data of the deep convolutional neural network before the final full-connection layer and the output prediction data of the deep gate control cyclic neural network before the final full-connection layer into a multi-feature fusion layer together, and fusing by utilizing multi-feature fusion, specifically: the addition layer is utilized to add and connect the two output predicted data in parallel to obtain information after combining characteristics, the information is flattened into one-dimensional data through the flattening layer and then is input into the final full-connection layer, and a predicted value result of the target variable parameter is output; and combining the high-dimensional characteristic data with the same number of samples, mapping all the characteristics of the data dimension and the characteristics of the time dimension into a high-dimensional characteristic space, and finally inputting the data into the fully-connected neural network to output a prediction result of a final target.
The output prediction data comprises a window number, a window size, a characteristic dimension and a plurality of prediction values corresponding to a plurality of window numbers.
The output prediction data of the deep convolutional neural network before the last fully-connected layer and the output prediction data of the deep gated cyclic neural network before the last fully-connected layer are high-dimensional characteristic data, and the high-dimensional characteristic is obtained by mapping the original data into a higher-dimensional space.
The two data sets are obtained through the processing of the steps, each data vector of the data set input into the deep convolutional neural network contains a reference variable parameter except a target variable parameter, and each data vector of the data set input into the deep gated convolutional neural network contains historical data of a fixed time step before the target variable parameter to be predicted; and respectively inputting the data set into a deep convolutional neural network and a deep gating loop network to extract the characteristics. After two networks, the two training samples are mapped to the same-dimension feature space, the two features are combined, and the high-dimension data is flattened into one-dimension data through a flat layer and is input to a full-connection layer for outputting a target variable result.
2.4, inputting the fusion features output by the multi-feature fusion layer into the full-connection layer of the last layer, training the deep convolutional neural network and the deep gating cyclic network by using a back propagation algorithm, adjusting weight parameters of the deep convolutional neural network and the deep gating cyclic network, realizing the approximation process of a predicted value and a true value, and optimizing to obtain the trained deep convolutional neural network and the deep gating cyclic network;
and 2.5, forming a trained multi-feature fusion depth neural network by combining the trained deep convolutional neural network and the depth gate control loop network with the multi-feature fusion layer, inputting the reference variable parameter data of the condition to be tested into the trained multi-feature fusion depth neural network, and outputting prediction to obtain the target variable parameter.
The deep convolution neural network captures data dimension characteristic data of the multivariate time sequence data, and output prediction data of the deep convolution neural network before the final full-connection layer is the data dimension characteristic data; the depth gating loop network captures time dimension characteristic data of target data, and output prediction data of the depth gating loop neural network before the last full connection layer is the time dimension characteristic data.
The utility model relates to periodic signal time series data acquired by a sensor in industrial production, which comprises but is not limited to important variables in blast furnace ironmaking processes such as blast furnace coal injection quantity, oxygen content, pressure difference in the furnace and the like.
The deep convolutional neural network is mainly composed of a plurality of convolutional blocks and a maximum pooling layer which are connected in sequence, each convolutional block is composed of a plurality of convolutional layers which are connected in sequence, time sequence data is normalized to eliminate dimensional relations among subsequences, then each convolutional block which is subjected to time domain convolution is processed, convolution calculation is carried out by a learnable convolution kernel, characteristics are obtained by using an activation function and mapped into a characteristic space, and a calculation formula is as follows:
C i ＝f(W i *C i-1 +b i )
in which W is i Shared weights representing the ith convolution block, and features C of the ith-1 convolution block i-1 Convolution, which is the sign of the convolution operation, b i Is the offset vector of the ith convolution block, f () represents the nonlinear activation function; obtaining the characteristic C of the ith convolution block by a nonlinear activation function relu i ；
After the time domain convolution processing of each convolution block, the output of the final maximum pooling layer is the data dimension characteristic data of the target characteristic space.
The deep gating circulating neural network is composed of three gating circulating layers which are sequentially connected, and a three-layer gating circulating network is formed.
The deep gating circulating neural network comprises three layers of processing modules: the first processing module comprises a first gating circulation layer, and the first gating circulation layer is sequentially transmitted to the second processing module through a reset gate and an update gate; the second processing module comprises a second gating circulation layer and a second Dropout layer, and the second gating circulation layer is sequentially transmitted to the third processing module through a reset gate and an update gate; the third processing module comprises a third gating circulation layer, and the third gating circulation layer is sequentially transmitted to the mapping layer through a reset gate and an update gate; the first gating circulation layer, the second gating circulation layer and the third gating circulation layer are mainly composed of sequential gating circulation units. The mapping layer serves as an intermediate output layer to output high-dimensional features of the original data time dimension.
