CN112163376B - Extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search - Google Patents
Extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search Download PDFInfo
- Publication number
- CN112163376B CN112163376B CN202011072976.1A CN202011072976A CN112163376B CN 112163376 B CN112163376 B CN 112163376B CN 202011072976 A CN202011072976 A CN 202011072976A CN 112163376 B CN112163376 B CN 112163376B
- Authority
- CN
- China
- Prior art keywords
- iteration
- whisker
- samples
- prediction
- training
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 241000254173 Coleoptera Species 0.000 title claims abstract description 34
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 24
- 238000002485 combustion reaction Methods 0.000 claims abstract description 16
- 238000010845 search algorithm Methods 0.000 claims abstract description 12
- 230000000694 effects Effects 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 33
- 238000012360 testing method Methods 0.000 claims description 21
- 241001481710 Cerambycidae Species 0.000 claims description 10
- 238000002790 cross-validation Methods 0.000 claims description 9
- 238000003066 decision tree Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 8
- 239000007789 gas Substances 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 238000011156 evaluation Methods 0.000 claims description 5
- 238000005096 rolling process Methods 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 4
- 230000009191 jumping Effects 0.000 claims description 4
- QVGXLLKOCUKJST-UHFFFAOYSA-N atomic oxygen Chemical compound [O] QVGXLLKOCUKJST-UHFFFAOYSA-N 0.000 claims description 3
- 229910052760 oxygen Inorganic materials 0.000 claims description 3
- 239000001301 oxygen Substances 0.000 claims description 3
- 238000013139 quantization Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 238000004088 simulation Methods 0.000 claims description 2
- 238000004364 calculation method Methods 0.000 claims 1
- 230000007246 mechanism Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 12
- 230000009286 beneficial effect Effects 0.000 description 4
- 239000012855 volatile organic compound Substances 0.000 description 4
- 230000008859 change Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 2
- 239000010815 organic waste Substances 0.000 description 2
- 239000007800 oxidant agent Substances 0.000 description 2
- 230000001172 regenerating effect Effects 0.000 description 2
- 239000002912 waste gas Substances 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000003776 cleavage reaction Methods 0.000 description 1
- 238000011217 control strategy Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000004043 dyeing Methods 0.000 description 1
- 238000003933 environmental pollution control Methods 0.000 description 1
- 238000009776 industrial production Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000003345 natural gas Substances 0.000 description 1
- 230000001590 oxidative effect Effects 0.000 description 1
- 239000003973 paint Substances 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000007017 scission Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/08—Thermal analysis or thermal optimisation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Feedback Control In General (AREA)
Abstract
An extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search belongs to the field of industrial combustion process furnace temperature prediction and control. The method is used for obtaining the predicted value of the furnace temperature in the combustion chamber in real time by establishing a regression relation between the key variable affecting the temperature and the temperature of the combustion chamber, so that the prediction accuracy of the furnace temperature reaches +/-1.5 ℃. Considering that the control accuracy of the original system is not high, designing a longhorn beetle whisker search algorithm, selecting a secondary performance index function commonly used in predictive control as an adaptability function of longhorn beetles, and searching to obtain an optimal control quantity through an olfactory search mechanism of the algorithm, so that the control effect of the whole system is better.
Description
Technical Field
The invention belongs to the field of furnace temperature prediction and control in an industrial combustion process, and particularly relates to an extreme random tree prediction control method based on longhorn beetle whisker search.
Background
With the increasingly stricter environmental pollution control, effective treatment of low-concentration volatile organic compounds (Volatile Organic Compounds, VOCs) from industrial production emissions of petrochemical industry, papermaking, paint, printing and dyeing and the like is receiving high attention in China. The most common equipment for treating VOCs in China at the present stage is a regenerative oxidizing furnace (Regenerative Thermal Oxidizer, RTO). While conventional RTOs employ simple PID control of the temperature within the combustion chamber, the oxidizer is essentially a large inertia, large hysteresis, and multivariable nonlinear system, and temperature fluctuations have a significant impact on the conversion of the organic exhaust gas combustion reaction as conditions within the combustion chamber change, thus requiring precise control of the furnace temperature within the combustion chamber.
The predictive control is widely applied to nonlinear systems due to good control effect and strong robustness. The main idea is to model the system, predict the future output state of the system, compare with the reference input after feedback correction of the predicted value, apply quadratic performance index to perform rolling optimization to obtain the optimal control amount, and complete the whole cycle. The classical predictive control algorithm generally builds a model of the system by obtaining step or impulse response parameters of the system to obtain an approximately linear model, and for a system with more nonlinearity and interference, the problem of lower accuracy of the system model exists.
