CN107974696A - Aluminium cell production process Anodic effect forecast method - Google Patents
Aluminium cell production process Anodic effect forecast method Download PDFInfo
- Publication number
- CN107974696A CN107974696A CN201610985259.5A CN201610985259A CN107974696A CN 107974696 A CN107974696 A CN 107974696A CN 201610985259 A CN201610985259 A CN 201610985259A CN 107974696 A CN107974696 A CN 107974696A
- Authority
- CN
- China
- Prior art keywords
- aluminium cell
- feature vector
- measured
- effect
- aluminium
- 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.)
- Granted
Links
Classifications
-
- C—CHEMISTRY; METALLURGY
- C25—ELECTROLYTIC OR ELECTROPHORETIC PROCESSES; APPARATUS THEREFOR
- C25C—PROCESSES FOR THE ELECTROLYTIC PRODUCTION, RECOVERY OR REFINING OF METALS; APPARATUS THEREFOR
- C25C3/00—Electrolytic production, recovery or refining of metals by electrolysis of melts
- C25C3/06—Electrolytic production, recovery or refining of metals by electrolysis of melts of aluminium
- C25C3/20—Automatic control or regulation of cells
Landscapes
- Chemical & Material Sciences (AREA)
- Engineering & Computer Science (AREA)
- Chemical Kinetics & Catalysis (AREA)
- Electrochemistry (AREA)
- Materials Engineering (AREA)
- Metallurgy (AREA)
- Organic Chemistry (AREA)
- Electrolytic Production Of Metals (AREA)
- Electrolytic Production Of Non-Metals, Compounds, Apparatuses Therefor (AREA)
Abstract
The invention discloses a kind of aluminium cell production process Anodic effect forecast method, is specially：Sample data is extracted from the creation data of aluminium cell；Processing is weighted to the feature vector of sample data using feature weight；Sample data is trained using support vector machines, obtains SVM effect forecasts model and model parameter；Extract the feature vector of aluminium cell to be measured and make weighting processing；The feature vector of aluminium cell to be measured is calculated with a distance from optimal hyperlane；If the feature vector of aluminium cell to be measured is more than or equal to predetermined threshold with a distance from optimal hyperlane, feature vector is sent into SVM effect forecast models, obtains forecast result；Otherwise, aluminium cell to be measured is forecast using k-nearest neighbor KNN.Effective forecast of different electrolytic cells anode effect under the conditions of various groove conditions can be achieved in the present invention, is conducive to stablize electrolytic cell operation, improves current efficiency.
Description
Technical field
The invention belongs to field of aluminum electrolysis, and in particular to a kind of aluminium cell production process Anodic effect forecast method.
Background technology
Anode effect is that a kind of phenomenon on anode occurs in Molten.When anode effect occurs, sun
Flashing around pole, bath voltage rise to tens of volts, power consumption increase, and aluminum amount and yield all decline, serious danger
The normal operation of evil electrolytic cell stability and aluminum electrolysis.Occur in order to avoid the accident of anode effect, to aluminium cell
Anode effect detects and forecast has great importance.
For aluminium cell when closing on anode effect, anodic current density reaches critical current density.Critical current density
Size represents the complexity that anode effect occurs, and numerical value is bigger, the easier generation of anode effect；Conversely, numerical value is smaller, anode
Effect is less susceptible to occur.Critical current density has substantial connection with alumina concentration, and critical current density is dense with aluminium oxide
Degree is reduced and reduced, thus when alumina concentration is relatively low, the easier generation anode effect of aluminium cell.
The cell resistance with alumina concentration of aluminium cell reduces and increases, since alumina concentration is unable to on-line measurement, because
This can reflect the change of alumina concentration with the change of cell resistance, and control and anode effect are pre- in real time as aluminium electrolysis process
The main feature of report.With cell resistance the caused error of potline current change can be reduced to analyze aluminium cell condition.In aluminium electricity
When solution groove is unstable, the noise jamming that the calculating of cell resistance is subject to is larger, only using Parameters Forecasting anodes such as tank voltage, cell resistances
Effect, which easily causes, to be failed to report or reports by mistake, therefore should increase other supplemental characteristics.
Prediction of Anode Effect main method is to carry out deep processing to retrievable data using INTELLIGENT IDENTIFICATION technology, mainly
The situation of anode electrolytic cell effect is forecast according to the signal magnitude of overall tank voltage.Prediction of Anode Effect research at this stage is also deposited
Prediction accuracy is high, the look-ahead time is shorter and the confidence level of model is relatively low the problems such as.
The content of the invention
In view of the problems of the existing technology, the present invention provides a kind of aluminium cell production process Anodic effect forecast side
Method, it is intended that accurate forecast can be carried out to the anode effect situation of the different electrolytic cells in different work areas, to be imitated to anode
The control answered, realize that aluminium cell low-voltage provides effective foundation without effect production, is conducive to stablize the operation of aluminium cell.