In said 2.4, for; the deep convolutional neural network and the deep gated recurrent neural network both establish the following loss functions:
wherein Loss represents a Loss value between a predicted value and a true value, m represents the number of sampling time points in one sample, y i Representing the i-th true value of the value,representing an i-th predicted value;
the predicted value is the data predicted and output by the network, and the actual value is the target variable parameter measured by actual hysteresis.
Taking input data as a sample, taking a weight derivative value of each convolution layer in each convolution block as a gradient of the sample in the deep convolution neural network, and taking a weight derivative value of a gating circulation unit in a first gating circulation layer as a gradient of the sample in the deep gating circulation neural network; the network parameters of the deep convolutional neural network and the deep gating cyclic neural network are all adjusted to weight values in the network through the following gradient update, and the optimizer is an Adam optimizer, and the gradient update method is as follows:
in the formula, the velocity/momentum at the moment t,the average gradient of all samples is the cumulative square gradient at the moment t, the exponential decay rate of the first moment estimation is the exponential decay rate of the second moment estimation, +.>For the speed after correction of the deviation, +.>The cumulative square gradient after the deviation correction is a network parameter at the time t, which is a non-zero parameter, specifically, a very small number, which avoids the situation that the denominator is 0, is a learning rate, and the ". As used herein, indicates Hadamard (Hadamard) product.
When the loss function tends to be stable and does not decline any more, the performance test of the whole network can be carried out by utilizing the test set data, and the method is used for predicting the key parameter variable of the target variable parameter in advance in the actual production process.
In the specific implementation of the utility model, the performance of the multi-feature fusion depth neural network in predicting the key parameters of the process industry which are difficult to directly measure is tested and checked by using the known multivariate signal time sequence data, the prediction result is counted, and the prediction accuracy is calculated.
Therefore, the utility model realizes the prediction of the key parameters of the process industry by collecting the multivariate time series signals, training the complete model offline and arranging the model into the online measurement equipment. The specific operation flow is as follows, firstly, the time series signals of the pair variables under normal production are collected, the process variable signals except the target variables are processed by utilizing data normalization to eliminate the dimensional relation among all the subsequences, and then the original data are divided into data fragments with equal sampling time intervals according to fixed time steps. And respectively inputting the two groups of sample data into a convolutional neural network and a gate control loop network for training, acquiring hidden state data of the feature of the fused data dimension and the time dimension, and then inputting the hidden state data containing the fused feature into a full-connection layer to output a result of the appointed dimension, namely a target variable prediction result. And reversely transmitting errors of the predicted value and the true value to two networks through a reverse propagation algorithm for optimizing network parameters of an offline model, so that the prediction progress is improved, and the prediction of the key parameters of the process industry is realized.
The method performs characteristic extraction on big data generated by the industrial seeds of the process, fuses time dimension and data dimension information, realizes the prediction of key parameters of the process industry, and can realize the prediction of key parameters in the process industry such as blast furnace ironmaking, debutanizer industry refining and the like, including but not limited to the silicon content of molten iron and the butane concentration at the bottom of the well.
The utility model decomposes the multivariate time series big data, which is used for inputting two different neural networks at the same time to capture the potential relation between variables in the original data and the time dependence of the target variable; and meanwhile, the high-dimensional characteristics obtained by the deep convolutional neural network and the deep gating cyclic neural network are fused, so that the high accuracy of the prediction result is realized by utilizing a multi-characteristic fusion technology.
The multi-feature fusion is specifically that the mapping results of the deep convolutional neural network and the deep gating cyclic neural network are combined into the same high-dimensional feature space, and then information in the feature space is input into a full-connection layer of the last layer to obtain a predicted value, and the parameters of the two networks are adjusted by back propagation errors, so that the loss function is minimum as much as possible.
The process industrial parameters refer to periodic signal time series data acquired by a sensor in industrial production, and include but are not limited to important variables in the blast furnace ironmaking process, such as blast furnace coal injection quantity, oxygen content, top pressure, furnace top temperature, top gas component content and the like.
The time dependence refers to the degree to which the current value of the target predicted variable is affected by the previous system state, and is generally classified into long-term dependence and short-term dependence.
In the implementation, the collected multi-variable time series data is complemented by a default value, the variable with low correlation degree with the target variable is eliminated, and the time step division original data is set to integrate sub-data with the same length according to a specific industrial system.