In recent years, the method of machine learning has been rapidly developed and widely used in model prediction. The key parameter in the combustion control system is temperature, and accurate prediction and control of temperature are important indexes. The extreme random tree (ET) algorithm has good generalization and nonlinear modeling capability, and can flexibly process various types of data.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a combustion chamber furnace temperature prediction method based on an extreme stochastic tree algorithm, and a regression relation between key variables influencing temperature and combustion chamber temperature is established, so that a prediction value of the combustion chamber furnace temperature is obtained in real time, and the prediction accuracy of the furnace temperature reaches +/-1.5 ℃. Considering that the control accuracy of the original system is not high, designing a longhorn beetle whisker search (Beetle Antennae Search, BAS) algorithm, selecting a secondary performance index function commonly used in predictive control as an adaptability function of longhorn beetles, and searching to obtain the optimal control quantity through an olfactory search mechanism of the algorithm, so that the control effect of the whole system is better.
The technical scheme adopted by the invention is as follows: the furnace temperature prediction part adopts an extreme random tree model prediction algorithm, the control quantity optimization part adopts a longhorn beetle whisker search algorithm, and the method mainly comprises the following implementation steps:
step 1, collecting control input and output data of past and current moments in the combustion process as an original data set O. The acquired data comprises: waste gas concentration, oxygen content, air quantity, proportional valve opening and temperature;
step 2, dividing the original data set O into a training set N by adopting a cross-validation mode 1 And test set N 2 ；
Step 3, training all training sets by using an extreme random tree ET algorithm, and establishing a regression prediction model;
and 4, carrying out simulation prediction on the test set by using the established regression prediction model to obtain a predicted value of the furnace temperature, and introducing a feedback correction link to correct the predicted furnace temperature in order to overcome the deviation between the predicted output and the actual output.
And 5, taking the set temperature of the hearth, the corrected predicted temperature and the historical control quantity as inputs, and solving an objective function J according to a longhorn beetle whisker search algorithm to obtain an optimal gas control quantity u (k), so as to realize rolling optimization of the control quantity.
The invention has the beneficial effects that: according to the invention, the nonlinear relation between the furnace temperature and the valve opening in the organic waste gas combustion process is modeled based on an extreme stochastic tree algorithm, so that the furnace temperature prediction is realized, the control variable is subjected to online rolling optimization through a longhorn beetle whisker search algorithm, and a method is provided for the prediction and control of the furnace temperature in the organic waste gas combustion process.
Based on the above technical solution, the solution will be described in further detail.
The specific steps of the cross-validation method described in the step 2 are as follows:
step 2.1, the original data set O is sampled through k times of layering to obtain k mutually exclusive subsets with similar sizes,
step 2.2 uses the union of k-1 mutually exclusive subsets at a time as training set and the remaining subset as test set. K groups of training sets and test sets are obtained, so that k times of modeling is performed, and finally, the average value of the test result is output.
The beneficial effect of above-mentioned scheme is: by hierarchical sampling, each subset D i The consistency of the data distribution is maintained as much as possible. The data set is divided into a training set and a testing set by setting the proportion of the testing set in the training set and the seeds of the random tree by adopting a cross-validation method based on a random function. The randomness of the training set and the testing set is guaranteed, the accuracy of training and prediction can be improved, over-fitting and under-fitting are avoided, and the reliability of the temperature predicted value is guaranteed.
Further, the ET algorithm training process described in step 3 is as follows:
for the acquired dataset Ω= { (a) 1 ,y 1 ),(A 2 ,y 2 )…(A i ,y i )…(A N ,y N ) }, wherein A i Is a row vector of dimension 1 x 5, is input for modeling samples, y i Is A i The real output values of the corresponding samples, i=1, 2 … N, N are the sample group numbers, and training is performed on the data based on an extreme random tree algorithm, wherein the specific training process is as follows:
step 3.1 a training data set is given. N is selected from the acquired data set omega by a cross-validation method 1 Omega corresponding to group data 1 As a training set, the data set is trained. Setting a maximum iteration number T, preferably 100, and initializing the iteration number t=1;
step 3.2 random partitioning of the node dataset. For the t-th iteration, from training set N 1 Sigma feature sequences { a }, are randomly selected 1 ,a 2 ,…a σ As the initial node, and randomly selecting a corresponding set of sequences { s } from the selected features 1 ,s 2 ,…,s σ As a threshold for the attribute of the feature, the threshold has a value ranging between the maximum and minimum values of the feature. A in the feature sequence x Is greater than s x Samples of x=1, 2, … and sigma are classified into left nodes of branches, and other samples are classified into right nodes, so that random division under the nodes is realized;
step 3.3, obtaining the optimal splitting value under the random characteristic. For the numerical characteristics of the system, the root mean square error is used as a quantization evaluation criterion. Traversing all the characteristic values on the left node and the right node, calculating the root mean square error of each characteristic value on the corresponding node, and selecting the characteristic value corresponding to the minimum root mean square error as a splitting point. Characterised by feature a i ∈{a 1 ,a 2 ,…a σ For example, after division in step 3.2, the samples on the left and right sides of the feature are a respectively il ＝(α 1 ,α 2 ,…,α i ,…,α p )，a ir ＝(β 1 ,β 2 ,…,β j ,…,β q ) Wherein: alpha i For samples assigned to the left, i=1, 2, … p; beta j For the samples assigned to the right, j=1, 2, … q; p and q are the left and right sample numbers, respectively. Taking the left node as an example, the Root Mean Square Error (RMSE) corresponding to each characteristic value i The method comprises the following steps:
wherein the method comprises the steps ofSelecting the feature with the minimum root mean square error for splitting;
and 3.4, judging whether a splitting end condition is met, if so, jumping to the step 3.5, otherwise, updating the iteration times, executing t=t+1, and repeating the steps 3.2-3.3. The conditions for ending the cleavage are as follows (any one of the conditions is satisfied):
(1) The maximum depth (max_depth) of the tree is preferably 50;
(2) The minimum number of samples required for internal node subdivision (min_samples_split), preferably 2;
(3) The minimum number of samples of leaf nodes (min_samples_leaf), preferably 1.
