CN114368768B - LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method - Google Patents
LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method Download PDFInfo
- Publication number
- CN114368768B CN114368768B CN202011106571.5A CN202011106571A CN114368768B CN 114368768 B CN114368768 B CN 114368768B CN 202011106571 A CN202011106571 A CN 202011106571A CN 114368768 B CN114368768 B CN 114368768B
- Authority
- CN
- China
- Prior art keywords
- data
- model
- prediction
- particle size
- seed
- 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 35
- WNROFYMDJYEPJX-UHFFFAOYSA-K aluminium hydroxide Chemical compound [OH-].[OH-].[OH-].[Al+3] WNROFYMDJYEPJX-UHFFFAOYSA-K 0.000 title claims abstract description 24
- 239000013078 crystal Substances 0.000 title claims abstract description 17
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 31
- 238000012549 training Methods 0.000 claims abstract description 28
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 15
- 238000010606 normalization Methods 0.000 claims abstract description 10
- 238000011156 evaluation Methods 0.000 claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims abstract description 7
- 238000005070 sampling Methods 0.000 claims abstract description 5
- 238000012360 testing method Methods 0.000 claims abstract description 5
- 238000004140 cleaning Methods 0.000 claims abstract description 4
- 239000002245 particle Substances 0.000 claims description 36
- ANBBXQWFNXMHLD-UHFFFAOYSA-N aluminum;sodium;oxygen(2-) Chemical compound [O-2].[O-2].[Na+].[Al+3] ANBBXQWFNXMHLD-UHFFFAOYSA-N 0.000 claims description 12
- 238000004880 explosion Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 229910001388 sodium aluminate Inorganic materials 0.000 claims description 10
- 238000013528 artificial neural network Methods 0.000 claims description 9
- CIWBSHSKHKDKBQ-JLAZNSOCSA-N Ascorbic acid Chemical compound OC[C@H](O)[C@H]1OC(=O)C(O)=C1O CIWBSHSKHKDKBQ-JLAZNSOCSA-N 0.000 claims description 8
- 230000015654 memory Effects 0.000 claims description 6
- 239000012452 mother liquor Substances 0.000 claims description 6
- 239000007787 solid Substances 0.000 claims description 6
- 210000000582 semen Anatomy 0.000 claims description 5
- 238000002425 crystallisation Methods 0.000 claims description 4
- 230000008025 crystallization Effects 0.000 claims description 4
- 210000002569 neuron Anatomy 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 3
- 230000001502 supplementing effect Effects 0.000 claims description 3
- 238000010801 machine learning Methods 0.000 claims description 2
- 238000003062 neural network model Methods 0.000 claims description 2
- 238000004519 manufacturing process Methods 0.000 abstract description 20
- PNEYBMLMFCGWSK-UHFFFAOYSA-N aluminium oxide Inorganic materials [O-2].[O-2].[O-2].[Al+3].[Al+3] PNEYBMLMFCGWSK-UHFFFAOYSA-N 0.000 description 16
- 239000000243 solution Substances 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 230000000737 periodic effect Effects 0.000 description 4
- 238000003556 assay Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000004131 Bayer process Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000000428 dust Substances 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000004907 flux Effects 0.000 description 2
- KRHYYFGTRYWZRS-UHFFFAOYSA-N Fluorane Chemical compound F KRHYYFGTRYWZRS-UHFFFAOYSA-N 0.000 description 1
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 1
- 229910052782 aluminium Inorganic materials 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 239000010419 fine particle Substances 0.000 description 1
- 229910000040 hydrogen fluoride Inorganic materials 0.000 description 1
- 229910017053 inorganic salt Inorganic materials 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000003446 memory effect Effects 0.000 description 1
- TWNQGVIAIRXVLR-UHFFFAOYSA-N oxo(oxoalumanyloxy)alumane Chemical compound O=[Al]O[Al]=O TWNQGVIAIRXVLR-UHFFFAOYSA-N 0.000 description 1
- 235000021110 pickles Nutrition 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000012047 saturated solution Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000001179 sorption measurement Methods 0.000 description 1
