CN114741838A - Method for predicting residual life of aluminum electrolytic cell - Google Patents

Method for predicting residual life of aluminum electrolytic cell Download PDF

Info

Publication number
CN114741838A
CN114741838A CN202210199501.1A CN202210199501A CN114741838A CN 114741838 A CN114741838 A CN 114741838A CN 202210199501 A CN202210199501 A CN 202210199501A CN 114741838 A CN114741838 A CN 114741838A
Authority
CN
China
Prior art keywords
degradation
state
hsmm
aluminum
model
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.)
Pending
Application number
CN202210199501.1A
Other languages
Chinese (zh)
Inventor
崔家瑞
苏成果
李擎
杨旭
李香泉
黄若愚
曹斌
路辉
唐伟荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guilin Honyuan Technology Co ltd
GUIYANG ALUMINUM MAGNESIUM DESIGN & RESEARCH INSTITUTE CO LTD
University of Science and Technology Beijing USTB
Original Assignee
Guilin Honyuan Technology Co ltd
GUIYANG ALUMINUM MAGNESIUM DESIGN & RESEARCH INSTITUTE CO LTD
University of Science and Technology Beijing USTB
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guilin Honyuan Technology Co ltd, GUIYANG ALUMINUM MAGNESIUM DESIGN & RESEARCH INSTITUTE CO LTD, University of Science and Technology Beijing USTB filed Critical Guilin Honyuan Technology Co ltd
Priority to CN202210199501.1A priority Critical patent/CN114741838A/en
Publication of CN114741838A publication Critical patent/CN114741838A/en
Pending legal-status Critical Current

Links

Images

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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/29—Graphical models, e.g. Bayesian networks
    • G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • 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
    • G06F2119/04—Ageing analysis or optimisation against ageing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Mathematical Analysis (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Algebra (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Operations Research (AREA)
  • Databases & Information Systems (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Electrolytic Production Of Metals (AREA)

Abstract

The invention provides a method for predicting the residual life of an aluminum electrolysis cell, belonging to the technical field of aluminum electrolysis. The method comprises the following steps: acquiring historical data capable of representing the degradation process of the aluminum electrolytic cell; training an HSMM model with all degradation states and an HSMM model with each degradation state by using the acquired historical data; wherein, HSMM represents a hidden semi-Markov model, and the probability distribution of state residence time in HSMM adopts Alron distribution; and identifying the current degradation state of the aluminum electrolysis cell by using all HSMM models in single degradation states obtained by training, and predicting the residual life of the aluminum electrolysis cell by using the HSMM models in all degradation states by using a forward algorithm added with state residence time according to the degradation state identification result. By adopting the method and the device, the prediction precision of the residual service life of the aluminum electrolytic cell can be improved.

Description

Method for predicting residual life of aluminum electrolysis cell
Technical Field
The invention relates to the technical field of aluminum electrolysis, in particular to a method for predicting the residual life of an aluminum electrolysis cell.
Background
With the continuous operation of the aluminum electrolytic cell in the production process, the electrolytic cell is continuously corroded in a complex electrolytic environment, the electrolytic cell is finally damaged, namely, a leakage cell occurs, and the service life of the electrolytic cell reaches the end point. The occurrence of the leakage groove not only seriously affects the safety of production personnel and equipment, but also causes huge economic loss of an aluminum plant due to the groove stopping and overhaul. In addition, some tanks may have a residual service Life (RUL) when planned to stop the tank for overhaul, and the shutdown caused by the advanced overhaul also causes economic loss. Therefore, the service life of the aluminum electrolysis cell is closely related to the economic benefit of an aluminum plant, and is one of the major concerns in the aluminum electrolysis industry at present.