The method can process the time series information in the multi-variable data and the linear and nonlinear relation between the variables, and the multi-layer extraction characteristics avoid the problem of dimension curse which is easy to occur when large data analysis occurs, and simultaneously maintain and improve the accuracy of target variable prediction, namely, the method has good accuracy when realizing data calculation simplification.
Compared with the traditional time sequence prediction method, the method adopts the convolutional neural network and the deep learning of the gating cycle to spontaneously extract a large amount of variable relationship and time characteristic information from the given sample, and does not need to rely on excessive flow industrial production information. The utility model only provides a training sample set and a test sample set for the model, the deep learning model can effectively extract the characteristic knowledge contained in the sample, automatically update the weight to obtain the influence degree of other variables on the key variable, learn the trend and periodicity of the key variable sequence from the past information, and finally obtain the predicted value of the current state of the key variable through the output layer. And continuously adjusting and optimizing the model parameters through reverse transfer errors to continuously improve the prediction accuracy and finally finish the prediction of key variables in the process industry.
Compared with the prior art, the utility model has the following beneficial effects:
1. by normalizing the information of other variables except the target variable, the off-line training time of the network is reduced, and faster model updating is realized.
2. By the multi-feature fusion technology, the convolution network which is good at processing the data dimension information plays the biggest role, makes up the deficiency of the gating circulation network which is good at processing the time dimension information, and improves the prediction accuracy of the final network.
3. Through the deep neural network structure, the convolutional neural network can learn the nonlinear relation between other variables and target variables, and the gating cyclic neural network can learn the nonlinear relation between the past moment state and the current moment state, so that the capability of the model for processing nonlinear problems is finally improved.
4. And the weight parameters of the two networks are updated simultaneously by using the error of the predicted value and the true value through a back propagation algorithm, so that the loss value is reduced, and the autonomous learning of the model is realized.
In summary, the method realizes the combination of data dimension and time dimension information through multi-feature fusion, thereby improving the prediction capability of the neural network and reducing the cost of measurement of key parameters of the process industry, so as to realize the prediction of the key parameters of the process industry and provide reliable and effective technical support for the online monitoring of the process industry.
Drawings
FIG. 1 is a schematic diagram of a blast furnace ironmaking structure according to the present utility model;
FIG. 2 is a block diagram of a multi-feature fusion depth neural network of the present utility model;
FIG. 3 is a block diagram of a deep convolutional neural network of the present utility model;
FIG. 4 is a block diagram of a deep gated recurrent neural network of the present utility model;
FIG. 5 is a graph of the model training set loss value versus the test set loss value for the present utility model;
FIG. 6 shows the result of predicting the silicon content of molten iron in a blast furnace according to the present utility model.
Detailed Description
The utility model is further described below with reference to the drawings and examples.
The embodiments implemented according to the complete method of the present disclosure and the implementation procedure are as follows:
the process industrial data for implementing the method is data measured by a blast furnace ironmaking process industrial production line of a steel-making foundry. Taking the process of predicting the silicon content of molten iron as an example, the prediction of key parameters of the flow industry is described in detail based on the specific processing process of the data.
In specific implementation, the utility model periodically uses new fault-free data to update model parameters, thereby avoiding model failure caused by accumulated error and prediction precision reduction. Abnormal data caused by artificial factors and sensor failure in the blast furnace ironmaking process can influence the accuracy of model prediction, and then quality evaluation is carried out on training data, so that the abnormal data and irrelevant variables are removed.
The process industrial data which is embodied by the utility model is the data measured by the blast furnace ironmaking process industrial production line of the steel making factory, but is not limited to the data, and the process industrial key parameters can be predicted by the utility model as long as continuous variable time series data related to the target key parameters can be measured in the process industrial system. The experimental data set consists of variables such as coal injection quantity, air inlet quantity, air permeability, air temperature, feeding speed, furnace top pressure difference and oxygen blowing quantity. The quality of the iron water produced during the test was relatively stable, with silicon content floating between 0.4 and 0.6, and individual special cases were outside this range.
As shown in fig. 1, the process realizes iron making from iron ore, in which the molten iron contains a small amount of impurities, and in which the silicon content is an important parameter for measuring the temperature in the blast furnace.
The process is specifically as follows: iron-containing raw materials (iron ore), fuel (coal dust and coke) and other auxiliary raw materials are charged into the top of a blast furnace according to a certain proportion, and hot air from a hot blast furnace is blown into the blast furnace from an air inlet at the lower part of the blast furnace to be used for assisting the combustion of the coke. Carbon and high temperature air contact reaction at high temperature to generate CO and H 2 . As the raw materials and fuel descend in the furnace, the ascending gas contacts the raw materials, and heat transfer, reduction, melting and decarburization sequentially occur to generate molten iron, and other components in the raw materials are combined with products after the reaction of the fuel to form slag. The whole process produces two byproducts, namely blast furnace gas and slag.