Step 3.5, after the splitting is finished, obtaining a decision tree f obtained by the t-th iteration t 。
Step 3.6 if t=t goes to step 3.7, otherwise t=t+1 goes to step 3.2;
step 3.7, combining the generated T decision trees by using a mean method to obtain a prediction modelThe method comprises the following steps: />
Inputting the test set into the prediction model to obtain prediction outputN 2 Is the number of groups of the test set, and the corresponding evaluation index is used for measuring the prediction effect.
The beneficial effect of above-mentioned scheme is: because the characteristics of the algorithm are randomly selected and the splitting threshold value is also randomly set, all samples are adopted for the establishment of each decision tree, so that the randomness of training based on the extremely random tree is stronger, the prediction effect of the trained regression prediction model is better, and the precision of the obtained furnace temperature prediction value is higher.
The longhorn beetle whisker search algorithm in the step 5 comprises the following specific steps:
step 5.1 gives the corresponding objective function:
wherein: t (T) r Is input for a reference terminal; t (T) p Outputting the corrected prediction; u is a control variable; λ is a weight coefficient (0 < λ < 1); d is the predicted number of steps.
Defining e as a model error vector:
e＝[T r (k+1)-T p (k+1),…,T r (k+d)-T p (k+d)] T
define deltau as the increment of the control quantity:
Δu＝[u(k)-u(k-1),…,u(k+d-1)-u(k+d-2)] T
according to the two formulas above, f is defined as:
f＝[e T λΔu T ] T 。
screening a d-dimensional control vector u * Let objective function J reach minimum, can obtain J and be the mathematical formula of minimum through the definition formula of f: minj=f T ×f。
And 5.2, reading data after the ET algorithm is executed at the last moment, substituting the data into an objective function, and initializing parameters of the longhorn beetle whisker search algorithm. Mainly comprises the following steps: the dimension of the problem is D, the distance between two whiskers is D 0 The initial step length is delta, and the attenuation coefficient of the distance between two whiskers is eta_D 0 The attenuation coefficient of the step length is eta delta, the iteration number is n, and the mass center coordinates are as follows:
wherein: u (u) 0 The amount to be optimized in the objective function is a random initial value in (0, 1); the rands () represents a randomly generated vector.
Step 5.3 if the left whisker coordinate is expressed as u left The right whisker coordinate is expressed as u right The relationship between the left and right whiskers can be expressed as: u (u) left -u right ＝D 0 Dir, dir represents a normalized random vector, and after l iterations, the two whisker coordinates are respectively:
wherein: u (u) 0 l Centroid coordinates at the first iteration; d (D) 0 l The distance between the two whiskers in the first iteration;
substituting the odor intensity of the left whisker and the right whisker into an objective function to determine the odor intensity of the left whisker and the right whisker: j (J) left ＝J(u left )、J right ＝J(u right ) And comparing the sizes of the two.
If J left ＜J right The longicorn advances in the left direction, and the barycenter coordinates are:
u 0 l+1 ＝u 0 l +δ l ·dir(J left -J right )
if J left ＞J right The longicorn advances in the right direction, and the barycenter coordinates are:
u 0 l+1 ＝u 0 l -δ l ·dir(J left -J right )
the above two equations can be uniformly written using the sign function:
u 0 l+1 ＝u 0 l -δ l ·dir·sign(J left -J right )
step 5.4, determining the odor intensity after the first +1 iteration, and updating the step length and the distance between two whiskers at the time:
δ l+1 ＝eta_δ·δ l
D 0 l+1 ＝eta_D 0 ·D 0 l
wherein: delta l+1 The moving step length of the longicorn in the first iteration is the moving step length of the longicorn in the first iteration (1+1); d (D) 0 l+1 Is the distance between the two beards of the longicorn at the first +1 iteration.