Currently, mainstream life prediction methods are mainly classified into two categories, namely mechanism-based life prediction and data-based life prediction, wherein although prediction accuracy of a mechanism-based model is higher, a modeling process of the mechanism-based model needs a large amount of mechanism knowledge and related data and is often difficult to realize. The data-driven prediction method does not depend on the failure mechanism of the equipment, and needs to monitor the operation process of the equipment and collect effective failure data or performance degradation data. With the rapid advance of technologies combining fault Prediction and Health Management (PHM), data-driven life prediction is increasingly gaining attention. However, currently, the research for analyzing the service life of the aluminum electrolysis cell based on data is less, and the accuracy of the prediction result is low.
Hidden Markov Models (HMM) are used as a probability statistical method, are suitable for statistical modeling of random process time sequences, and are widely concerned in the aspects of speech recognition, life prediction, equipment degradation recognition and the like. Carey firstly migrates the HMM from the speech recognition field and applies the HMM to the equipment fault diagnosis field, and creates a brand new research direction of fault diagnosis; ma et al conclude that HMMs need to satisfy the irrationality of their markov property under the premise of obeying exponential distribution, and therefore cannot be directly used for life prediction of equipment.
Disclosure of Invention
The embodiment of the invention provides a method for predicting the residual life of an aluminum electrolytic cell, which can improve the prediction precision of the residual life of the aluminum electrolytic cell. The technical scheme is as follows:
the embodiment of the invention provides a method for predicting the residual life of an aluminum electrolytic cell, which comprises the following steps:
acquiring historical data capable of representing the degradation process of the aluminum electrolytic cell;
training an HSMM model with all degradation states and an HSMM model with each degradation state by using the acquired historical data; wherein, HSMM represents hidden semi-Markov model, and the probability distribution of state residence time in HSMM adopts Alron distribution;
and identifying the degradation state of the aluminum electrolysis cell by using all the trained HSMM models in the single degradation state, and predicting the residual life of the aluminum electrolysis cell by using the HSMM models with all the degradation states by using a forward algorithm added with state residence time according to the identification result of the degradation state.
Further, the obtaining historical data that can characterize the aluminum reduction cell degradation process comprises:
analyzing the degradation process of the aluminum electrolytic cell, determining parameters capable of representing the degradation process of the aluminum electrolytic cell, and acquiring historical data corresponding to the parameters; wherein the parameters include: the working voltage of the aluminum electrolytic cell, the iron content of the electrolyte and the silicon content of the electrolyte.
Further, before training with the acquired historical data to obtain an HSMM model with all degradation states and an HSMM model with each degradation state, the method comprises:
the probability distribution of the state residence time in the traditional HSMM is changed from an exponential distribution form to an Iranian distribution based on queuing theory.
Further, the acquired historical data capable of characterizing the aluminum reduction cell degradation process comprises: historical data of all degradation states of the aluminum electrolysis cell;
the training of the HSMM model with all degradation states and the HSMM model of each degradation state using the acquired historical data comprises:
training an HSMM model with all degradation states by using the acquired historical data of all degradation states of the aluminum electrolysis cell;
and training the HSMM model of the corresponding degradation state by using the acquired historical data of each degradation state of the aluminum electrolysis cell.
Further, the identifying the degradation state of the aluminum electrolysis cell by using the trained HSMM models in all single degradation states, and according to the identification result of the degradation state, the predicting the remaining life of the aluminum electrolysis cell by using the HSMM model with all degradation states by using a forward algorithm of adding state residence time comprises:
inputting the data to be tested into the HSMM model of each degradation state, and identifying the current degradation state of the aluminum electrolytic cell by calculating the maximum likelihood probability;
according to the degradation state identification result, the HSMM model with all degradation states determines the probability distribution of the remaining residence time of the aluminum electrolytic cell in the current degradation state by using a forward algorithm of adding state residence time;
and determining the remaining life of the aluminum electrolysis cell according to the determined probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state.
Further, the forward algorithm has a recurrence formula as follows:
Figure BDA0003526984860000031
wherein alpha ist+1(i, d) represents a degraded state at time t +1 as SiAnd the remaining dwell time of the degraded state is the probability of d; p is a radical of formulai(d) Indicating a state of degradation siProbability of lower duration being d; alpha is alphat(i, d +1) represents that the state of degradation at time t is SiAnd the probability that the remaining dwell time of the degraded state is d + 1;
Figure BDA0003526984860000032
indicating a degraded state SiThe time observation value is OtThe probability of (d); s. thet(i) Is represented by an observed value of OtThe remaining dwell time τ of the current state of degradation under the condition of (1)tIs 1 and the state at the next moment is SiThe probability of (c).
