CN110347192B - Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder - Google Patents
Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder Download PDFInfo
- Publication number
- CN110347192B CN110347192B CN201910523327.XA CN201910523327A CN110347192B CN 110347192 B CN110347192 B CN 110347192B CN 201910523327 A CN201910523327 A CN 201910523327A CN 110347192 B CN110347192 B CN 110347192B
- Authority
- CN
- China
- Prior art keywords
- time
- encoder
- temperature
- real
- glass
- 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
- 239000011521 glass Substances 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 29
- 230000007246 mechanism Effects 0.000 title claims abstract description 27
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 claims abstract description 76
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims abstract description 38
- 239000003345 natural gas Substances 0.000 claims abstract description 38
- 239000001301 oxygen Substances 0.000 claims abstract description 38
- 229910052760 oxygen Inorganic materials 0.000 claims abstract description 38
- 238000002844 melting Methods 0.000 claims abstract description 33
- 230000008018 melting Effects 0.000 claims abstract description 33
- 238000004519 manufacturing process Methods 0.000 claims abstract description 10
- 230000014509 gene expression Effects 0.000 claims abstract description 7
- 239000013604 expression vector Substances 0.000 claims abstract description 6
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000013598 vector Substances 0.000 claims description 48
- 238000005259 measurement Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 14
- 238000003723 Smelting Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 10
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 3
- 238000005265 energy consumption Methods 0.000 abstract 1
- 239000011159 matrix material Substances 0.000 description 7
- 230000008859 change Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012549 training Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 238000007619 statistical method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 241001504469 Anthus Species 0.000 description 1
- 239000003034 coal gas Substances 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000015654 memory Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000017105 transposition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D23/00—Control of temperature
- G05D23/19—Control of temperature characterised by the use of electric means
- G05D23/1927—Control of temperature characterised by the use of electric means using a plurality of sensors
- G05D23/193—Control of temperature characterised by the use of electric means using a plurality of sensors sensing the temperaure in different places in thermal relationship with one or more spaces
- G05D23/1931—Control of temperature characterised by the use of electric means using a plurality of sensors sensing the temperaure in different places in thermal relationship with one or more spaces to control the temperature of one space
Landscapes
- Engineering & Computer Science (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Feedback Control In General (AREA)
Abstract
The invention provides an attention mechanism and self-encoder based intelligent prediction control method for glass furnace temperature, which comprises the steps of collecting production history data related to prediction control of the temperature of a pretreated glass furnace; obtaining an input variable with attention weight at each moment of each furnace temperature by adopting an attention mechanism according to the input variable obtained by preprocessing; obtaining two expression vectors comprising relative Euclidean distance and cosine similarity according to the reconstruction error of the depth self-encoder, and combining potential expressions generated by an encoder in the depth self-encoder to form final low-dimensional expression; according to the low-dimensional representation, obtaining a predicted value of the temperature of the melting furnace in a plurality of time step lengths later by adopting an LSTM prediction model; according to a control mode combining an LSTM prediction model and a statistical strategy, the natural gas flow and the oxygen flow of the glass melting furnace are intelligently adjusted on line, so that the fluctuation of the temperature of the glass melting furnace is intelligently controlled, the quality of glass products is improved, and the energy consumption is reduced.
Description
Technical Field
The invention relates to the technical field of automatic control of the temperature of a melting tank of a glass kiln, in particular to an intelligent prediction control method of the temperature of a glass furnace based on an attention mechanism and an auto-encoder.
Background
The temperature control effect of the melting tank of the glass kiln is directly related to the quality of glass products, and the rate of finished glass product rate is further influenced. The stability of the temperature control of the glass kiln is of great importance. At present, in glass production, a computer is generally adopted for controlling the temperature of a kiln, but the most common control method is still ordinary PID control (including single loop, cascade loop, range control and the like which all adopt PID as a basic control algorithm), and some improved methods comprise Smith estimation compensation, PID control, fuzzy control and the like.