- 230000009897 systematic effect Effects 0.000 description 1
Classifications
-
- C—CHEMISTRY; METALLURGY
- C01—INORGANIC CHEMISTRY
- C01F—COMPOUNDS OF THE METALS BERYLLIUM, MAGNESIUM, ALUMINIUM, CALCIUM, STRONTIUM, BARIUM, RADIUM, THORIUM, OR OF THE RARE-EARTH METALS
- C01F7/00—Compounds of aluminium
- C01F7/02—Aluminium oxide; Aluminium hydroxide; Aluminates
- C01F7/04—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom
- C01F7/06—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom by treating aluminous minerals or waste-like raw materials with alkali hydroxide, e.g. leaching of bauxite according to the Bayer process
- C01F7/0666—Process control or regulation
-
- C—CHEMISTRY; METALLURGY
- C01—INORGANIC CHEMISTRY
- C01F—COMPOUNDS OF THE METALS BERYLLIUM, MAGNESIUM, ALUMINIUM, CALCIUM, STRONTIUM, BARIUM, RADIUM, THORIUM, OR OF THE RARE-EARTH METALS
- C01F7/00—Compounds of aluminium
- C01F7/02—Aluminium oxide; Aluminium hydroxide; Aluminates
- C01F7/04—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom
- C01F7/06—Preparation of alkali metal aluminates; Aluminium oxide or hydroxide therefrom by treating aluminous minerals or waste-like raw materials with alkali hydroxide, e.g. leaching of bauxite according to the Bayer process
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- C—CHEMISTRY; METALLURGY
- C01—INORGANIC CHEMISTRY
- C01P—INDEXING SCHEME RELATING TO STRUCTURAL AND PHYSICAL ASPECTS OF SOLID INORGANIC COMPOUNDS
- C01P2004/00—Particle morphology
- C01P2004/60—Particles characterised by their size
- C01P2004/61—Micrometer sized, i.e. from 1-100 micrometer
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention relates to an LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and a method. The method comprises the following steps: s1: collecting historical data; s2: processing the data sampling interval t; s3: cleaning and interpolating the data; s4: carrying out moving average processing on the data; s5: carrying out maximum and minimum value normalization processing on the data; s6: constructing an input matrix and an output matrix of model training data; s7: constructing a seed crystal granularity refinement burst prediction model; s8: according to 7:3, dividing the ratio into a training set and a testing set; s9: setting model parameters; s10: performing inverse normalization and recovering to be a normal index value; s11: evaluating the model using root mean square error; s12: adjusting model parameters until the model evaluation index is at a better value; s13: and finishing model training by using the historical data, and storing the model training as a fixed file. Realizes the stable production of seed decomposition process, improves the system yield and optimizes the product index.
Description
Technical Field
The invention relates to an aluminum hydroxide seed crystal granularity refinement burst prediction model and method, in particular to an LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method in the technical field of aluminum oxide production.
Background
Seed decomposition is one of the key processes for producing alumina by the bayer process, and the production condition of the seed decomposition process determines the quality of metallurgical grade sandy alumina. The sandy alumina is mainly used as a raw material for electrolytic aluminum production, and is required to have the advantages of coarse granularity, good fluidity, strong hydrogen fluoride adsorption capacity and the like. Therefore, the quality of metallurgical grade sandy alumina is improved, the production efficiency of alumina is improved, and the metallurgical grade sandy alumina is more and more paid attention to alumina production enterprises. However, there is an absolute conflict between the improvement of the decomposition rate and the improvement of the quality of the alumina product, so that the yield of the alumina product is improved as much as possible on the premise of meeting the quality of the alumina product, which is one of the problems to be solved by alumina manufacturers.