Further, the determining the remaining life of the aluminum reduction cell according to the determined remaining residence time probability distribution of the aluminum reduction cell in the current degradation state comprises:
the degradation state at the moment t +1 t +1 determined according to the forward algorithm is SiAnd the probability alpha of the remaining dwell time of the degraded state being dt+1(i, d) determining the current degradation state S at time tiOf the remaining residence time Dt(Si) Expressed as:
Figure BDA0003526984860000033
wherein D represents the maximum residence time;
according to D obtainedt(Si) Determining the degradation state S of the aluminum electrolysis celliThe rest after the dwell time t isEffective lifetime RUL is:
Figure BDA0003526984860000034
where N denotes the number of degenerate states, D (S)j) Representing a state of degradation sjThe remaining residence time of.
Further, after training the HSMM model with all degradation states and the HSMM model for each degradation state using the acquired historical data, the method includes:
and verifying the trained HSMM model with all degradation states and the HSMM model of each degradation state.
The technical scheme provided by the embodiment of the invention has the beneficial effects that at least:
in the embodiment of the invention, historical data capable of representing the degradation process of the aluminum electrolytic cell is obtained; modeling the degradation process of the aluminum electrolysis cell by adopting an HSMM model based on the acquired historical data, and adopting an Ellang distribution as the probability distribution of state residence time in the HSMM so as to improve the residual life prediction effect of the HSMM and improve the residual life prediction precision of the aluminum electrolysis cell; and an improved forward algorithm of adding state residence time is used during the life prediction, so that the calculation of the life prediction is simplified, the calculation efficiency is improved, and the accuracy of the HSMM prediction result is further improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic flow chart of a method for predicting the remaining life of an aluminum electrolysis cell according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of an aluminum electrolysis cell according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of an HSMM provided by an embodiment of the invention;
fig. 4 is a detailed flowchart of the method for predicting the remaining life of an aluminum electrolysis cell according to the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in detail with reference to the accompanying drawings.
Example one
As shown in fig. 1, an embodiment of the present invention provides a method for predicting remaining life of an aluminum electrolysis cell, including:
s101, acquiring historical data capable of representing the degradation process of the aluminum electrolytic cell;
in this embodiment, a schematic structural diagram of an aluminum electrolysis cell is shown in fig. 2. By analyzing the influence of the aluminum electrolysis production process on the aluminum electrolysis cell, the degradation of the aluminum electrolysis cell can be summarized into the following main reasons by combining various literatures and expert experiences:
(1) sodium infiltration: the main signals for cell bottom lining failure are the formation of coarse fluoride crystals and the cracking of the carbon block, and the formation of this cracking force is mainly due to the expansion of the osmotic crystals and the reaction of sodium with the electrolyte upon cell start-up.
(2) Air infiltration damages the liner by oxidation: because of the untight sealing, air enters the lining, and then sodium-carbon-air reaction is directly generated under the cathode lining, thereby continuously causing the oxidative corrosion of the lining.
(3) Electrolyte leakage: the refractory bricks at the bottom are corroded by the melt, the electrolyte leaks to melt the steel bar, and if the crack depth of the cathode carbon block reaches the top of the steel bar, high-temperature aluminum liquid or electrolyte liquid can continuously permeate, so that the anode steel bar is rapidly melted and corroded, and the groove leakage accident can be caused quickly.
(4) And (3) bulging of the cathode carbon block: when the roasting is started, the cathode bulges, the electrolyte permeates into the inner lining of the cathode carbon block to cause the volume expansion of the cathode carbon block, and the electrolyte infiltrates along cracks between carbon seams and blocks until the electrolyte infiltrates to the bottom of the cathode carbon block, so that the bulging is aggravated and the cracking of the carbon block is finally caused.