Dynamic property analysis of the kiln temperature shows that the glass kiln is a variable parameter system with large inertia, large hysteresis and nonlinear characteristics in general. And the kiln can be influenced by various disturbance factors such as coal gas pressure fluctuation, feeding quality fluctuation, kiln heat preservation performance change, working environment temperature change and the like in the operation process. In view of the above characteristics of glass furnaces, the current control methods are difficult to meet the high performance requirements for furnace temperature control. If the simple PID control has poor control effect on the kiln with larger lag, the change of object parameters is difficult to adapt, and the contradiction between rapidity and overshoot exists. The design of the Smith control method is also heavily dependent on the precise model of the controlled object, and the control performance deteriorates significantly when the parameters change or the model error is large.
Disclosure of Invention
The invention aims to solve the technical problem of providing an attention mechanism and self-encoder based intelligent prediction control method for glass furnace temperature.
The invention adopts the technical scheme that the intelligent predictive control method for the glass furnace temperature based on the attention mechanism and the self-encoder comprises the following steps,
step 1, collecting production history data related to the temperature prediction control of a pretreated glass kiln;
step 2, obtaining an input variable with attention weight at each moment of each smelting furnace temperature by adopting an attention mechanism according to the input variable obtained by preprocessing;
step 3, obtaining two expression vectors including relative Euclidean distance and cosine similarity according to the reconstruction error of the depth self-encoder, and combining potential expressions generated by an encoder in the depth self-encoder to form final low-dimensional expression;
step 4, obtaining predicted values of the temperature of the melting furnace in a plurality of following time steps by adopting an LSTM prediction model according to the low-dimensional representation;
and 5, performing online intelligent adjustment on the natural gas flow and the oxygen flow of the glass melting furnace according to a control mode combining an LSTM prediction model and a statistical strategy.
In step 1, acquiring a real-time measured value of the arch top temperature of the glass melting furnace, a real-time measured value of the pool bottom temperature of the glass melting furnace, a real-time measured value of natural gas flow and a real-time measured value of oxygen flow;
carrying out min-max normalization processing on the arch top temperature of the glass melting furnace and the pool bottom temperature of the glass melting furnace;
carrying out log normalization processing on the real-time measured value of the natural gas flow and the real-time measured value of the oxygen flow;
and forming an input variable X by using the real-time measurement value of the temperature of the arch top of the smelting furnace, the real-time measurement value of the temperature of the pool bottom of the smelting furnace, the real-time measurement value of the natural gas flow and the real-time measurement value of the oxygen flow which are subjected to normalization processing at a plurality of moments.
Furthermore, step 2 is implemented in such a way that,
obtaining the input variable at time t after weighting each feature by introducing the input variable X into the attention mechanism module
Wherein,
a value representing the ith feature at time t,
represents the weight of the ith feature at the t-th time instant,
obtaining an input vector with attention weight at each time
Wherein the input variable with weighting at time t is used
And the hidden state of the LSTM at time t-1 to obtain the hidden state of the LSTM at time t.
Furthermore, the implementation of step 3 is as follows,
according to input vector
And the output vector
Extracting two expression vectors from the reconstruction error;
obtaining a representative vector z from the relative Euclidean distance1，
Obtaining a representation vector z according to the cosine similarity2，
From potential representations z of encoder output in a depth-dependent encodercAnd represents a vector z1And z2Determining the lower dimension to represent z, z ═ z (z)c,z1,z2)。
Furthermore, step 4 is implemented in such a way that,
transmitting the low-dimensional representation z into an LSTM prediction model of 3 layers to obtain a predicted value x of the next momentT+1Repeating the operation p times to obtain p predicted values (x) of time stepsT+1,xT+2,…,xT+p+1)。
Moreover, the step 5 implementation includes the following steps,
1) according to the allowable fluctuation range of the temperature of the smelting furnace required by the industrial glass production, when the temperature value at the current time T +1 is obtained through prediction of a prediction model and is abnormal, determining the maximum and minimum set values corresponding to the natural gas flow and the oxygen flow;
2) determining set values of natural gas flow and oxygen flow to be adjusted;
comparing the real-time values of the natural gas flow and the oxygen flow at the time T +1 according to the maximum and minimum set values of the natural gas flow and the oxygen flow at the time T +1, adjusting the following steps,
wherein OPT+1Real-time values of the natural gas flow rate and the oxygen flow rate at time T +1 are shown,
and
to representThe maximum set value and the minimum set value within the process allowable range of the natural gas flow and the oxygen flow counted at the T +1 moment are obtained.