The grain size change of the seed decomposition procedure is actually a process of crystallizing and separating out aluminum hydroxide from supersaturated sodium aluminate solution, and the seed crystal decomposition process of the supersaturated sodium aluminate solution is a complex physicochemical process unlike the crystallization of a conventional inorganic salt saturated solution. In terms of sodium aluminate solution seed crystal decomposition theory, a great deal of research work is also being done by global researchers, but no unification is currently underway.
In recent years, the granularity and strength of alumina produced by Bayer process in China are improved greatly, but the influence of periodical refinement is not eliminated, namely, a large amount of aluminum hydroxide fine particles (less than 45 mu m) are exploded out in a seed decomposition system in a certain production stage. During refinement, the seed filtration vertical disc-level aluminum hydroxide filtration flat disc has poor filtration effect, the productivity is reduced, more dust can be generated in the downstream roasting process, and the energy and dust collection are greatly influenced.
Different alumina production enterprises have different production equipment and production working conditions, and in the production process, the influence weights of various factors on the decomposition granularity of seeds are also greatly different, so that an accurate mathematical mechanism model is difficult to build aiming at the change of the decomposition granularity. The experimental data obtained in the laboratory are less, and are only effective under specific experimental conditions, and lack systemicity and pertinence. This results in that it is difficult for each alumina producer to find an effective solution in controlling the decomposition granularity, and even if a certain enterprise finds a certain control rule in long-term production, it is difficult to popularize it into other alumina producers.
Therefore, we need to switch the particle size control concept, starting with the refinement of seed particle size that disrupts production. The periodic refinement of the seed crystal decomposition step includes periodic two words, but the timing and degree of refinement cannot be accurately predicted.
In the seed decomposition production process, if a seed particle size refinement prediction model can be established according to the production conditions and historical data of a decomposition process, the change trend of the seed particle size can be mastered in advance, the refinement degree of the seed particle size is known, a regulation and control means is adopted as early as possible, the amplitude of periodic refinement is weakened, the seed particle size refinement is even eliminated, and finally, the stable production of the seed decomposition process is realized, the system yield is improved, and the product index is optimized.
Disclosure of Invention
In order to solve the problems, the invention provides an LSTM-based aluminum hydroxide seed particle size refinement burst prediction model and method, and aims to grasp the seed particle size change trend in advance, know the refinement degree of the seed particle size, take a regulation and control means as soon as possible, weaken the periodical refinement amplitude and even eliminate the seed particle size refinement, finally realize the stable production of a seed decomposition procedure, improve the system yield and optimize the product index.
The technical scheme adopted by the invention is as follows:
an LSTM-based aluminum hydroxide seed particle size refinement burst prediction model comprising the steps of:
s1: collecting historical data of all relevant variables of the predicted seed grain refinement burst;
s2: processing the data sampling interval delta t to ensure that the data length of all variables is consistent and the time interval is consistent, namely, the data time scale is unified;
s3: cleaning and interpolating the data, deleting abnormal values, supplementing the missing values, and constructing continuous time sequence data of the multidimensional features;
s4: because the collected data has a certain system error, the data is subjected to moving average processing;
s5: the input variables are prevented from being unbalanced in weight due to the huge difference of the orders of magnitude, and the maximum and minimum normalization processing is carried out on the data, so that the convergence rate in the training process is improved;
s6: the seed crystal granularity refinement condition of how many days in the future cannot be predicted accurately, so that parameters are required to be continuously adjusted in the training process of the model, the prediction step number q is improved as much as possible on the premise of ensuring the prediction accuracy, and an input matrix and an output matrix of model training data are constructed;
s7: constructing a seed crystal granularity refinement burst prediction model;
s8: the multi-dimensional feature continuous time series data constructed in the step S5 are processed according to the following steps: 3, dividing the ratio into a training set and a testing set;
s9: setting model parameters;
s10: model training is carried out to obtain a model prediction result, and the prediction result is reversely normalized and restored to be a normal index value;
s11: evaluating the model by using a model evaluation index and adopting a root mean square error, and measuring the deviation between the predicted value and the true value;
s12: repeating the step S9 repeatedly, and adjusting the model parameters until the model evaluation index accords with the expectation;
s13: and finishing model training by using the historical data, and storing the model training as a fixed file.