(5) Engineering construction and roasting start control: the construction technology of prebaking groove furnace building is unqualified, so that the current distribution is not uniform after electrification, local overheating is formed, and the cathode carbon block is deformed due to overheating to generate cracks. Or the individual cathode carbon blocks are heated unevenly in the roasting starting process of the pre-roasting tank, so that large-gap cracks are generated too early and the individual cathode carbon blocks are damaged.
The characteristics of the parameters related to the degradation process of the electrolytic cell are shown as follows: the temperature of the cathode steel bar rises gradually, the working voltage of the electrolytic cell rises gradually, the cathode voltage drop becomes large, the needle vibration phenomenon occurs in the electrolytic cell, the temperature of the cell shell rises gradually, the iron content and the silicon content in the electrolyte rise, the aluminum output quantity drops and the like.
Therefore, by analyzing the degradation process of the aluminum electrolytic cell, parameters capable of representing the degradation process of the aluminum electrolytic cell can be determined, and historical data corresponding to the parameters are obtained; wherein the parameters include: the working voltage of the aluminum electrolytic cell, the iron content of the electrolyte and the silicon content of the electrolyte.
In this embodiment, the failure range index of the key parameter for the degradation modeling of the aluminum electrolysis cell is obtained according to the expert experience and the actual statistical result, and is shown in table 1.
TABLE 1 aluminium cell degradation modeling key parameter failure range index
Figure BDA0003526984860000051
S102, training a Hidden Semi Markov Model (HSMM) with all degradation states and an HSMM model of each degradation state by using the acquired historical data;
in this embodiment, in order to predict the remaining life of the aluminum electrolysis cell, a model of the degradation process of the aluminum electrolysis cell needs to be first modeled, and specifically, the modeling of the degradation process of the aluminum electrolysis cell using the improved HSMM model includes: training an HSMM model with all degradation states by using the acquired historical data of all degradation states of the aluminum electrolysis cell; training an HSMM model of a corresponding degradation state by using the acquired historical data of each degradation state of the aluminum electrolytic cell; and then, realizing the prediction of the residual service life of the aluminum electrolytic cell by a certain prediction algorithm.
In this embodiment, in order to better understand the HSMM model, first, it is briefly described:
unlike hidden markov models, which correspond to one observation for one state, HSMM models can correspond to multiple observations, which can be considered as one observation unit. Assuming N observation units, define qnIs the time index of a single observation point in the nth observation unit (N is more than or equal to 1 and less than or equal to N), and N represents the number of the degradation states. Let the hidden state at time t be StO is the observation sequence corresponding to the state, and for the nth observation unit, these observations are
Figure BDA0003526984860000061
And they have the same microscopic state mark
Figure BDA0003526984860000062
hnShowing the nth degradation state, the block diagram of which is shown in fig. 3.
The simplified hidden semi-markov model can be described by 4 probability matrices, which can be represented as:
λ＝(π，A，B，P) (1)
wherein pi represents the initial state probability distribution vector of the aluminum electrolysis cell, pi ═ pi (pi)1，π2，π3，...，πN)，πiDenotes qtThe time model is in a degraded state SiI.e.:
πi＝P(qt＝Si)，1≤i≤N，
Figure BDA0003526984860000063
a represents a probability matrix (short for: state transition matrix) of mutual transition among various degradation states of the aluminum electrolysis cell,A＝{aij}，1≤i，j≤N。aijrepresenting a degraded state siTransition to degenerate state sjI.e.:
aij＝P(qt＝Sj|qt-1＝Si)，
Figure BDA0003526984860000064
b represents probability matrix of observation value of aluminum cell (observation value probability matrix for short), B ═ Bi(Vm)}，1≤i≤N，1≤m≤M。bi(Vm) Representing a state of degradation siThe time observation value is VmI.e.:
bi(Vm)＝P(Ot＝Vm|qt＝Si)，1≤i≤N，
Figure BDA0003526984860000065
wherein M represents the number of observed values, OtRepresents an observed value, P (O)t＝Vm|qt＝Si) Denotes qtThe time of day model is in a degraded state siUnder the condition of (1) an observed value of VmThe probability of (d);
p represents the probability distribution matrix of the state duration of the aluminum electrolysis cell (short for: the state residence time probability distribution matrix):
Pi(d)＝P(d|qt＝Si)
，1≤i≤N，1≤d≤D (5)
where D represents the maximum time that any one of the degradation states may last (shortly: maximum dwell time), Pi(d) Indicating a state of degradation siThe probability that the lower duration is d (i.e., the state dwell time d), P (d | q)tI) denotes qtThe time model is in a degraded state SiWith a probability of duration d in this degraded state.