In order to realize the technology of the present invention purpose, the present invention provides a kind of aluminium cell production process Anodic effect forecast side
Method, including off-line training step and online forecasting stage；
The specific implementation step of the off-line training step is：
11) sample data is extracted from the creation data of multiple aluminium cells in multiple work areas；
12) processing is weighted to the feature vector of sample data using feature weight；
13) sample data is trained using support vector machines, obtains SVM effect forecasts model and obtain mould
Shape parameterAnd b*；
The specific implementation step in the online forecasting stage is：
21) feature vector of aluminium cell to be measured is extracted, processing is weighted to feature vector using feature weight；
22) feature vector of aluminium cell to be measured is calculated with a distance from optimal hyperlaneWherein, K
(xi, x) be support vector machines kernel function, x be aluminium cell weighting to be measured processing after feature vector, xiFor support for i-th to
Amount, n are supporting vector number, yiFor the classification of i-th of supporting vector, the y if anode effect does not occuri=-1, if anode effect occurs
Answer then yi=1；
If 23) feature vector of aluminium cell to be measured is more than or equal to predetermined threshold from optimal hyperlane distance g (x),
Feature vector after aluminium cell weighting to be measured is handled is sent into SVM effect forecast models, obtains forecast result；Otherwise, use
K-nearest neighbor KNN forecasts aluminium cell to be measured.
Further, the feature vector of the sample data includes cell resistance, tank voltage.
Further, it is horizontal, electric to further include electrolysis temperature, aluminum yield, molecular proportion, aluminium for the feature vector of the sample data
Solve matter level, feeding quantity.
The present invention contrasts existing Prediction of Anode Effect method, has following innovative point：
A. the forecast of aluminum cell anode effect is realized using support vector machines, in order to achieve this, needing from different works
Sample of each electrolytic cell extraction in area for Training Support Vector Machines, establishes multiple corresponding effect forecast models；
B. support vector machines is combined with KNN, and sample characteristics is weighted with processing, arrived according to feature vector to be sorted
The distance of optimal hyperlane, selects SVM effect forecasts model and KNN algorithms to treat characteristic of division vector and classify.
The present invention can forecast the anode effect situation of the different electrolytic cells in different work areas, effect forecast success rate
Up to more than 90 percent, for the control to anode effect, realize aluminium cell low-voltage without effect production provide effectively according to
According to being conducive to stablize the operation of aluminium cell, reach energy-saving, improve the effect of current efficiency.
Brief description of the drawings
Fig. 1 is the method for the present invention flow chart.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to the accompanying drawings and embodiments, it is right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in each embodiment of invention described below
Not forming conflict each other can be mutually combined.
As shown in Figure 1, a kind of aluminium cell production process Anodic effect forecast method of the present invention includes off-line training rank
Section and online forecasting stage：
The specific implementation step of the off-line training step is：
11) sample data is extracted from the Production database of multiple aluminium cells in multiple work areas, sample data is with groove
Resistance and tank voltage are main judging characteristic, and other supplemental characteristics include：Electrolysis temperature, aluminum yield, molecular proportion, aluminium are horizontal, electricity
Solve matter level, feeding quantity.In this step, according to a kind of better embodiment, also cell resistance is filtered and smoothing processing,
The sample to be standardized, and sample is normalized.
In example, the sample data that the present invention extracts includes normal sample and effect sample, effect sample extraction effect hair
Before the raw moment 0.5 it is small when data, when normal sample is extracted no generating effect day 8,14 when, 20 when data.
12) feature weight of sample is determined using sample data, sample data feature is weighted using feature weight
Processing.
Determining the feature weight of sample can use in feature selecting algorithm (Relief), Principal Component Analysis, Information Entropy etc.
Any one.Relief algorithms assign feature different weights according to the correlation of each feature and classification, and efficiency is very high,
Therefore the present invention carries out determining for sample characteristics weight using Relief algorithms.
13) sample training SVM effect forecast models are utilized.
Choosing is optimized to the penalty coefficient C and nuclear parameter g of model using grid-search algorithms (Grid Search) algorithm
Select.For SVM model parameters optimization method mainly have grid-search algorithms (Grid Search), particle cluster algorithm (PSO),
Genetic algorithm (GA) etc..Grid-search algorithms can simultaneously scan for multiple parameter values, to mutually independent parameter to being searched parallel
Rope, therefore the present invention carries out the optimum choice of model parameter using the method.