The collecting data for model prediction in step S1 includes: initial decomposition temperature, final decomposition temperature, first decomposition tank solid content, final decomposition tank solid content, sodium aluminate fine solution alpha k Mother liquor alpha k Sodium aluminate semen N k Seed precipitation mother liquor N k Sodium aluminate semen flux, crystallization aid (CGM) addition, seed particle size data (f 1.92, f3.55, f5.34, f8.87, f10.87, -9 μm, -14 μm, -29 μm, -45 μm, -59 μm, -80 μm, -101 μm, -125 μm, -150 μm, -162 μm), 25 variables in total.
In the step S5, the formula for normalization selection is as followsIn (1) the->Is x i,j Normalized value, ++>And->Is the variable x i Maximum and minimum values in the sequence.
The input and output matrix in the step S6 is as follows:
the seed crystal granularity refined burst prediction model constructed in the step S7 belongs to a cyclic neural network model for supervised machine learning, and the model comprises the following components: input layer, hidden layer, output layer, model training and model prediction; wherein:
a. the input layer is configured to seed grain refinement the historical data of all relevant variables of the burst;
b. the hidden layer consists of a plurality of long-period memory model nodes h (t0 Is composed of a forgetting door f (t) Input door (i) (t) ANDa (t) ) Output door o (t) And memory cell C (t) A function;
c. and the output layer outputs the grain size refinement index of the aluminum hydroxide seed crystal from the last layer of the hidden layer through the full-connection layer.
The parameters to be set in the step 9 include: model prediction step number q, neural network layer number, neuron number of each layer, loss function selection, optimizer selection, iteration number and training time window size.
The model evaluation index formula in the step S11 is as follows:
where q is the number of samples, y i And->Representing the true value and the predicted value of the ith sample, respectively.
The prediction method of the aluminum hydroxide seed grain refinement explosion prediction model based on LSTM comprises the steps of constructing an aluminum hydroxide seed grain refinement explosion prediction model, collecting historical data and latest data of all relevant variables of the prediction seed grain refinement explosion, processing the data according to the steps S2-S5, and constructing a prediction model Input data Input according to the finally determined prediction step number q when the prediction model is established; and inputting data by using a prediction model to obtain a prediction result, and carrying out inverse normalization on the prediction result to obtain a final prediction result.
The above-mentioned construction prediction model Input data Input:
wherein m is the number of prediction related variables, n is the data length, and x 1,n …x m,n Is the most current set of data.
The above prediction result Y predict =[y 1 …y q ]。
The invention has the beneficial effects that:
according to the production conditions of the decomposition process and the historical data, a seed grain refinement prediction model is established, so that the change trend of the seed grain size can be mastered in advance, the refinement degree of the seed grain size is known, a regulation and control means is adopted as soon as possible, the amplitude of periodic refinement is weakened, the seed grain refinement is even eliminated, the stable production of the seed decomposition process is finally realized, the system yield is improved, and the product index is optimized.
Drawings
FIG. 1 is a schematic diagram of a single-node design of an LSTM (long short term memory neural network).
FIG. 2 is a graph comparing the results of the seed particle size refinement burst assay with the results of model predictions.
Detailed description of the preferred embodiments
The present invention will be described in detail with reference to specific examples. The present embodiment is implemented on the premise of the technical scheme of the present invention, and a detailed implementation manner and a specific operation process are given, but the protection scope of the present invention is not limited to the following examples.