In the hidden semi-markov model, the distribution of the state residence time is generally selected as an exponential distribution, and the effect is not ideal in some cases from the practical effect. In order to improve the prediction effect of the HSMM, in the embodiment, the probability distribution of the state residence time in the conventional HSMM is changed from an exponential distribution form to an erlang distribution based on the queuing theory, so as to improve the accuracy of the residual life prediction.
In probability statistics, an Erlang distribution is a continuous probability distribution that, like an exponential distribution, can be used to represent the time intervals at which independent random events occur. The Erlang distribution is widely used in queuing theory, and changing the parameters of the Erlang distribution can approximate almost all continuous distributions, relative to exponential distributions, and thus can better fit realistic data.
If k random variables XiI 1, 2.. k, respectively, are subject to an exponential distribution, then the random variable X1+X2+...+XkObeying an Erlang distribution. Namely: the random variable with an Erlang distribution of order k can be viewed as the sum of the independent k random variables with the same exponential distribution.
The probability density f (x; k, μ) of the Erlang distribution is:
Figure BDA0003526984860000071
wherein k represents an order and is a positive integer; mu represents 1/lambda, lambda represents an exponential distribution parameter, mu > 0, and the mean and variance are k mu and k mu, respectively2(ii) a x is an independent variable.
Xi (i, J, d) is defined as the degradation state S of the aluminum cell at the time t-1iThe degradation state at time t is SjAnd the observed observation value under the condition that the residence time is d is O1，O2，...，OTProbability xi oft(i，j，d)：
ξt(i，j，d)＝P(qt＝Si，qt+1＝Sj|O，Ci＝d) (7)
Wherein, CiIs shown in a degraded state SiThe duration of time;
then, when performing parameter estimation on the data, the mean μ (i) and the variance σ (i) can be calculated by the following two equations:
Figure BDA0003526984860000072
Figure BDA0003526984860000073
wherein T represents the total time;
the average residence time of the state is then:
Figure BDA0003526984860000074
defining the degradation state of the model at time t as SiThe probability of (D) can be recorded as gammat(i) Namely:
Figure BDA0003526984860000081
wherein ξt(i, j) represents that the state of the aluminum electrolytic cell at the time t-1 is SiThe state at time t is SjThe probability of (d); according to the definition of the model parameters, other parameters of the model are updated by using intermediate variables, and the updating formula is as follows:
πiwhen initial time t is 1, the model is in state SiIs equal to gamma1(i) (12)
Figure BDA0003526984860000082
Figure BDA0003526984860000083
Wherein the content of the first and second substances,
Figure BDA0003526984860000084
indicating a degraded state SjThe time observation value is VmThe probability of (d); the form δ (x, y) is a decision function, with the output being 1 when x equals y, otherwise 0.
S103, identifying the degradation state of the aluminum electrolysis cell by using all the trained HSMM models in the single degradation state, and predicting the residual life of the aluminum electrolysis cell by using the HSMM models in all the degradation states by using a forward algorithm added with state residence time according to the identification result of the degradation state, wherein the method specifically comprises the following steps:
inputting the data to be tested into the HSMM model of each degradation state, and identifying the current degradation state of the aluminum electrolytic cell by calculating the maximum likelihood probability;
according to the degradation state identification result, the HSMM model with all degradation states determines the probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state by using a forward algorithm of adding state residence time;
and determining the remaining life of the aluminum electrolytic cell according to the determined probability distribution of the remaining residence time of the aluminum electrolytic cell in the current degradation state.