The feeding speed, the coal injection quantity, the blast quantity, the oxygen blowing quantity and the like are controllable variables, the blast temperature, the pressure difference in the furnace and the like are required to be measured by a sensor, the values of the controllable variables are stored at equal time intervals, the variable values related to the silicon content of molten iron in the process industry are sampled at equal time intervals, and all measured data files are stored in a csv format.
Based on the above, the present example uses 7 kinds of blast furnace state sample data related to silicon content of molten iron in total, and the variable numbers and specific variable names are shown in table 1.
TABLE 1
Wherein the target variable parameter is represented, u i Representing other variable parameters than the target variable. As shown in fig. 2, the implementation process conveniently adapts to the input requirements of two deep neural networks by dividing the multivariate data into two data sets, and the prediction process of the key variables of the flow industry comprises the following steps:
step 1, a sensor is used for sampling and collecting time series signal data with high correlation with a target variable parameter in a process industrial system through intervals to serve as a training sample;
and step 2, processing the time sequence signal data to obtain training samples which can be input into a deep convolutional neural network and a deep gating cyclic network. Normalizing the time sequence data of the process variable signals except the target variable to eliminate the dimension relation among the subsequences; dividing the signal time sequence data of the target variable into sub-data with the same length according to time sequence, wherein each sub-data comprises the same number of sampling points, as shown in fig. 2;
step 3, processing the multivariable signal time sequence data of the training samples by adopting the step 2 to obtain two samples, wherein each sample of the training samples input into the deep convolutional neural network comprises other variable values except key parameters, and the structure of the deep convolutional neural network is shown in figure 3; each sample of training samples input into the deep gating circulating neural network contains historical data of a fixed time step before a target key parameter to be predicted, the structure of the deep gating circulating neural network is shown in fig. 4, and the two training samples are respectively input into the deep convolutional neural network and the deep gating circulating network for feature extraction; and finally, fusing information output by the two neural networks into the same feature space through a connecting layer, merging the two parts of features, flattening the high-dimensional data into one-dimensional data through a flat layer, and inputting the one-dimensional data into a full-connection layer for outputting a target variable result.
And 4, after combining the characteristics of the deep convolutional neural network and the deep gating cyclic neural network, inputting the hidden state into an output layer of the fully-connected network to output a result, simultaneously adjusting parameters in the convolutional neural network and the gating cyclic neural network according to a loss value returned in a reverse transmission process, judging whether the network training degree is good or not according to the result obtained by the loss function, and controlling the stopping of training. The model training results are shown in fig. 5 and 6 for the blast furnace ironmaking process data. The obtained model is used for an intelligent on-line monitoring device, the key parameters of the flow industry are predicted by using the information measured in real time, and operators can adjust the flow industry system in advance.
For comparison, the same data set was model evaluated using a conventional Deep Belief Network (DBN), long and short term memory neural network (LSTM), separate Convolutional Neural Network (CNN), and gated cyclic neural network (GRU), respectively, using the mean square error MSE to evaluate the difference between the model predicted and true values, using the decision coefficient R 2 To evaluate the fit of the model. The results of the model performance evaluation are shown in table 2.
TABLE 2
From Table 2, it can be seen that the other methods, whether the fitting degree or the mean square error, are not as good as the method proposed by the present utility model, and the method proposed by the present utility model determines the coefficient R 2 Much higher than other methods. The reliability and the practicability of the calculation method provided by the utility model are verified. DBN networks are also commonly used for target variable parameter prediction in the process industry, but the method provided by the utility model is improved by 34.6% in the batch of data relative to DBN networks. LSTM network contrastThe GRU network is more complex, but it does not perform well in the process industry problem. When the GRU network or the CNN network is singly used for training test, the defects of the network structures of the GRU network and the CNN network can not sense the data dimension characteristics and the time dimension characteristics at the same time, so that the fitting result is not ideal, and after the Concate layer structure is added, the data dimension characteristics and the time dimension characteristics can be captured at the same time, so that the performance of the whole network model is improved, and more complicated process industrial conditions can be conveniently processed.
The implementation shows that the method provided by the utility model can be used for predicting the key parameters in the process industry based on the multi-feature fusion depth neural network, and has higher accuracy. This demonstrates the feasibility of on-line health monitoring and critical variable prediction in such process industries.