The difference between the invention and the prior art and the corresponding technical effects are as follows: the high-dimensional degree of an input vector and two data characteristics of a time sequence in the temperature control of a glass kiln are not considered in the prior art, the characteristic that the input vector contains a continuous time sequence is firstly considered, and the input vector can be selectively gathered in certain main characteristics at each moment through an attention force mechanism instead of the traditional consideration of all the characteristics at each moment; secondly, the problem of high dimension of the input vector is considered, and if the input vector is directly taken for model training, the problem of dimension disaster is caused, so that dimension reduction processing is carried out on the input vector through a depth self-encoder to obtain low-dimension representation with important information of the input vector, and then the kiln temperature is predicted in a low-dimension space; finally, the invention realizes the intelligent control of the kiln temperature by a method combining prediction, feedback and statistics. By the technology, the intelligent prediction and dynamic control of the temperature of the glass kiln can be realized, so that the quality of glass is improved. The invention is particularly suitable for large-scale glass manufacturing and has important market value.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is explained in detail in the following by combining the drawings and the embodiment.
The embodiment of the invention provides an attention mechanism and self-encoder based intelligent prediction control method for glass furnace temperature, which comprises the following steps:
step (1), data preprocessing: production history data related to predictive control of the temperature of the pretreated glass kiln is collected.
Step (1.1) historical data acquisition of glass melting furnaces
The invention collects production information related to the temperature control of the glass furnace, because more than 1000 sensors are deployed on the glass furnace for detecting and adjusting variable indexes, after analysis, the required characteristic data of the invention mainly comprises: the real-time measurement value of the arch top temperature of the glass melting furnace, the real-time measurement value of the pool bottom temperature of the glass melting furnace, the real-time measurement value of the natural gas flow and the real-time measurement value of the oxygen flow;
step (1.2) of normalization of characteristics
Filling and aligning data per minute on the collected real-time measured value of the arch top temperature of the glass melting furnace, the real-time measured value of the pool bottom temperature of the glass melting furnace, the real-time measured value of the natural gas flow and the real-time measured value of the oxygen flow to obtain four furnace characteristics with the same time interval; then, carrying out statistical analysis on the characteristics of the melting furnace to obtain respective maximum value and minimum value; because the dimensions of different features in the data are different, the data needs to be normalized:
in an embodiment of the present invention,
carrying out statistical analysis on the collected real-time measured value of the arch top temperature of the glass melting furnace, the real-time measured value of the pool bottom temperature of the glass melting furnace, the real-time measured value of the natural gas flow and the real-time measured value of the oxygen flow to obtain the respective maximum value and minimum value; because the dimensions of different features in the data are different, the data needs to be normalized:
carrying out min-max normalization on the real-time measured value of the arch top temperature of the glass melting furnace and the real-time measured value of the pool bottom temperature of the glass melting furnace, wherein the specific formula is as follows:
wherein x represents a real-time measurement value of the arch top temperature of the melting furnace or a real-time measurement value of the pool bottom temperature of the glass melting furnace, xminRepresents the minimum value, x, of the featuremaxMaximum value, x, representing the characteristic*Represents the value of the feature after min-max normalization;
carrying out logarithmic normalization on the real-time measured values of the natural gas flow and the oxygen flow, wherein the specific formula is as follows:
y*＝log(y+1)
wherein y represents a real-time measurement of natural gas flow or oxygen flow, log represents a logarithm based on a natural number e, y*Representing the value of the feature after logarithmic normalization;
step (1.3) of obtaining an input sequence
Forming an input sequence X by data at a plurality of moments, wherein the data at each moment comprises a real-time measured value of the arch top temperature of the smelting furnace, a real-time measured value of the pool bottom temperature of the smelting furnace, a real-time measured value of the natural gas flow and a real-time measured value of the oxygen flow which are subjected to normalization processing at the moment:
wherein,
representing a real number, n representing the number of features in the input sequence, n in the embodiment being 4, T representing the size of the time window, suggested values of 3, 5, 10 or 15,
representing transposition, so that the input sequence X is a two-dimensional vector consisting of n features, each feature consisting of data at T instants;
by using
K represents the kth feature vector with the time window size T, k represents the kth feature of the input sequence X, k takes one of {1,2, …, n },
a value representing the 1 st time instant of the kth feature; and use
To representA vector consisting of n features at time T, wherein the variable T represents time T, and T is 1,2, … and T;
determining features with attention weights at each moment of the attention mechanism;
in an embodiment, the data X for time t can be obtained by inputting the input vector X into the attention mechanism moduletNew input variables with each feature weighted
Wherein
Represents the weight of the ith feature at the t-th time, an 
Values representing the ith characteristic at time t, using the input variables at time t with weighting
And the hidden layer state of the LSTM at the t-1 moment to obtain the hidden layer state of the LSTM at the t moment, namely receiving the back propagation of the LSTM prediction model.