An LSTM-based aluminum hydroxide seed particle size refinement burst prediction model comprising the steps of:
s1: collecting historical data of all relevant variables predicting seed grain refinement outbreaks, including: initial decomposition temperature, final decomposition temperature, first decomposition tank solid content, final decomposition tank solid content, sodium aluminate fine solution alpha k Mother liquor alpha k Sodium aluminate semen N k Seed precipitation mother liquor N k Sodium aluminate semen flux, crystallization aid (CGM) addition, seed particle size data (f 1.92, f3.55, f5.34, f8.87, f10.87, -9 μm, -14 μm, -29 μm, -45 μm, -59 μm, -80 μm, -101 μm, -125 μm, -150 μm, -162 μm), 25 variables in total.
S2: and processing the data sampling interval delta t to ensure that the data length of all variables is consistent and the time interval is consistent, namely, the data time scale is unified.
S3: and cleaning and interpolating the data, deleting abnormal values, supplementing missing values, and constructing continuous time sequence historical data of a group of 25-dimensional features.
S4: since the data collected in step S1 has a certain systematic error, such as a sampling assay error, a sliding average process is performed on the data, and the sliding window size w=5.
S5: the input variables are prevented from causing weight unbalance due to huge difference of orders of magnitude, maximum and minimum normalization processing is carried out on smoothed data, and the data is mapped between [0,1] by the following formula, so that convergence speed in the training process is improved.
In the method, in the process of the invention,is x i,j Normalized value, ++>And->Is a variant ofQuantity x i Maximum and minimum values in the sequence.
S6: input and Output of burst prediction model training data are refined by using the seed granularity constructed by the data:
s7: the LSTM structure adopted by the model is an improved cyclic neural network, and is mainly characterized by being capable of processing time series data and still having a memory effect on data before a longer time. And the LSTM structure can solve the problems of gradient disappearance (Gradient Vanishing) and gradient explosion (Gradient Exploding) of the common circulating neural network. LSTM differs from a normal recurrent neural network in that a simple hidden layer node h is used (t) Improvement is carried out by adding the forgetting door f (t) Input door (i) (t) ANDa (t) ) Output door o (t) And memory cell C (t) Function. FIG. 1 is a schematic diagram of an LSTM single-node design. The functional expressions in the single-node structure of FIG. 1 are as follows:
f (t) =σ(W f [h (t-1) ,x (t) ]+b f )
i (t) =σ(W i [h (t-1) ,x (t) ]+b i )
a (t) =tanh(W a [h (t-1) ,x (t) ]+b a )
o (t) =σ(W o [h (t-1) ,x (t) ]+b o )
where W and b are the weight and bias of the corresponding function, respectively, σ is the activation function, typically sigmoid, and tanh is an activation function. Therefore, we can obtain the memory cell C of the node output (t) And node expression h (t) :
C (t) =f (t) ·C (t-1) +i (t) ·a (t)
h (t) =o (t) ·tanh(C (t) )
The model has a larger data range and data dimension, so that the expression capacity of the model can be enhanced, the neural network can be helped to express complex relations under the premise of limited data, and the model is particularly suitable for time sequence prediction, namely multi-step prediction is carried out on a certain variable value or a certain variable value. This characteristic is very suitable for the problems of fine prediction of seed particle size in the decomposition process in the production of alumina.
S8: dividing the continuous time series history data of the 25-dimensional feature constructed in the step S5 according to 7: the scale of 3 is divided into training and test sets.
S9: setting model parameters, wherein the parameters to be set mainly comprise: model prediction step number q, neural network layer number, neuron number of each layer, loss function selection, optimizer selection, iteration number and training time window size.
S10: model training is carried out to obtain a model prediction result, and the prediction result is reversely normalized and restored to be a normal index value.
S11: the model is evaluated by using a model evaluation index Root Mean Square Error (RMSE), the deviation between the seed grain refinement predicted value and the historical test true value is measured, and the RMSE formula is as follows:
where q is the number of samples, y i Andrepresenting the true value and the predicted value of the ith sample, respectively.