In this embodiment, the hidden semi-markov model is also used for lifetime prediction, and three problems, namely, the recognition problem, the decoding problem, and the learning problem need to be solved. To solve the above three problems, a more classical algorithm is usually used for solving, which are a Forward-backward algorithm (Forward-backward algorithm), a Viterbi algorithm, and a Baum-Welch algorithm, respectively, wherein the derivation of the Forward-backward algorithm is the basis of the following two algorithms.
In this embodiment, the prediction of the remaining lifetime is implemented by using an improved forward algorithm, wherein the improved forward variable is defined as:
Figure BDA0003526984860000085
forward variableαt(i, d) characterisation in a known observation sequence OtUnder the condition that the degradation state at time t is SiAnd the remaining dwell time tau of the degraded statetA probability of d;
the recursion formula of the forward algorithm is as follows:
Figure BDA0003526984860000091
wherein alpha ist+1(i, d) represents a degraded state at time t +1 as SiAnd the remaining dwell time of the degraded state is the probability of d; p is a radical ofi(d) Indicating a degraded state SiProbability of lower duration being d; alpha (alpha) ("alpha")t(i, d +1) represents that the state of degradation at time t is SiAnd the probability that the remaining dwell time of the degraded state is d + 1;
Figure BDA0003526984860000092
indicating a degraded state SiThe time observation value is OtThe probability of (d); st(i) Is represented by an observed value of OtThe remaining dwell time τ of the current state of degradation under the condition of (1)tIs 1 and the state at the next moment is SiThe probability of (d); st(i) And
Figure BDA0003526984860000093
is an intermediate calculation value of the forward algorithm. Therefore, at any time t, the current state S of the aluminum electrolysis cell can be solved based on the model parameters and the known observation sequencetThe distribution probability of the remaining residence time.
At the time t, the maximum likelihood probability is calculated according to a Viterbi algorithm, and then the degradation state q of the aluminum electrolytic cell at present can be estimatedt＝Si(ii) a Then, the degradation state at the moment t +1 is obtained as S according to a forward algorithmiAnd the probability alpha of the remaining dwell time of the degraded state being dt+1(i, D) determining the remaining dwell time D of the current state of degradation at time tt(Si) Expressed as:
Figure BDA0003526984860000094
according to D obtainedt(Si) Determining the degradation state S of the aluminum electrolytic celliThe remaining useful life RUL after the dwell time t is:
Figure BDA0003526984860000095
where N denotes the number of degenerate states, D (S)j) Indicating a degraded state SjThe remaining residence time of.
As shown in fig. 4, the method for predicting the remaining life of the aluminum electrolysis cell based on the HSMM comprises the following steps:
Figure BDA0003526984860000096
Figure BDA0003526984860000101
in this embodiment, after training the HSMM model having all degradation states and the HSMM model of each degradation state using the acquired historical data, the method includes:
and verifying the trained HSMM model with all degradation states and the HSMM model of each degradation state.
In this example, the actual data of a certain electrolytic aluminum plant was used for verification, specifically:
the daily report data and the laboratory data actually recorded in the electrolytic aluminum workshop production are selected, and two electrolytic tanks (2605 tank, 2616 tank of 6 work areas, 2720 tank of 7 work areas and 2732 tank of 7 work areas respectively) of two work areas subjected to overhaul are selected from the data. Because the set voltage, the working mode and the like of different work areas are different, the modeling analysis of the aluminum electrolysis cells of different work areas is needed.
3.1 degradation Process modeling of aluminum reduction cells
Selecting three parameters of the working voltage of the aluminum electrolytic cell, the iron content of the electrolyte and the silicon content of the electrolyte, which can represent the degradation process of the aluminum electrolytic cell, as model training data for preprocessing aiming at daily report data and assay data; among these, cell 2720 and cell 2605 recorded data from the start to days 837 and 771, when the aluminum electrolysis cell was stopped. And (3) carrying out variable point detection and segmented clustering on the preprocessed data and processing by using a maximum and minimum redundancy algorithm, and finally determining that the aluminum electrolysis cell experiences 4 degradation states in total.