In the embodiment, the specific implementation process of step 2 is as follows:
setting initial values of related variables in the attention mechanism, wherein the initial values specifically comprise the following steps:
the weight matrix W is a matrix of weights,
m represents the number of hidden nodes, and the suggested value is 16, 32, 64, 128 or 256;
the weight matrix U is a matrix of weights,
the weight matrix V is a matrix of weights,
let t be 1 and t be equal to 1,
t-1 hidden state S of LSTMt-1，
Where N denotes a batch size, i.e. each cycle is trained using N input sequences, e.g. X1＝(x1,x2,…,xT),X2＝(x2,x3,…,xT+1),…,XN＝(xN,xN+1,…,xN+T) Wherein X is1Representing the first input sequence, …, XNRepresenting the Nth input sequence, and suggesting N to be 16, 32, 64, 128, 256 or 512 according to the size of the data set;
hidden state h of LSTM at t-1 momentt-1，
Wherein LSTM represents Long Short-Term Memory, i.e. Long Short-Term Memory neural network element.
In particular, these variables may be pre-set as hyper-parameters of the model, for example, based on a feature analysis of the data set.
Step (2.2) of obtaining an input vector of the features with weights at time t
The specific process is as follows:
step (2.2.1) of setting the kth feature of the input vector X and obtaining the attention initial weight of the kth feature at the time t
The specific formula is as follows:
wherein
Represents the transpose of the weight matrix V, [ h ]t-1；St-1]Means to splice together two hidden states of LSTM, xkA kth feature representing the input vector X; k is one of {1,2, …, n }.
In this way, the attention initial weights of all features can be obtained
And (2.2.2) performing normalization processing (softmax) on the obtained attention initial weights of the n features at the time t, wherein for the kth feature of the input vector X, the normalization processing is specifically as follows:
wherein exp represents an exponential function with a natural constant e as the base,
represents the attention weight of the kth feature at time t, an
Thus, can obtain
And is
Step (2.2.3) use of attention weight
Updating an input vector xtFor new input vectors with attention weights
The specific formula is as follows:
wherein
Representing the input vector x at time ttThe ith feature of (1);
step (2.2.4): updating hidden state S of LSTM at time ttAnd htThe concrete formula is as follows:
wherein LSTM () represents the calling of long-short term memory neural network elements;
step (2.3): and (3) returning to execute the step (2.2) by taking T as T +1 until T is T, and iteratively executing the step (2.2) for T times to obtain an input vector with attention weight at each moment
Based on the attention mechanism, the new input vector can selectively focus on a certain feature, rather than looking at the same for all input feature sequences;
and (3): potential representation determination with important information based on depth autoencoder: obtaining two representation vectors according to the reconstruction error of the depth self-encoder, and combining potential representations generated by the encoder in the depth self-encoder to form a final low-dimensional representation;
attention-attracting mechanism module output
The vector is transmitted into a depth self-encoder to obtain a reduced-dimension vector with input
Is potentially represented by zcThe specific process is as follows:
output vector of attention mechanism module
After passing through the encoder of the depth self-encoder, a reduced-dimension potential representation z with important information is obtainedcWherein the encoder is a 3-layer fully-connected neural network;
output potential representation z of the encodercAfter passing through the decoder of the depth self-encoder, a reconstructed input vector is obtained
The purpose of the depth self-encoder is to obtain a sum input vector
Reconstructed vectors as similar as possible
The output of such an encoder potentially represents zcA large amount of important information can be carried;
in the present invention, the decoder and encoder are symmetrical structures, for example, the encoder is 3 layers, where the number of nodes in each layer is (120,60,1), then the decoder is (1,60,120), where 1 is common, i.e. the output of the encoder is the input of the decoder.