Step 5.5, judging whether the condition of iteration cut-off is reached: a certain number of iterations n=5000 is reached, or the centroid coordinates are unchanged 200 times in succession. If the iteration cut-off condition is met, outputting a global optimal value u * Namely, the optimal control quantity is obtained,
if not, repeating the operation steps 5.2-5.4 until the condition is met.
The beneficial effect of above-mentioned scheme is: because the characteristics of the algorithm are randomly selected and the splitting threshold value is also randomly set, all samples are adopted for the establishment of each decision tree, so that the randomness of training based on the extremely random tree is stronger, the prediction effect of the trained regression prediction model is better, and the precision of the obtained furnace temperature prediction value is higher.
Drawings
FIG. 1 is a block diagram of an extreme random tree furnace temperature predictive control system based on longhorn beetle whisker search.
FIG. 2 is an overall flowchart of extreme random tree predictive control based on longhorn beetle whisker search.
Fig. 3 is a schematic flow chart of an extreme random tree algorithm.
Fig. 4 is a flow chart of the principle of the longhorn beetle whisker search algorithm.
Fig. 5 is a graph comparing the effect of prediction errors of 4 algorithms. Wherein (a) is an RF prediction error map; (b) is a KNN prediction error map; (c) is an ET prediction error map; (d) is a SVR prediction error map.
Fig. 6 is a control strategy comparison chart.
Detailed Description
The following describes the embodiments of the present invention further with reference to the drawings.
The predictive model of FIG. 1 is generated by an extreme stochastic tree algorithm with inputs including gas control u (k), temperature T (k), and M (k), where M (k) is a 1X 3 dimensional vectorThe group consists of air quantity, waste gas concentration and oxygen content. Output T of predictive model m (k+1) obtaining T through feedback correction p (k+1) input to an optimization step of performing rolling optimization on the control variable based on the longhorn beetle whisker search, wherein T r (k+1) is the reference input.
The specific process for modeling the furnace temperature based on the extreme stochastic tree algorithm of longhorn beetle whisker optimization is as follows:
step 1 acquiring production data of historical lots through a real-time database of a control system for an acquired raw dataset { (A) 1 ,y 1 ),(A 2 ,y 2 )…(A i ,y i )…(A N ,y N ) }, wherein A i Is a row vector of dimension 1 x 5, is a set of inputs to the modeled samples, y i Is A i The corresponding training samples have true values, i=1, 2 … N, N being the number of sample sets.
Step 2, selecting an original data set N=3000, dividing the data set by a cross-validation method, selecting k=10 to obtain 10 mutually exclusive subsets, and determining a training sample set N according to the setting of a training set and a test set 9:1 1 =2700, test sample set N 2 =300. Setting a maximum iteration number T, and initializing the iteration number t=1;
step 3, taking the root mean square error as a quantization evaluation criterion to split the features, and selecting the features with the minimum root mean square error to split;
and step 4, judging whether the splitting end condition is met, if so, jumping to step 5, otherwise, jumping to step 3. The condition for satisfying the end of splitting is set as follows: maximum depth of tree (max_depth) =50; the internal node subdivides the minimum number of samples required (min_samples_split) =2; leaf node minimum number of samples (min_samples_leaf) =1.
Step 5, after splitting, obtaining a decision tree f obtained by the t-th iteration t 。
Step 6, if t=t, go to step 7, otherwise t=t+1, go to step 2;
step 7, combining the generated T decision trees by using a mean method to obtain a prediction modelThe method comprises the following steps: />
Step 8 gives the corresponding objective function:obtaining optimal u by using longhorn beetle whisker search algorithm * 。
Examples:
by adopting the extreme random tree furnace temperature prediction control method based on the longhorn beetle whisker search, which is provided by the invention, on-site data of 5 days are randomly derived from a database, and the change range of natural gas flow and furnace temperature of 8 days and 10 days in 2018 is found to be very large, so that all possible working conditions on site are almost covered. A total of 32600 sets of data were collected on the same day, with a sampling period of t=0.1 s. Randomly selecting 3000 groups from the 32600 groups of data, and selecting 2700 groups as training set omega of the extreme random tree algorithm by a cross validation method 1 ＝{(A 1 ,y 1 ),(A 2 ,y 2 )…(A 2700 ,y 2700 )}，A i Is a row vector of dimension 1 x 5, is a set of input quantities of modeling samples, y i Is A i The true output of the corresponding modeling sample, i=1, 2, …,2700; the remaining 300 groups are used as test set Ω 2 ＝{(B 1 ,y 1 ),(B 2 ,y 2 )…(B 300 ,y 300 )}，B i Is a row vector of dimension 1 x 5, is a set of input quantities of modeling samples, y i Is B i The true output of the corresponding modeling sample, i=1, 2, …,300. The specific implementation mode is as follows:
initializing various parameters, setting maximum iteration times T=100 and maximum feature numbersMaximum depth of tree (max_depth) =50; the internal node subdivides the minimum number of samples required (min_samples_split) =2; leaf node minimum sample number [ ]min_samples_leaf) =1. And randomly selecting delta features from the input training set A as initial nodes to split according to the detailed step 3, so as to obtain a regression prediction model. The test data set is input to carry out overdriving on the model prediction effect, and as can be seen from the figure 5, the extreme random tree furnace temperature prediction control method based on the longhorn beetle whisker search has smaller prediction error and highest prediction precision compared with SVR, KNN, RF.