Claims (5)

1. A multi-feature fusion depth neural network industrial production process target data prediction method is characterized by comprising the following steps of: the method specifically comprises the following steps:
step 1, sampling industrial production processes under normal working conditions at different moments through equal time intervals to obtain production state data containing target variable parameters, and preprocessing the production state data to obtain a data set;
step 2, inputting the data set into a multi-feature fusion depth neural network to obtain a prediction result of the target variable parameter;
the method specifically comprises the following steps:
step 1, sampling industrial production processes under normal working conditions at different moments by using equal time intervals to obtain production state data, wherein the production state data are multivariate parameters obtained by detection of different sensors, the acquired production state data are respectively segmented according to the moments and the sensors to obtain time sequence data and sensor classification data which are used as two data sets of a multi-feature fusion depth neural network comprising two kinds of neural networks and are subsequently and simultaneously input into the two data sets; the production state data refer to variable parameters obtained through sensor detection, including data of target variable parameters and reference variable parameters, wherein the target variable parameters are variable parameters which cannot be obtained through the sensor direct real-time accurate detection and can only be obtained through sensor lag detection, and the reference variable parameters are variable parameters which can be obtained through the sensor direct real-time accurate detection;
step 2, the multi-feature fusion depth neural network comprises a depth convolution neural network, a depth gating circulation neural network and a multi-feature fusion layer, wherein the output ends of the depth convolution neural network and the depth gating circulation neural network are connected to the multi-feature fusion layer;
2.1, each sensor of the time series data at the same moment detects directly obtained variable parameters, takes a reference variable parameter in the time series data as input, takes a target variable parameter in the time series data as output and inputs the target variable parameter into a deep convolutional neural network together for processing, learns to obtain the relation between different variable parameters at the same moment, and obtains the relation between the target variable parameter and the variable parameter of the production state data;
2.2, setting time steps, sampling a part corresponding to the target variable parameter in the sensor classification data according to the time steps, inputting the sampled data into a single-step predicted depth gate control cyclic neural network, and learning the relation between different moments of the target variable parameter to obtain the relation between a plurality of historical moment data of the target variable parameter and the current moment data;
2.3, inputting the output prediction data of the deep convolutional neural network before the final full-connection layer and the output prediction data of the deep gate control cyclic neural network before the final full-connection layer into a multi-feature fusion layer together, and fusing by utilizing multi-feature fusion, specifically: adding and connecting two output prediction data in parallel by using an addition layer to obtain information after combining characteristics, flattening the information into one-dimensional data by using a flattening layer, and inputting the one-dimensional data into a final full-connection layer;
2.4, inputting the fusion features output by the multi-feature fusion layer into the full-connection layer of the last layer, training the deep convolutional neural network and the deep gating cyclic network by using a back propagation algorithm, adjusting weight parameters of the deep convolutional neural network and the deep gating cyclic network, and optimizing to obtain the trained deep convolutional neural network and the trained deep gating cyclic network;
and 2.5, forming a trained multi-feature fusion depth neural network by combining the trained deep convolutional neural network and the depth gate control loop network with the multi-feature fusion layer, inputting production state data of a condition to be tested into the trained multi-feature fusion depth neural network, and outputting prediction to obtain target variable parameters.
2. The method for predicting industrial process target data of the multi-feature fusion depth neural network according to claim 1, wherein the method comprises the following steps of: the deep convolution neural network captures data dimension characteristic data of the multivariate time sequence data, and output prediction data of the deep convolution neural network before the final full-connection layer is the data dimension characteristic data; the depth gating loop network captures time dimension characteristic data of target data, and output prediction data of the depth gating loop neural network before the last full connection layer is the time dimension characteristic data.
3. The method for predicting industrial process target data of the multi-feature fusion depth neural network according to claim 1, wherein the method comprises the following steps of: the deep convolutional neural network comprises a plurality of convolutional blocks and a maximum pooling layer which are sequentially connected, each convolutional block is formed by sequentially connecting the plurality of convolutional layers, time sequence data is subjected to normalization processing, then each convolutional block subjected to time domain convolution is processed, characteristics are obtained by using an activation function, the characteristics are mapped into a characteristic space, and a calculation formula is as follows:
C i ＝f(W i *C i-1 +b i )
in which W is i Shared weights representing the ith convolution block, and features C of the ith-1 convolution block i-1 Convolution, which is the sign of the convolution operation, b i Is the offset vector of the ith convolution block, f () represents the nonlinear activation function; obtaining the characteristic C of the ith convolution block by a nonlinear activation function relu i ；
After the time domain convolution processing of each convolution block, the output of the final maximum pooling layer is the data dimension characteristic data of the target characteristic space.
4. The method for predicting industrial process target data of the multi-feature fusion depth neural network according to claim 1, wherein the method comprises the following steps of: the deep gating circulating neural network is composed of three gating circulating layers which are sequentially connected, and a three-layer gating circulating network is formed.
5. The method for predicting industrial process target data of the multi-feature fusion depth neural network according to claim 4, wherein the method comprises the following steps of: the deep gating circulating neural network comprises three layers of processing modules: the first processing module comprises a first gating circulation layer, and the first gating circulation layer is sequentially transmitted to the second processing module through a reset gate and an update gate; the second processing module comprises a second gating circulation layer and a second Dropout layer, and the second gating circulation layer is sequentially transmitted to the third processing module through a reset gate and an update gate; the third processing module comprises a third gating circulation layer, and the third gating circulation layer is sequentially transmitted to the mapping layer through a reset gate and an update gate;
the first gating cycle layer, the second gating cycle layer and the third gating cycle layer all comprise gating cycle units.
CN202010744152.8A 2020-07-29 2020-07-29 Industrial production process target data prediction method of multi-feature fusion depth neural network Active CN112001527B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010744152.8A CN112001527B (en) 2020-07-29 2020-07-29 Industrial production process target data prediction method of multi-feature fusion depth neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010744152.8A CN112001527B (en) 2020-07-29 2020-07-29 Industrial production process target data prediction method of multi-feature fusion depth neural network