And (4): determination of the low-dimensional representation z;
step (4.1): according to the input vector in the step (3)
And the output vector
Extracting two expression vectors from the reconstruction error;
obtaining a representative vector z from the relative Euclidean distance1The concrete formula is as follows:
wherein | · |)2Expressed is a 2-norm;
obtaining a representation vector z according to the cosine similarity2The concrete formula is as follows:
step (4.2): according to the potential representation z obtained in step (2)cAnd step (4.1) obtaining a representative vector z1And z2And determining the low-dimensional expression z by splicing, wherein the specific formula is as follows:
z＝z(zc,z1,z2)
note that: z is a radical ofc,z1,z2Splicing according to the last dimension; namely, assume that
The three vectors are now spliced according to the last dimension (here the second dimension) to obtain
And (5): LSTM prediction model: according to the low-dimensional representation, obtaining a predicted value of the temperature of the smelting furnace of p time steps later by adopting an LSTM prediction model;
transmitting the low-dimensional representation z obtained in the step (4) into an LSTM prediction model of 3 layers, and obtaining a predicted value x of the next momentT+1Repeating the step p times (p represents a cyclic variable and the suggested value is less than 10), and obtaining the predicted values (x) of p time stepsT+1,xT+2,…,xT+p+1)；
And (6): determining an objective function of the model:
where N represents a batch size, j represents a loop variable,
representing the reconstruction error of the depth autocoder, L (x)T+1,x′T+1) Representing the next time prediction x obtained by the LSTM prediction modelT+1(e.g., next moment furnace temperature value) and actual observed value x'T+1An error of (2); and the model is back propagated through Adam;
and (7): intelligent control of furnace temperature through a combination of predictive models and statistical strategies: and (3) carrying out online intelligent adjustment on the natural gas flow and the oxygen flow of the glass melting furnace according to a control method combining an LSTM prediction model and a statistical strategy.
The embodiment is realized as follows:
step (7.1): according to the allowable fluctuation range (such as fluctuation not greater than plus or minus 2 ℃) of the furnace temperature required by industrial glass production, appointing the maximum and minimum set values corresponding to the arch top temperature and the pool bottom temperature of the furnace; when the predicted value of the temperature of the melting furnace output in the step (5) is within the range of the set maximum and minimum set values, the current temperature of the melting furnace is well controlled; when the predicted value of the furnace temperature output in the step (5) is out of the range of the set maximum and minimum set values, the furnace temperature value obtained at the current moment is abnormal, and the switching values of the related natural gas flow and oxygen flow need to be adjusted, so that the temperature is restored to be in the range of the set values.
Assuming that the temperature value at the current time T +1 is not normal through prediction of the prediction model, but the temperature value at the previous time T is normal, therefore, the real-time values of the natural gas flow and the oxygen flow corresponding to the previous time T are searched, and the maximum and minimum set values corresponding to the natural gas flow and the oxygen flow at the time T +1 are finally determined according to the allowable fluctuation ranges (such as plus or minus 0.5) of the natural gas flow and the oxygen flow opening of the glass process. Adjusting the parameter value of the corresponding control quantity by judging the real natural gas flow and oxygen flow opening at the time of T +1 and the corresponding maximum and minimum set values, thereby realizing the normal temperature at the time of T + 1;
step (7.2): determining set values of natural gas flow and oxygen flow to be adjusted;
obtaining the maximum and minimum set values of the natural gas flow and the oxygen flow at the time T +1 according to the step (7.1), and comparing the real-time values of the natural gas flow and the oxygen flow at the time T +1 to adjust the values as follows:
wherein OPT+1Real-time values of the natural gas flow rate and the oxygen flow rate at time T +1 are shown,
and
the maximum set value and the minimum set value within the process allowable range of the natural gas flow and the oxygen flow obtained by statistics at the T +1 moment are shown;
in specific implementation, the automatic operation of the above processes can be realized by adopting a computer software technology, and a device for operating the processes also should be within the protection scope of the present invention.