Before modeling, initial parameters of the model are determined, and normally because the aluminum electrolytic cell is in an undegraded state at an initial moment, an initial state probability distribution vector pi is [1,0,0,0 ]. Since the degradation of the aluminum reduction cell can only be transferred from the current state to the next degradation state or stay in the current state, and not be transferred to the previous degradation state, i.e. the HSMM model of the left-right type, the initial state transfer matrix can be defined as the following form:
Figure BDA0003526984860000102
generally, the determination of the observation value probability matrix B is troublesome, and in this embodiment, the initial value is estimated by using a K-means clustering method. The method is characterized in that divided data in one state are gathered into F classes by using a K-means clustering algorithm, and then mean vectors and covariance matrixes of the data in the F classes are respectively calculated, so that F normal distribution parameters are obtained. And dividing the total number of the state data by the number of the data contained in each class to obtain the mixing coefficient of the density functions of various classes. Finally, the probability density function of a certain state can be obtained by linearly adding the F normal distribution functions. The state residence time probability distribution matrix P is estimated using the Erlang distribution and using equations (6) to (10).
The 2720 tank data and 2605 tank data are input into the HSMM respectively for training to obtain model output parameters respectively, so that a degradation model of the complete data cycle of the electrolytic cell is obtained, the state transition paths of the two tanks respectively experience 4 states of a degradation state 2, a degradation state 3 and a degradation state 4 from the initial stable operation stage (called as a degradation state 1) to the last data point, and the aluminum electrolytic cell stops to reach the end point of the service life. Meanwhile, the HSMM model corresponding to each degradation state is trained by using the data of the four degradation states of each electrolytic cell respectively, so that the verification data/data to be tested can be conveniently input into each degradation model in the following process, the degradation state of the verification data/data to be tested is determined, and the residual life of the aluminum electrolytic cell can be predicted by using a forward algorithm added with the state residence time through the degradation state and the state residence time.
3.2 prediction of remaining Life of aluminum reduction cells
Selecting data of the 2732 cell and the 2616 cell as verification sets, inputting the verification data into a trained HSMM model, judging the degradation state of the aluminum electrolysis cell by comparing the maximum likelihood probability value, and then calculating the residual life of the aluminum electrolysis cell in the current state according to the current degradation state and the state residence time by using an equation (17) and an equation (18). By verification, we get: the prediction accuracy of the method for predicting the residual life of the aluminum electrolytic cell provided by the embodiment of the invention is more accurate than the result of the conventional aluminum electrolytic cell life prediction.
The method for predicting the residual life of the aluminum electrolytic cell, provided by the embodiment of the invention, comprises the steps of obtaining historical data capable of representing the degradation process of the aluminum electrolytic cell; based on the acquired historical data, an HSMM model is adopted to model the degradation process of the aluminum electrolysis cell, and in order to improve the residual life prediction effect of the HSMM, the probability distribution of state residence time in the HSMM adopts Alron distribution so as to improve the residual life prediction accuracy of the aluminum electrolysis cell; and an improved forward algorithm of adding state residence time is used during the life prediction, so that the calculation of the life prediction is simplified, the calculation efficiency is improved, and the accuracy of the HSMM prediction result is further improved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (8)

1. A method for predicting the residual life of an aluminum electrolysis cell is characterized by comprising the following steps:
acquiring historical data capable of representing the degradation process of the aluminum electrolytic cell;
training an HSMM model with all degradation states and an HSMM model with each degradation state by using the acquired historical data; wherein, HSMM represents hidden semi-Markov model, and the probability distribution of state residence time in HSMM adopts Alron distribution;
and identifying the current degradation state of the aluminum electrolysis cell by using all HSMM models in single degradation states obtained by training, and predicting the residual life of the aluminum electrolysis cell by using the HSMM models in all degradation states by using a forward algorithm added with state residence time according to the degradation state identification result.