Publications (2)

Publication Number Publication Date
CN112001527A CN112001527A (en) 2020-11-27
CN112001527B true CN112001527B (en) 2024-01-30

Family

ID=73463963

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010744152.8A Active CN112001527B (en) 2020-07-29 2020-07-29 Industrial production process target data prediction method of multi-feature fusion depth neural network

Country Status (1)

Country Link
CN (1) CN112001527B (en)

Families Citing this family (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113009906B (en) * 2021-03-04 2022-08-02 青岛弯弓信息技术有限公司 Big data prediction analysis method and system based on industrial Internet
CN112985096B (en) * 2021-03-05 2021-09-07 广州东兆信息科技有限公司 Kiln ceramic quality monitoring system and method based on Internet of things cloud platform
CN113360851B (en) * 2021-06-22 2023-03-03 北京邮电大学 Industrial flow line production state detection method based on Gap-loss function
CN113642234A (en) * 2021-08-09 2021-11-12 贵州电网有限责任公司 Power grid icing prediction method based on multi-source characteristic time convolution deep learning
CN114015825B (en) * 2021-11-09 2022-12-06 上海交通大学 Method for monitoring abnormal state of blast furnace heat load based on attention mechanism
CN114332474A (en) * 2021-11-25 2022-04-12 中国计量大学 Method for predicting maintenance time of key part of aircraft engine
CN114216349B (en) * 2021-12-06 2022-09-16 浙江大学 Sintering end point forecasting method based on coding and decoding network
CN114519610A (en) * 2022-02-16 2022-05-20 支付宝(杭州)信息技术有限公司 Information prediction method and device
CN114638117B (en) * 2022-03-29 2022-11-08 厦门链建科技有限公司 Data processing method and system based on artificial intelligence and cloud platform
CN115457367B (en) * 2022-09-22 2023-09-26 淮阴工学院 Light-Net-based lightweight target detection method and system
CN115713987B (en) * 2022-11-17 2023-06-13 广州瑞博新材料技术研究有限公司 Polycaprolactone test data analysis method and system
CN116612836B (en) * 2023-07-19 2023-10-03 福建德尔科技股份有限公司 Tail gas amount prediction method and system for trifluoromethane production
CN117035559B (en) * 2023-10-08 2024-01-23 国网江苏省电力有限公司电力科学研究院 Electrical equipment multi-parameter transmitter simulation installation evaluation method and system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018156314A1 (en) * 2017-02-24 2018-08-30 Qualcomm Incorporated Method and apparatus for multi-dimensional sequence prediction
CN108985376A (en) * 2018-07-17 2018-12-11 东北大学 It is a kind of based on convolution-Recognition with Recurrent Neural Network rotary kiln sequence operating mode's switch method
CN111222798A (en) * 2020-01-13 2020-06-02 湖南师范大学 Soft measurement method for key indexes of complex industrial process
CN111325112A (en) * 2020-01-31 2020-06-23 贵州大学 Cutter wear state monitoring method based on depth gate control circulation unit neural network