And reading the related data after the ET algorithm is executed at the previous moment, substituting the related data into an objective function, and initializing parameters of the longhorn beetle whisker search algorithm. Setting the problem dimension d=3, the initial step delta=30, the distance between the initial step and the two whiskers c=4, the attenuation coefficient eta=0.95 and the iteration number n=5000. The method is characterized in that the longhorn beetle carries out iterative search in a global range and finally outputs the optimal control quantity, and as can be seen from the figure 6, for the step signal randomly given by the system, the control effect of the extreme random tree predictive control algorithm based on longhorn beetle whisker search is superior to that of generalized predictive control (Generalized Predictive Control, GPC) which is put into production on site, and the system can respond to the variable quantity more quickly under the condition that the temperature is suddenly changed by the extreme random tree predictive control method based on longhorn beetle whisker search, and has shorter rise time, smaller overshoot and smaller steady-state error.
The foregoing is a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and any simple modification, equivalent variation and variation of the above embodiment according to the technical substance of the present invention falls within the scope of the technical solution of the present invention.
Claims (5)
1. An extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search is characterized by comprising the following steps:
step 1, collecting control input and output data of past and current moments in the combustion process as an original data set O;
step 2, dividing the original data set O into training sets N by adopting a cross-validation mode 1 And test set N 2 ；
Step 3, training all training sets by using an extreme random tree ET algorithm, and establishing a regression prediction model;
step 4, performing simulation prediction on the test set by using the established regression prediction model to obtain a predicted value of the hearth temperature, introducing a feedback correction link, and correcting the predicted hearth temperature;
step 5, taking the set temperature of the hearth, the corrected predicted temperature and the historical control quantity as inputs, and solving an objective function J according to a longhorn beetle whisker search algorithm to obtain an optimal gas control quantity u (k), so as to realize rolling optimization of the control quantity;
the specific steps of the cross-validation method described in the step 2 are as follows:
step 2.1, the original data set O is subjected to layered sampling for k times to obtain k mutually exclusive subsets with similar sizes,
step 2.2, using a union set of k-1 mutually exclusive subsets as a training set each time, and using the rest subset as a test set; k groups of training sets and test sets are obtained altogether, so that k times of modeling is carried out, and finally, the average value of the test result is output;
for the acquired dataset Ω= { (a) 1 ,y 1 ),(A 2 ,y 2 )…(A i ,y i )…(A N ,y N ) }, wherein A i Is a row vector of dimension 1 x 5, is input for modeling samples, y i Is A i The real output values of the corresponding samples, i=1, 2 … N, N are the sample group numbers, and training is performed on the data based on an extreme random tree algorithm, wherein the specific training process is as follows:
step 3.1, given training data set: n is selected from the acquired data set omega by a cross-validation method 1 Omega corresponding to group data 1 Training the data set as a training set; setting a maximum iteration number T, and initializing the iteration number t=1;
step 3.2, random division of node data sets: for the t-th iteration, from training set N 1 Sigma feature sequences { a }, are randomly selected 1 ,a 2 ,…a σ As the initial section }Points and randomly selecting a corresponding set of sequences { s } from the selected features 1 ,s 2 ,…,s σ A threshold value of the characteristic attribute, wherein the value range of the threshold value is between the maximum value and the minimum value of the characteristic; a in the feature sequence x Is greater than s x Samples of x=1, 2, … and sigma are classified into left nodes of branches, and other samples are classified into right nodes, so that random division under the nodes is realized;
step 3.3, obtaining an optimal splitting value under random characteristics: taking root mean square error as a quantization evaluation criterion for the numerical characteristics of the method; traversing all the characteristic values on the left node and the right node, calculating the root mean square error of each characteristic value on the corresponding node, and selecting the characteristic value corresponding to the minimum root mean square error as a splitting point;
step 3.4, judging whether the splitting end condition is met, if yes, jumping to step 3.5, otherwise updating the iteration times, executing t=t+1, and repeating steps 3.2-3.3; the splitting end condition may be any one of the following conditions:
(1) Maximum depth max_depth of tree;
(2) The internal node subdivides the minimum sample number min_samples_split required;
(3) The minimum number of samples of leaf nodes is min_samples_leaf;
step 3.5, after splitting is finished, obtaining a decision tree f obtained by the t-th iteration t ；
Step 3.6, if t=t goes to step 3.7, otherwise t=t+1 goes to step 3.2;