The invention has the following advantages: by means of the attention mechanism module, the input vector can be selectively focused on a certain feature, and not all input feature sequences are viewed identically; the potential representation of the input vector can be obtained through the depth self-encoder, the potential representation is a low-dimensional representation with important characteristic information of the input vector, and for data with excessive number of such characteristics of a melting furnace, the low-dimensional representation can save a large amount of model training time; finally, the performance of the attention mechanism module, the depth self-encoder and the LSTM prediction model can be effectively enhanced through end-to-end joint training, and meanwhile, potential representations with a large amount of important information can be learned through the reconstruction error of the self-encoder which is as low as possible.
While specific embodiments of the invention have been described, it will be understood by those skilled in the art that the specific embodiments described are illustrative only and are not limiting to the scope of the invention, and that equivalent modifications and variations can be made by those skilled in the art without departing from the spirit of the invention, which is to be limited only by the claims appended hereto.
It should be emphasized that the described embodiments of the present invention are illustrative and not restrictive. Therefore, the present invention includes, but is not limited to, the examples described in the detailed description, and all other embodiments that can be derived from the technical solutions of the present invention by those skilled in the art also belong to the protection scope of the present invention.
Claims (6)
1. An attention mechanism and self-encoder based intelligent prediction control method for glass furnace temperature is characterized in that: comprises the following steps of (a) carrying out,
step 1, collecting production history data related to the temperature prediction control of a pretreated glass kiln;
step 2, obtaining an input variable with attention weight at each moment of each smelting furnace temperature by adopting an attention mechanism according to the input variable obtained by preprocessing;
step 3, obtaining two expression vectors including relative Euclidean distance and cosine similarity according to the reconstruction error of the depth self-encoder, and combining potential expressions generated by an encoder in the depth self-encoder to form final low-dimensional expression;
step 4, obtaining predicted values of the temperature of the melting furnace in a plurality of following time steps by adopting an LSTM prediction model according to the low-dimensional representation;
and 5, performing online intelligent adjustment on the natural gas flow and the oxygen flow of the glass melting furnace according to a control mode combining an LSTM prediction model and a statistical strategy.
2. The intelligent predictive control method for glass furnace temperature based on attention mechanism and self-encoder as claimed in claim 1, characterized in that: step 1, acquiring a real-time measured value of the arch top temperature of the glass melting furnace, a real-time measured value of the pool bottom temperature of the glass melting furnace, a real-time measured value of natural gas flow and a real-time measured value of oxygen flow;
carrying out min-max normalization processing on the arch top temperature of the glass melting furnace and the pool bottom temperature of the glass melting furnace;
carrying out log normalization processing on the real-time measured value of the natural gas flow and the real-time measured value of the oxygen flow;
and forming an input variable X by using the real-time measurement value of the temperature of the arch top of the smelting furnace, the real-time measurement value of the temperature of the pool bottom of the smelting furnace, the real-time measurement value of the natural gas flow and the real-time measurement value of the oxygen flow which are subjected to normalization processing at a plurality of moments.
3. The intelligent predictive control method for glass furnace temperature based on attention mechanism and self-encoder as claimed in claim 2, characterized in that: the step 2 is realized in a mode that,
obtaining the input variable at time t after weighting each feature by introducing the input variable X into the attention mechanism module
Wherein,
a value representing the ith feature at time t,
represents the weight of the ith feature at the t-th time instant,
obtaining an input vector with attention weight at each time
Wherein the input variable with weighting at time t is used
And the concealment of the LSTM at the time t-1The layer state is used to obtain the hidden layer state of LSTM at t.