S12: and (5) repeatedly repeating the step S9 to adjust the model parameters until the model evaluation index is at a better value. Finally, the parameters of this example were chosen as follows:
model prediction step number q:80
Layer number of neural network: 3
Number of neurons per layer: 256. 128, 64
Loss function selection: mae
The optimizer selects: adam (Adam)
Iteration number: 50
Training time window size: 80
S13: and finishing model training by using the historical data, and storing the model training as a fixed file. This embodiment is implemented using python, the model is saved as a pickle file.
The prediction method of the aluminum hydroxide seed grain refinement explosion prediction model based on LSTM comprises the steps of constructing an aluminum hydroxide seed grain refinement explosion prediction model, collecting historical data and latest data of all relevant variables of the prediction seed grain refinement explosion, processing the data according to the steps S2-S5, and constructing a prediction model Input data Input according to the finally determined prediction step number q when the prediction model is established;
where n is the total length of the data, x 1,n …x 25,n A set of data up to date for the 25 variables.
Obtaining a prediction result Y by using the input data of the prediction model new =[y 1 …y 80 ]
Will predict result Y new And performing inverse normalization to obtain a final prediction result.
Finally, if the prediction effect of the model and the prediction method is required to be checked, the data trend is compared, and the data is required to be segmented. The n groups of data are respectively transmitted into an established seed crystal granularity refinement prediction model, and n groups of prediction results Y can be obtained predict ,
Y predict ={[y 1,1 ,y 1,2 ,…,y 1,80 ],[y 2,1 ,y 2,2 ,…,y 2,80 ],…,[y n,1 ,y n,2 ,…,y n,80 ]}
Each group of predicted results has the length of q=80, and the n groups of predicted results are extracted from every 80 groups to form a new sequence
New sequences are to be addedCurves were plotted separately from the-45 μm real assay results of the seeds, as shown in FIG. 2.
What is not described in detail in this specification is prior art known to those skilled in the art.
Claims (6)
1. The LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model is characterized by comprising the following steps of:
s1: collecting historical data of all relevant variables of the predicted seed grain refinement burst;
s2: processing the data sampling interval delta t to ensure that the data length of all variables is consistent and the time interval is consistent, namely, the data time scale is unified;
s3: cleaning and interpolating the data, deleting abnormal values, supplementing the missing values, and constructing continuous time sequence data of the multidimensional features;
s4: carrying out moving average processing on the data;
s5: carrying out maximum and minimum normalization processing on the data, and improving the convergence rate in the training process;
s6: constructing an input matrix and an output matrix of model training data;
s7: constructing a seed crystal granularity refinement burst prediction model;
s8: constructing continuous time series data of the multidimensional feature in the step S5 according to 7:3, dividing the ratio into a training set and a testing set;
s9: setting model parameters;
s10: model training is carried out to obtain a model prediction result, and the prediction result is reversely normalized and restored to be a normal index value;
s11: evaluating the model by using a model evaluation index and adopting a root mean square error, and measuring the deviation between the predicted value and the true value;
s12: repeating the step S9 repeatedly, and adjusting the model parameters until the model evaluation index accords with the expectation;
s13: training the model by using historical data, and storing the model as a fixed file;
the collecting data for model prediction in step S1 includes: initial decomposition temperature, final decomposition temperature, first decomposition tank solid content, final decomposition tank solid content, sodium aluminate fine solution alpha k Mother liquor alpha k Sodium aluminate semen N k Seed precipitation mother liquor N k Sodium aluminate liquor flow, crystallization aid addition, f1.92 seed particle size data, f3.55 seed particle size data, f5.34 seed particle size data, f8.87 seed particle size data, f10.87 seed particle size data, -9 μm seed particle size data, -14 μm seed particle size data, -29 μm seed particle size data, -45 μm seed particle size data, -59 μm seed particle size data, -80 μm seed particle size data, -101 μm seed particle size data, -125 μm seed particle size data, -150 μm seed particle size dataDegree data, -162 μm seed particle size data, 25 variables total;
in the step S5, the formula for normalization selection is as followsIn (1) the->Is x i,j Normalized value, ++>And->Is the variable x i Maximum and minimum values in the sequence;
the parameters to be set in the step 9 include: model prediction step number q, neural network layer number, neuron number of each layer, loss function selection, optimizer selection, iteration number and training time window size;
the model evaluation index formula in the step S11 is as follows:
where q is the number of samples, y i And->Representing the true value and the predicted value of the ith sample, respectively.