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11544259B2 (en) * 2018-11-29 2023-01-03 Koninklijke Philips N.V. CRF-based span prediction for fine machine learning comprehension

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2018156314A1 (en) * 2017-02-24 2018-08-30 Qualcomm Incorporated Method and apparatus for multi-dimensional sequence prediction
CN108985376A (en) * 2018-07-17 2018-12-11 东北大学 It is a kind of based on convolution-Recognition with Recurrent Neural Network rotary kiln sequence operating mode's switch method
CN111222798A (en) * 2020-01-13 2020-06-02 湖南师范大学 Soft measurement method for key indexes of complex industrial process
CN111325112A (en) * 2020-01-31 2020-06-23 贵州大学 Cutter wear state monitoring method based on depth gate control circulation unit neural network

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Qiangqiang Niu等.A High-performance Web Attack Detection Method based on CNN-GRU Model.《2020 IEEE 4th Information Technology,Networking,Electronic and Automation Control Conference (ITNEC 2020)》.2020,804-808. *
深度学习在流程工业过程数据建模中的应用;袁小峰;《智能科学与技术学报》;107-115 *

Also Published As

Publication number Publication date
CN112001527A (en) 2020-11-27

Similar Documents

Publication Publication Date Title
CN112001527B (en) Industrial production process target data prediction method of multi-feature fusion depth neural network
CN110066895B (en) Stacking-based blast furnace molten iron quality interval prediction method
CN111949700B (en) Intelligent safety guarantee real-time optimization method and system for petrochemical device
CN109884892B (en) Process industrial system prediction model based on cross correlation time-lag grey correlation analysis
CN109935280B (en) Blast furnace molten iron quality prediction system and method based on ensemble learning
CN107526927B (en) Blast furnace molten iron quality online robust soft measurement method
CN105821170A (en) Soft measuring system and method for quality indexes of multielement molten iron of blast furnace
Liu et al. Model fusion and multiscale feature learning for fault diagnosis of industrial processes
Jiang et al. A multisensor cycle-supervised convolutional neural network for anomaly detection on magnetic flux leakage signals
CN114015825A (en) Method for monitoring abnormal state of blast furnace heat load based on attention mechanism
CN113761787A (en) Blast furnace molten iron silicon content online prediction method and system based on deep migration network
CN113450880A (en) Desulfurization system inlet SO2Intelligent concentration prediction method
CN116957120A (en) Device state history trend anomaly prediction method based on data analysis
CN111983140B (en) Carbon monoxide measuring system and method for dry quenching production
Liu et al. Temporal hypergraph attention network for silicon content prediction in blast furnace
CN113858566A (en) Injection molding machine energy consumption prediction method and system based on machine learning
CN113517037A (en) Method and system for predicting sintering ore FeO by fusing data and knowledge
CN111798023A (en) Method for predicting comprehensive coke ratio in steelmaking sintering production
CN108998653B (en) Intelligent heating control method for deformed steel bar
CN108709426B (en) Sintering machine air leakage fault online diagnosis method based on frequency spectrum characteristic bilateral detection method
CN114279494B (en) Fiber bragg grating blast furnace shell temperature compensation detection method based on improved LSTM
CN116305985A (en) Local intelligent ventilation method based on multi-sensor data fusion
CN113836813B (en) Blast furnace tuyere water leakage detection method based on data analysis
KR101858860B1 (en) Apparatus for controlling heat of blast furnace
CN109840722B (en) Gold hydrometallurgy whole-flow operation state online evaluation method and system

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