In order to ensure sample is linear separability, it is necessary to introduce nonlinear mapping function K (x), by the data of the input space
It is mapped to high dimensional attribute space.RBF kernel functions are selected, radial direction base RBF (Radial Basic Function) kernel function can be with
Sample is mapped to the space of a more higher-dimension, sample when being non-linear relation can be handled between class label and feature, and
Still there is good generalization ability in the case of priori poor information.
RBF kernel functions are selected, whole sample data set is trained using optimal parameter C and g, it is pre- to obtain SVM effects
Model is reported, obtains model parameterAnd b*And the set of supporting vector.
The specific implementation step in online forecasting stage is：
21) feature vector of aluminium cell to be measured is extracted, feature vector is weighted using the feature weight of step 12)
Processing.
22) feature vector of aluminium cell to be measured is calculated with a distance from optimal hyperlaneWherein, K
(xi, x) be support vector machines kernel function, x be aluminium cell weighting to be measured processing after feature vector, xiFor support for i-th to
Amount, n are supporting vector number, yiFor the classification of i-th of supporting vector, the y if anode effect does not occuri=-1, if anode effect occurs
Answer then yi=1.
23) if the feature vector of aluminium cell to be measured is more than or equal to predetermined threshold with a distance from optimal hyperlane, will treat
The feature vector surveyed after aluminium cell weighting processing is sent into SVM effect forecast models, obtains forecast result；Otherwise, use is most adjacent
Nearly algorithm (KNN) forecasts aluminium cell to be measured.
Aluminium cell to be measured is forecast using k-nearest neighbor (KNN), its basic thought is the weighted input spy that looks for novelty
The sign vector samples most adjacent with k in training set, classification of most classifications as new samples in this k sample, in view of KNN
Algorithm belongs to the prior art, and details are not described herein.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, all any modification, equivalent and improvement made within the spirit and principles of the invention etc., should all include
Within protection scope of the present invention.
Claims (3)
1. aluminium cell production process Anodic effect forecast method, it is characterised in that including off-line training step and pre- online
The report stage；
The specific implementation step of the off-line training step is：
11) sample data is extracted from the creation data of multiple aluminium cells in multiple work areas；
12) processing is weighted to the feature vector of sample data using feature weight；
13) sample data is trained using support vector machines, obtains SVM effect forecasts model and obtain model ginseng
NumberAnd b*；
The specific implementation step in the online forecasting stage is：
21) feature vector of aluminium cell to be measured is extracted, processing is weighted to feature vector using feature weight；
22) feature vector of aluminium cell to be measured is calculated with a distance from optimal hyperlaneWherein, K (xi,
X) be support vector machines kernel function, x be aluminium cell weighting to be measured processing after feature vector, xiFor i-th of supporting vector,
N is supporting vector number, yiFor the classification of i-th of supporting vector, the y if anode effect does not occuri=-1, if anode effect occurs
Then yi=1；
23) if the feature vector of aluminium cell to be measured is more than or equal to predetermined threshold from optimal hyperlane distance g (x), will treat
The feature vector surveyed after aluminium cell weighting processing is sent into SVM effect forecast models, obtains forecast result；Otherwise, use is most adjacent
Nearly algorithm KNN forecasts aluminium cell to be measured.
2. aluminium cell production process Anodic effect forecast method according to claim 1, it is characterised in that the sample
The feature vector of notebook data includes cell resistance, tank voltage.
3. aluminium cell production process Anodic effect forecast method according to claim 2, it is characterised in that the sample
The feature vector of notebook data further includes electrolysis temperature, aluminum yield, molecular proportion, aluminium level, electrolyte level, feeding quantity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610985259.5A CN107974696B (en) | 2016-10-25 | 2016-10-25 | Anode effect forecasting method in aluminum electrolysis cell production process |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610985259.5A CN107974696B (en) | 2016-10-25 | 2016-10-25 | Anode effect forecasting method in aluminum electrolysis cell production process |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107974696A true CN107974696A (en) | 2018-05-01 |
| CN107974696B CN107974696B (en) | 2020-06-26 |
Family
ID=62005198
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610985259.5A Active CN107974696B (en) | 2016-10-25 | 2016-10-25 | Anode effect forecasting method in aluminum electrolysis cell production process |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107974696B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111058061A (en) * | 2019-10-28 | 2020-04-24 | 上海大学 | Method for improving current efficiency of industrial aluminum electrolysis production |