2. The LSTM-based aluminum hydroxide seed particle size refined burst prediction model of claim 1, wherein the input and output matrices in step S6 are:
3. the LSTM-based aluminum hydroxide seed particle size refinement burst prediction model according to claim 1, wherein the seed particle size refinement burst prediction model constructed in step S7 belongs to a supervised machine learning cyclic neural network model, and the model comprises: input layer, hidden layer, output layer, model training and model prediction; wherein:
a. the input layer is configured to seed grain refinement the historical data of all relevant variables of the burst;
b. the hidden layer consists of a plurality of long-period memory model nodes h (t) Is composed of a forgetting door f (t) Input door (i) (t) ANDa (t) ) Output door o (t) And memory cell C (t) A function;
c. and the output layer outputs the grain size refinement index of the aluminum hydroxide seed crystal from the last layer of the hidden layer through the full-connection layer.
4. The prediction method of the aluminum hydroxide seed grain refinement explosion prediction model based on LSTM according to claim 1, which is characterized by constructing an aluminum hydroxide seed grain refinement explosion prediction model, collecting historical data and latest data of all relevant variables of the prediction seed grain refinement explosion, processing the data according to the steps S2-S5, and constructing a prediction model Input data Input according to the prediction step number q finally determined when the prediction model is established; and inputting data by using a prediction model to obtain a prediction result, and carrying out inverse normalization on the prediction result to obtain a final prediction result.
5. The prediction method of the LSTM-based aluminum hydroxide seed particle size refinement burst prediction model according to claim 4, characterized by constructing a prediction model Input data Input:
wherein m is the number of prediction related variables, n is the data length, and x 1,n …x m,n Is the most current set of data.
6. The prediction method of LSTM-based aluminum hydroxide seed particle size refinement explosion prediction model according to claim 4, wherein the prediction result Y predict =[y 1 … y q ]。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011106571.5A CN114368768B (en) | 2020-10-15 | 2020-10-15 | LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011106571.5A CN114368768B (en) | 2020-10-15 | 2020-10-15 | LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114368768A CN114368768A (en) | 2022-04-19 |
CN114368768B true CN114368768B (en) | 2023-10-24 |
Family
ID=81137992
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011106571.5A Active CN114368768B (en) | 2020-10-15 | 2020-10-15 | LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114368768B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117668740A (en) * | 2024-02-02 | 2024-03-08 | 浙江晶盛机电股份有限公司 | Sapphire long-speed abnormality detection method, device, electronic device and storage medium |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1974395A (en) * | 2005-06-23 | 2007-06-06 | 中国铝业股份有限公司 | Seed separation granularity controlling method in alumina producing Bayer process |
CN101070169A (en) * | 2007-05-30 | 2007-11-14 | 山东铝业股份有限公司 | Method for resolving aluminium-oxide granularity periodic and explosive refining |
CN101654270A (en) * | 2009-09-10 | 2010-02-24 | 沈阳工业大学 | Method for eliminating periodic thinning of granularity of seed product |
CN108764568A (en) * | 2018-05-28 | 2018-11-06 | 哈尔滨工业大学 | A kind of data prediction model tuning method and device based on LSTM networks |
CN110909953A (en) * | 2019-12-03 | 2020-03-24 | 浙江科技学院 | Parking position prediction method based on ANN-LSTM |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7771681B2 (en) * | 2006-12-29 | 2010-08-10 | Nalco Company | Method for improved aluminum hydroxide production |