step 3.7, combining the generated T decision trees by using a mean method to obtain a prediction modelThe method comprises the following steps: />
Inputting the test set into the prediction model to obtain prediction outputN 2 The number of the groups of the test set is the number of the groups, and the corresponding evaluation index is used for measuring the prediction effect;
in step 3.3, the selected feature is a i ∈{a 1 ,a 2 ,…a σ When in the process of }, after the division in the step 3.2, the samples on the left and right sides of the feature are respectively a il ＝(α 1 ,α 2 ,…,α i ,…,α p )，a ir ＝(β 1 ,β 2 ,…,β j ,…,β q ) Wherein: alpha i For samples assigned to the left, i=1, 2, … p; beta j For the samples assigned to the right, j=1, 2, … q; p and q are the number of samples on the left and right sides respectively; the root mean square error RMSE corresponding to each eigenvalue of the left node i The method comprises the following steps:
wherein the method comprises the steps ofSelecting the feature with the minimum root mean square error for splitting; the calculation mode of the right node is the same as that of the left node;
the longhorn beetle whisker search algorithm in the step 5 comprises the following specific steps:
step 5.1, giving a corresponding objective function:
wherein: t (T) r Is input for a reference terminal; t (T) p Outputting the corrected prediction; u is a control variable; lambda is a weight coefficient, and lambda is more than 0 and less than 1; d is the predicted step number;
defining e as a model error vector:
e＝[T r (k+1)-T p (k+1),…,T r (k+d)-T p (k+d)] T
define deltau as the increment of the control quantity:
Δu＝[u(k)-u(k-1),…,u(k+d-1)-u(k+d-2)] T
according to the two formulas above, f is defined as:
f＝[e T λΔu T ] T ；
screening a d-dimensional control vector u * Let objective function J reach minimum, can obtain J and be the mathematical formula of minimum through the definition formula of f: min j=f T ×f；
Step 5.2, reading data after the ET algorithm is executed at the last moment, substituting the data into an objective function, and initializing parameters of the longhorn beetle whisker search algorithm; the parameter initialization includes: the dimension of the problem is D, the distance between two whiskers is D 0 The initial step length is delta, and the attenuation coefficient of the distance between two whiskers is eta_D 0 The attenuation coefficient of the step length is eta delta, the iteration number is n, and the mass center coordinates are as follows:
wherein: u (u) 0 The amount to be optimized in the objective function is a random initial value in (0, 1); the fields () represents a randomly generated vector;
step 5.3, let the left whisker coordinate be expressed as u left The right whisker coordinate is expressed as u right The relationship between the left and right whiskers can be expressed as: u (u) left -u right ＝D 0 Dir, dir represents a normalized random vector, and after l iterations, the two whisker coordinates are respectively:
wherein: u (u) 0 l Centroid coordinates at the first iteration; d (D) 0 l The distance between the two whiskers in the first iteration;
substituting the odor intensity of the left whisker and the right whisker into an objective function to determine the odor intensity of the left whisker and the right whisker: j (J) left ＝J(u left )、J right ＝J(u right ) And comparing the sizes of the two;
if J left ＜J right The longicorn advances in the left direction, and the barycenter coordinates are:
u 0 l+1 ＝u 0 l +δ l ·dir(J left -J right )
if J left ＞J right The longicorn advances in the right direction, and the barycenter coordinates are:
u 0 l+1 ＝u 0 l -δ l ·dir(J left -J right )
the above two equations can be uniformly written using the sign function:
u 0 l+1 ＝u 0 l -δ l ·dir·sign(J left -J right )
step 5.4, determining the odor intensity after the first (1) th iteration, and updating the step length and the distance between two whiskers at the moment:
δ l+1 ＝eta_δ·δ l
D 0 l+1 ＝eta_D 0 ·D 0 l
wherein: delta l+1 The moving step length of the longicorn in the first iteration is the moving step length of the longicorn in the first iteration (1+1); d (D) 0 l+1 Is the distance between the two beards of the longicorn in the first iteration and the second iteration;
step 5.5, judging whether the condition of iteration cut-off is reached: reaching a certain iteration number, or continuously unchanged centroid coordinates; if the iteration cut-off condition is met, outputting a global optimal value u * And (3) if the control quantity is the optimal control quantity, repeating the operation steps 5.2 to 5.4 until the control quantity meets the condition.
2. The method for controlling furnace temperature prediction of an extreme random tree based on longhorn beetle whisker search according to claim 1, wherein in the step 1, the collected data comprises: exhaust gas concentration, oxygen content, air quantity, proportional valve opening and temperature.
3. The method for controlling furnace temperature prediction of an extreme random tree based on longhorn beetle whisker search according to claim 1, wherein in the step 3.1, the maximum iteration number is 100.
4. The method for predicting and controlling the furnace temperature of the extreme random tree based on the longhorn beetle whisker search according to claim 1, wherein in the step 3.4, the splitting end condition is satisfied by any one of the following conditions:
(1) Maximum depth max_depth=50 of the tree;
(2) The internal node subdivides the minimum number of samples required min_samples_split=2;
(3) Leaf node minimum number of samples min_samples_leaf=1.