4. The intelligent predictive control method for glass furnace temperature based on attention mechanism and self-encoder as claimed in claim 1, characterized in that: the implementation of step 3 is as follows,
according to input vector
And the output vector
Extracting two expression vectors from the reconstruction error;
obtaining a representative vector z from the relative Euclidean distance1，
Obtaining a representation vector z according to the cosine similarity2，
From potential representations z of encoder output in a depth-dependent encodercAnd represents a vector z1And z2Determining the lower dimension to represent z, z ═ z (z)c,z1,z2)。
5. The intelligent predictive control method for glass furnace temperature based on attention mechanism and self-encoder as claimed in claim 4, characterized in that: the step 4 is realized in a way that,
transmitting the low-dimensional representation z into an LSTM prediction model of 3 layers to obtain a predicted value x of the next momentT+1Repeating the operation p times to obtain p predicted values (x) of time stepsT+1,xT+2,…,xT+p+1)。
6. The intelligent predictive control method for glass furnace temperature based on attention mechanism and self-encoder as claimed in claim 5, characterized in that: the step 5 implementation includes the following steps,
1) according to the allowable fluctuation range of the temperature of the smelting furnace required by the industrial glass production, when the temperature value at the current time T +1 is obtained through prediction of a prediction model and is abnormal, determining the maximum and minimum set values corresponding to the natural gas flow and the oxygen flow;
2) determining set values of natural gas flow and oxygen flow to be adjusted;
comparing the real-time values of the natural gas flow and the oxygen flow at the time T +1 according to the maximum and minimum set values of the natural gas flow and the oxygen flow at the time T +1, adjusting the following steps,
wherein OPT+1Real-time values of the natural gas flow rate and the oxygen flow rate at time T +1 are shown,
and
the maximum set value and the minimum set value within the process allowable range of the natural gas flow and the oxygen flow counted at the time T +1 are shown.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910523327.XA CN110347192B (en) | 2019-06-17 | 2019-06-17 | Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910523327.XA CN110347192B (en) | 2019-06-17 | 2019-06-17 | Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110347192A CN110347192A (en) | 2019-10-18 |
| CN110347192B true CN110347192B (en) | 2020-11-03 |
Family
ID=68182127
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910523327.XA Active CN110347192B (en) | 2019-06-17 | 2019-06-17 | Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110347192B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111815561B (en) * | 2020-06-09 | 2024-04-16 | 中海石油(中国)有限公司 | Pipeline defect and pipeline assembly detection method based on depth space-time characteristics |
| CN112279507B (en) * | 2020-10-23 | 2021-09-17 | 南京玻璃纤维研究设计院有限公司 | Temperature control method and device, electronic equipment and system |
| CN112817289A (en) * | 2021-02-09 | 2021-05-18 | 上海凯盛朗坤信息技术股份有限公司 | Glass factory data analysis and intelligent prediction system |
| CN113190974B (en) * | 2021-04-09 | 2022-06-21 | 广东工业大学 | Roller kiln temperature field multipoint prediction method based on deep learning |
| CN116909233B (en) * | 2023-07-28 | 2024-05-07 | 安徽华容药品包装有限公司 | Method and system for producing medicinal glass bottle |
| CN117132170B (en) * | 2023-10-24 | 2024-01-26 | 苏州科尔珀恩机械科技有限公司 | Intelligent prediction method for production energy consumption of digital kiln |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324862A (en) * | 2013-07-11 | 2013-09-25 | 中国石油大学(华东) | Coal-fired boiler optimization method based on improved neural network and genetic algorithm |
| CN106865958A (en) * | 2017-02-28 | 2017-06-20 | 华中科技大学 | A kind of glass melter combustion flame optimal control method |
| CN108604310A (en) * | 2015-12-31 | 2018-09-28 | 威拓股份有限公司 | Method, controller and system for controlling a dispensing system using a neural network architecture |
| CN109285562A (en) * | 2018-09-28 | 2019-01-29 | 东南大学 | Speech-emotion recognition method based on attention mechanism |
| CN109765906A (en) * | 2019-03-04 | 2019-05-17 | 武汉理工大学 | A kind of intelligent ship tracking method based on Compound Orthogonal Neural Network PREDICTIVE CONTROL |
| CN109829396A (en) * | 2019-01-16 | 2019-05-31 | 广州杰赛科技股份有限公司 | Recognition of face motion blur processing method, device, equipment and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11327475B2 (en) * | 2016-05-09 | 2022-05-10 | Strong Force Iot Portfolio 2016, Llc | Methods and systems for intelligent collection and analysis of vehicle data |