KR20180066456A (en) * | 2016-12-09 | 2018-06-19 | 주식회사 삼화양행 | Manufacturing method for high density bead of high purity alumina |
-
2020
- 2020-10-15 CN CN202011106571.5A patent/CN114368768B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1974395A (en) * | 2005-06-23 | 2007-06-06 | 中国铝业股份有限公司 | Seed separation granularity controlling method in alumina producing Bayer process |
CN101070169A (en) * | 2007-05-30 | 2007-11-14 | 山东铝业股份有限公司 | Method for resolving aluminium-oxide granularity periodic and explosive refining |
CN101654270A (en) * | 2009-09-10 | 2010-02-24 | 沈阳工业大学 | Method for eliminating periodic thinning of granularity of seed product |
CN108764568A (en) * | 2018-05-28 | 2018-11-06 | 哈尔滨工业大学 | A kind of data prediction model tuning method and device based on LSTM networks |
CN110909953A (en) * | 2019-12-03 | 2020-03-24 | 浙江科技学院 | Parking position prediction method based on ANN-LSTM |
Also Published As
Publication number | Publication date |
---|---|
CN114368768A (en) | 2022-04-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108629144B (en) | Bridge health assessment method | |
Dai et al. | Hybrid parallel stochastic configuration networks for industrial data analytics | |
CN112987664B (en) | Flow shop scheduling method based on deep reinforcement learning | |
CN109670625B (en) | NOx emission concentration prediction method based on unscented Kalman filtering least square support vector machine | |
CN113591374B (en) | Cement finished product specific surface area prediction method based on convolution simple circulation unit network | |
CN111310348A (en) | Material constitutive model prediction method based on PSO-LSSVM | |
CN112317110A (en) | Ore grinding granularity prediction system and method based on deep learning | |
CN114368768B (en) | LSTM-based aluminum hydroxide seed crystal granularity refinement burst prediction model and method | |
CN110045606A (en) | A kind of increment space-time learning method for distributed parameter system line modeling | |
CN111832703B (en) | Irregular sampling dynamic sequence modeling method for process manufacturing industry | |
CN111985825A (en) | Crystal face quality evaluation method for roller mill orientation instrument | |
CN106547899B (en) | Intermittent process time interval division method based on multi-scale time-varying clustering center change | |
CN113203953A (en) | Lithium battery residual service life prediction method based on improved extreme learning machine | |
Xiong et al. | Fuzzy pruning based LS-SVM modeling development for a fermentation process | |
CN111444963A (en) | Blast furnace molten iron silicon content prediction method based on SSA-SVR model | |
CN116646022A (en) | Prediction method, device, equipment and medium for organic solid waste pyrolysis tristate product | |
CN116484275A (en) | Aluminum electrolysis anode effect monitoring method based on self-adaptive space-time diagram convolutional neural network | |
CN111066562A (en) | Grape downy mildew prediction method and system | |
CN113377075B (en) | Method and device for real-time optimization of rare earth extraction process and computer readable storage medium | |
CN115525697A (en) | Process optimization method based on traditional Chinese medicine production data mining | |
CN112364527B (en) | Debutanizer soft measurement modeling method based on ALIESN online learning algorithm | |
CN110794809B (en) | VSI EWMA control chart economic design method based on preventive maintenance and quality loss function | |
CN113191614A (en) | Polypropylene product quality real-time monitoring method based on slow characteristic regression analysis | |
Zhang et al. | Anode Effect prediction based on Expectation Maximization and XGBoost model | |
Gong et al. | Multi-class classification method based on support vector machine with hyper-sphere for steel surface defects |
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 |