| CN111763958A (en) * | 2020-08-24 | 2020-10-13 | 沈阳铝镁设计研究院有限公司 | Anode effect detection method based on anode guide rod vibration |
| CN111910217A (en) * | 2020-08-24 | 2020-11-10 | 常州机电职业技术学院 | High-efficiency intelligent control system for aluminum electrolysis production |
| CN114574905A (en) * | 2022-02-22 | 2022-06-03 | 北京科技大学 | Distributed multi-point blanking control method for aluminum electrolysis cell |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101498012A (en) * | 2008-01-30 | 2009-08-05 | 贵阳铝镁设计研究院 | Computer quenching method for aluminum cell anode effect |
| CN101967658A (en) * | 2010-11-18 | 2011-02-09 | 北方工业大学 | Aluminum cell anode effect prediction device |
| CN103668328A (en) * | 2013-12-14 | 2014-03-26 | 云南云铝润鑫铝业有限公司 | Device and method for eliminating anode effect of aluminum electrolysis cell |
| CN104619014A (en) * | 2015-01-09 | 2015-05-13 | 中山大学 | SVM-KNN (Support Vector Machine-K Nearest Neighbor)-based indoor positioning method |
-
2016
- 2016-10-25 CN CN201610985259.5A patent/CN107974696B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101498012A (en) * | 2008-01-30 | 2009-08-05 | 贵阳铝镁设计研究院 | Computer quenching method for aluminum cell anode effect |
| CN101967658A (en) * | 2010-11-18 | 2011-02-09 | 北方工业大学 | Aluminum cell anode effect prediction device |
| CN103668328A (en) * | 2013-12-14 | 2014-03-26 | 云南云铝润鑫铝业有限公司 | Device and method for eliminating anode effect of aluminum electrolysis cell |
| CN104619014A (en) * | 2015-01-09 | 2015-05-13 | 中山大学 | SVM-KNN (Support Vector Machine-K Nearest Neighbor)-based indoor positioning method |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111058061A (en) * | 2019-10-28 | 2020-04-24 | 上海大学 | Method for improving current efficiency of industrial aluminum electrolysis production |
| CN111763958A (en) * | 2020-08-24 | 2020-10-13 | 沈阳铝镁设计研究院有限公司 | Anode effect detection method based on anode guide rod vibration |
| CN111910217A (en) * | 2020-08-24 | 2020-11-10 | 常州机电职业技术学院 | High-efficiency intelligent control system for aluminum electrolysis production |
| CN114574905A (en) * | 2022-02-22 | 2022-06-03 | 北京科技大学 | Distributed multi-point blanking control method for aluminum electrolysis cell |
| CN114574905B (en) * | 2022-02-22 | 2023-11-24 | 北京科技大学 | Distributed multipoint blanking control method for aluminum electrolysis cell |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107974696B (en) | 2020-06-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111091143B (en) | Distribution transformer weight overload early warning method based on deep belief network and K-means clustering | |
| WO2022135265A1 (en) | Failure warning and analysis method for reservoir dispatching rules under effects of climate change | |
| CN107974696A (en) | Aluminium cell production process Anodic effect forecast method | |
| CN106251001A (en) | A kind of based on the photovoltaic power Forecasting Methodology improving fuzzy clustering algorithm | |
| CN111401460B (en) | Abnormal electric quantity data identification method based on limit value learning | |
| CN103390200A (en) | Photovoltaic power station electricity generation output power forecasting method based on similar days | |
| CN104376371B (en) | A kind of distribution based on topology is layered load forecasting method | |
| CN105184402B (en) | A kind of personalized user short-term load forecasting algorithm based on decision tree | |
| CN108133225A (en) | A kind of icing flashover fault early warning method based on support vector machines | |
| CN111104972A (en) | Method for identifying low-voltage risk of distribution room based on genetic algorithm optimization support vector machine multi-classifier | |
| CN109325631A (en) | Electric car charging load forecasting method and system based on data mining | |
| CN113435492B (en) | Power system dominant instability mode discrimination method based on active learning | |
| CN112270349B (en) | Individual position prediction method based on GCN-LSTM | |
| Fang et al. | A statistical approach to estimate imbalance-induced energy losses for data-scarce low voltage networks | |
| CN109558467A (en) | The recognition methods of electricity consumption class of subscriber and system | |
| CN104361365A (en) | Oil-well pump running condition recognition method and device | |
| Varghese et al. | Smart grid consumer behavioral model using machine learning | |
| CN117094754B (en) | Macroscopic and microscopic combined medium-long term electric quantity prediction method | |
| CN110570042A (en) | Short-term electric vehicle charging load prediction method and system | |
| CN110516849A (en) | A kind of load classification evaluation of result method based on typical day load curve | |
| CN113361776A (en) | Power load probability prediction method based on user power consumption behavior clustering | |
| CN109767353A (en) | A kind of photovoltaic power generation power prediction method based on probability-distribution function | |
| Cai et al. | Short-term load forecasting for city holidays based on genetic support vector machines | |
| Wang et al. | A Novel Multi‐Input AlexNet Prediction Model for Oil and Gas Production | |
| CN106960260B (en) | Wind power prediction system convenient for power dispatching |
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 |