| CN106773682B (en) * | 2016-12-05 | 2019-11-22 | 清华大学 | The glass furnace bottom of pond temperature intelligent forecast Control Algorithm being dynamically determined based on time lag |
| US10474709B2 (en) * | 2017-04-14 | 2019-11-12 | Salesforce.Com, Inc. | Deep reinforced model for abstractive summarization |
| CN109685246B (en) * | 2018-11-13 | 2024-04-23 | 平安科技（深圳）有限公司 | Environment data prediction method and device, storage medium and server |
| CN109740148B (en) * | 2018-12-16 | 2023-12-22 | 北京工业大学 | Text emotion analysis method combining BiLSTM with Attention mechanism |
-
2019
- 2019-06-17 CN CN201910523327.XA patent/CN110347192B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324862A (en) * | 2013-07-11 | 2013-09-25 | 中国石油大学(华东) | Coal-fired boiler optimization method based on improved neural network and genetic algorithm |
| CN108604310A (en) * | 2015-12-31 | 2018-09-28 | 威拓股份有限公司 | Method, controller and system for controlling a dispensing system using a neural network architecture |
| CN106865958A (en) * | 2017-02-28 | 2017-06-20 | 华中科技大学 | A kind of glass melter combustion flame optimal control method |
| CN109285562A (en) * | 2018-09-28 | 2019-01-29 | 东南大学 | Speech-emotion recognition method based on attention mechanism |
| CN109829396A (en) * | 2019-01-16 | 2019-05-31 | 广州杰赛科技股份有限公司 | Recognition of face motion blur processing method, device, equipment and storage medium |
| CN109765906A (en) * | 2019-03-04 | 2019-05-17 | 武汉理工大学 | A kind of intelligent ship tracking method based on Compound Orthogonal Neural Network PREDICTIVE CONTROL |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110347192A (en) | 2019-10-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110347192B (en) | Glass furnace temperature intelligent prediction control method based on attention mechanism and self-encoder | |
| KR102267180B1 (en) | Air conditioning system and controlling method thereof | |
| WO2022121944A1 (en) | Method and apparatus for intelligent forecasting of industrial process operation indicators, device, and storage medium | |
| CN110598929B (en) | Wind power nonparametric probability interval ultrashort term prediction method | |
| CN110689171A (en) | Turbine health state prediction method based on E-LSTM | |
| CN111045326A (en) | Tobacco shred drying process moisture prediction control method and system based on recurrent neural network | |
| CN113988449B (en) | Wind power prediction method based on transducer model | |
| CN110097929A (en) | A kind of blast furnace molten iron silicon content on-line prediction method | |
| CN112288137A (en) | LSTM short-term load prediction method and device considering electricity price and Attention mechanism | |
| Suryo et al. | Improved time series prediction using LSTM neural network for smart agriculture application | |
| CN117784852B (en) | Multi-mode sensor temperature control method based on fish scale bionic optimization algorithm | |
| CN116562454A (en) | Manufacturing cost prediction method applied to BIM long-short-time attention mechanism network | |
| CN101285816A (en) | Copper matte air refining procedure parameter soft sensing instrument and its soft sensing method | |
| CN113204918B (en) | Modeling method for cement kiln production energy consumption | |
| Marenbach et al. | Data-driven structured modelling of a biotechnological fed-batch fermentation by means of genetic programming | |
| CN117034755B (en) | Cold-rolled steel mechanical property prediction method integrating multi-head attention mechanism | |
| CN113177355A (en) | Power load prediction method | |
| CN118331210A (en) | Intelligent optimization method for coking production process based on knowledge distillation | |
| CN113962432B (en) | Wind power prediction method and system integrating three-dimensional convolution and lightweight convolution threshold unit | |
| Liu et al. | A PSO-RBF neural network for BOD multi-step prediction in wastewater treatment process | |
| CN115860232A (en) | Steam load prediction method, system, electronic device and medium | |
| Lu et al. | A New Power Load Forecasting Model (SIndRNN): independently recurrent neural network based on softmax kernel function | |
| CN116954086B (en) | Intelligent prediction control method and device for adjusting system of pumped storage unit | |
| CN117132170B (en) | Intelligent prediction method for production energy consumption of digital kiln | |
| CN115879569B (en) | Online learning method and system for IoT observation data |
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 |