5. The method for controlling furnace temperature prediction of an extreme random tree based on longhorn beetle whisker search according to claim 1, wherein in the step 5.5, the number of iterations reached is 5000, or the centroid coordinates are cut off iteratively when 200 consecutive times are unchanged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011072976.1A CN112163376B (en) | 2020-10-09 | 2020-10-09 | Extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011072976.1A CN112163376B (en) | 2020-10-09 | 2020-10-09 | Extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112163376A CN112163376A (en) | 2021-01-01 |
| CN112163376B true CN112163376B (en) | 2024-03-12 |
Family
ID=73866416
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011072976.1A Active CN112163376B (en) | 2020-10-09 | 2020-10-09 | Extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112163376B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113069124A (en) * | 2021-03-09 | 2021-07-06 | 浙江工业大学 | Electrocardio monitoring method based on CNN-ET model |
| CN114719333A (en) * | 2022-04-14 | 2022-07-08 | 兰州理工大学 | Multi-inverter electromagnetic heating system and output power prediction control method thereof |
| CN115028301B (en) * | 2022-05-31 | 2024-04-26 | 河北工程大学 | Intelligent purifying swimming pool water recycling system and method |
| CN115186582B (en) * | 2022-07-05 | 2023-04-18 | 科大智能物联技术股份有限公司 | Steel rolling heating furnace control method based on machine learning model |
| CN116625134B (en) * | 2023-07-24 | 2023-10-24 | 苏州弘皓光电科技有限公司 | Electric furnace temperature monitoring control method and system based on 5G technology |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101457264A (en) * | 2008-12-29 | 2009-06-17 | 杭州电子科技大学 | Blast furnace temperature optimization control method |
| CN109473148A (en) * | 2018-10-26 | 2019-03-15 | 武汉工程大学 | A kind of ion concentration prediction technique, device and computer storage medium |
| CN109709795A (en) * | 2018-12-24 | 2019-05-03 | 东华大学 | A kind of PID controller parameter setting method based on longicorn palpus searching algorithm |
| CN110175195A (en) * | 2019-04-23 | 2019-08-27 | 哈尔滨工业大学 | Mixed gas detection model construction method based on extreme random tree |
| CN110532674A (en) * | 2019-08-27 | 2019-12-03 | 东北电力大学 | A kind of coal-fired power station boiler fire box temperature measurement method |
| WO2020000248A1 (en) * | 2018-06-27 | 2020-01-02 | 大连理工大学 | Space reconstruction based method for predicting key performance parameters of transition state acceleration process of aircraft engine |
| CN110766137A (en) * | 2019-10-18 | 2020-02-07 | 武汉大学 | Power electronic circuit fault diagnosis method based on longicorn whisker optimized deep confidence network algorithm |
| CN110783962A (en) * | 2019-04-17 | 2020-02-11 | 上海电机学院 | Control method of virtual synchronous generator grid-connected inverter |
| CN110852902A (en) * | 2019-11-14 | 2020-02-28 | 合肥工业大学 | Photovoltaic power generation power prediction method based on BAS-BP |
| CN111158237A (en) * | 2019-12-25 | 2020-05-15 | 南京理工大学 | Industrial furnace temperature multi-step prediction control method based on neural network |
| CN111173573A (en) * | 2020-01-08 | 2020-05-19 | 上海电力大学 | Identification method for power object model of steam turbine regulating system |
| CN111562744A (en) * | 2020-06-15 | 2020-08-21 | 安徽工业大学 | Boiler combustion implicit generalized predictive control method based on PSO algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2273327A1 (en) * | 2009-06-24 | 2011-01-12 | ABB Research Ltd. | Estimating initial states of a system model for controlling an industrial process |
-
2020
- 2020-10-09 CN CN202011072976.1A patent/CN112163376B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101457264A (en) * | 2008-12-29 | 2009-06-17 | 杭州电子科技大学 | Blast furnace temperature optimization control method |
| WO2020000248A1 (en) * | 2018-06-27 | 2020-01-02 | 大连理工大学 | Space reconstruction based method for predicting key performance parameters of transition state acceleration process of aircraft engine |
| CN109473148A (en) * | 2018-10-26 | 2019-03-15 | 武汉工程大学 | A kind of ion concentration prediction technique, device and computer storage medium |
| CN109709795A (en) * | 2018-12-24 | 2019-05-03 | 东华大学 | A kind of PID controller parameter setting method based on longicorn palpus searching algorithm |
| CN110783962A (en) * | 2019-04-17 | 2020-02-11 | 上海电机学院 | Control method of virtual synchronous generator grid-connected inverter |
| CN110175195A (en) * | 2019-04-23 | 2019-08-27 | 哈尔滨工业大学 | Mixed gas detection model construction method based on extreme random tree |
| CN110532674A (en) * | 2019-08-27 | 2019-12-03 | 东北电力大学 | A kind of coal-fired power station boiler fire box temperature measurement method |
| CN110766137A (en) * | 2019-10-18 | 2020-02-07 | 武汉大学 | Power electronic circuit fault diagnosis method based on longicorn whisker optimized deep confidence network algorithm |
| CN110852902A (en) * | 2019-11-14 | 2020-02-28 | 合肥工业大学 | Photovoltaic power generation power prediction method based on BAS-BP |
| CN111158237A (en) * | 2019-12-25 | 2020-05-15 | 南京理工大学 | Industrial furnace temperature multi-step prediction control method based on neural network |
| CN111173573A (en) * | 2020-01-08 | 2020-05-19 | 上海电力大学 | Identification method for power object model of steam turbine regulating system |
| CN111562744A (en) * | 2020-06-15 | 2020-08-21 | 安徽工业大学 | Boiler combustion implicit generalized predictive control method based on PSO algorithm |
Non-Patent Citations (8)
| Title |
|---|
| Xiaojun Wang.Ladle furnace temperature prediction model based on large-scale data with random forest.《IEEE/CAA Journal of Automatica Sinica》.2016,第770-774页. * |
| Xingrui Chen.et al.Load Forecasting Based On Elman Neural Network Optimized By Beetle Antennae Search Optimization.《2019 1st International Conference on Industrial Artificial Intelligence (IAI)》.2019,第1-6页. * |
| 应健锋等.一种多Ｘ值输入下测试覆盖率损失的预测方法.《微电子学与计算机》.2020,第37卷(第4期),第12-18页. * |
| 李大中等.一种利用混合算法选择变量的天牛须优化神经网络风速预测方法.《电力科学与工程》.2019,第35卷(第12期),第28-33页. * |
| 瞿红春等.基于IBAS-Elman网络的滚动轴承故障诊断研究.机床与液压.2020,第48卷(第16期),第201-205页. * |
| 薛美盛等.神经网络预测控制在加热炉炉温控制中 的仿真研究.《化工自动化及仪表》.2018,第45卷(第8期),第590-594页. * |
| 闭应洲等.《数据挖掘与机器学习》.浙江科学技术出版社,2020,第51页. * |
| 韩瑞等.基于Boruta和极端随机树方法的森林蓄积量估测.《林业资源管理》.2020,(第4期),第127-133页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112163376A (en) | 2021-01-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112163376B (en) | Extreme random tree furnace temperature prediction control method based on longhorn beetle whisker search | |
| Yuan et al. | Stacked enhanced auto-encoder for data-driven soft sensing of quality variable | |
| CN111804146B (en) | Intelligent ammonia injection control method and intelligent ammonia injection control device | |
| CN111650834B (en) | Sewage treatment process prediction control method based on extreme learning machine | |
| CN114547974A (en) | Dynamic soft measurement modeling method based on input variable selection and LSTM neural network | |
| Meddage et al. | Tree-based regression models for predicting external wind pressure of a building with an unconventional configuration | |
| CN107403196A (en) | Instant learning modeling method based on spectral clustering analysis | |
| Wu et al. | Learning soft sensors using time difference–based multi-kernel relevance vector machine with applications for quality-relevant monitoring in wastewater treatment | |
| CN117289652A (en) | Numerical control machine tool spindle thermal error modeling method based on multi-universe optimization | |
| CN114186472B (en) | Design method of multi-input multi-output urban solid waste incineration process model | |
| 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 | |
| CN110276478B (en) | Short-term wind power prediction method based on segmented ant colony algorithm optimization SVM | |
| CN109101683B (en) | Model updating method for pyrolysis kettle of coal quality-based utilization and clean pretreatment system | |
| CN110909492A (en) | Sewage treatment process soft measurement method based on extreme gradient lifting algorithm | |
| Tian et al. | Coke oven flue temperature control based on improved implicit generalized predictive control | |
| Li et al. | Adaptive Incremental Learning Stochastic Configuration Network and Its NIR Modeling Application | |
| Liu et al. | A nonlinear predictive control algorithm based on fuzzy online modeling and discrete optimization | |
| CN113610260A (en) | Method for predicting concentration of flue gas components in incineration process of municipal domestic waste | |
| Mohammad et al. | An Optimal Nonlinear Type-2 Fuzzy FOPID Control Design Based on Integral Performance Criteria Using FSM | |
| Zhang et al. | Data‐Driven Optimization Framework for Nonlinear Model Predictive Control | |
| Eksin et al. | Design of optimum fuzzy controller using genetic algorithms | |
| CN115146833B (en) | Prediction method for generation concentration of nitrogen oxides of boiler | |
| CN114282614B (en) | Medium-long runoff prediction method for optimizing CNN-GRU based on random forest and IFDA | |
| CN116187187A (en) | Method for predicting oxygen content of flue gas in urban solid waste incineration process in real time |
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 |