2. The method for predicting the remaining life of an aluminum reduction cell according to claim 1, wherein the obtaining historical data characterizing the degradation process of the aluminum reduction cell comprises:
analyzing the degradation process of the aluminum electrolytic cell, determining parameters capable of representing the degradation process of the aluminum electrolytic cell, and acquiring historical data corresponding to the parameters; wherein the parameters include: the working voltage of the aluminum electrolytic cell, the iron content of the electrolyte and the silicon content of the electrolyte.
3. The method for predicting the remaining life of an aluminum reduction cell according to claim 1, wherein before training a HSMM model having all degradation states and an HSMM model for each degradation state using the acquired historical data, the method comprises:
the probability distribution of the state residence time in the traditional HSMM is changed from an exponential distribution form to an Iranian distribution based on queuing theory.
4. The method for predicting the residual life of the aluminum reduction cell according to claim 1, wherein the acquired historical data capable of characterizing the degradation process of the aluminum reduction cell comprises: historical data of all degradation states of the aluminum electrolysis cell;
the training of the HSMM model with all degradation states and the HSMM model of each degradation state using the acquired historical data comprises:
training an HSMM model with all degradation states by using the acquired historical data of all degradation states of the aluminum electrolysis cell;
and training the HSMM model of the corresponding degradation state by using the acquired historical data of each degradation state of the aluminum electrolysis cell.
5. The method for predicting the residual life of the aluminum electrolysis cell according to claim 1, wherein the step of identifying the current degradation state of the aluminum electrolysis cell by using the trained HSMM models in all single degradation states and predicting the residual life of the aluminum electrolysis cell by using a forward algorithm added with state residence time according to the degradation state identification result by using the HSMM models in all degradation states comprises the steps of:
inputting the data to be tested into the HSMM model of each degradation state, and identifying the current degradation state of the aluminum electrolytic cell by calculating the maximum likelihood probability;
according to the degradation state identification result, the HSMM model with all degradation states determines the probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state by using a forward algorithm of adding state residence time;
and determining the remaining life of the aluminum electrolysis cell according to the determined probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state.
6. The method for predicting the remaining life of an aluminum electrolysis cell according to claim 5, wherein the forward algorithm has a recursion formula as follows:
Figure FDA0003526984850000021
wherein alpha ist+1(i, d) represents a degraded state at time t +1 as SiAnd the remaining dwell time of the degraded state is the probability of d; p is a radical ofi(d) Indicating a degraded state SiProbability of lower duration being d; alpha (alpha) ("alpha")t(i, d +1) represents that the state of degradation at time t is SiAnd the probability that the remaining dwell time of the degraded state is d + 1;
Figure FDA0003526984850000022
indicating a degraded state SiThe time observation value is OtThe probability of (d); st(i) Is represented by an observed value of OtThe remaining dwell time τ of the current state of degradation under the condition of (1)tIs 1 and the state at the next time is SiThe probability of (c).
7. The method for predicting the remaining life of the aluminum reduction cell as recited in claim 5, wherein the determining the remaining life of the aluminum reduction cell according to the determined remaining residence time probability distribution of the aluminum reduction cell in the current degradation state comprises:
the degradation state at the moment t +1 t +1 determined according to the forward algorithm is SiAnd the probability alpha of the remaining dwell time of the degraded state being dt+1(i, d) determining the current degradation state S at time tiResidual residence time D oft(Si) Expressed as:
Figure FDA0003526984850000023
wherein D represents the maximum residence time;
according to D obtainedt(Si) Determining the degradation state S of the aluminum electrolysis celliThe remaining useful life RUL after the dwell time t is:
Figure FDA0003526984850000024
where N denotes the number of degenerate states, D (S)j) Indicating a degraded state SjThe remaining residence time of the catalyst.
8. The method of predicting remaining life of an aluminum reduction cell according to claim 1, wherein after training the HSMM model having all degradation states and the HSMM model for each degradation state using the acquired historical data, the method comprises:
and verifying the trained HSMM model with all degradation states and the HSMM model of each degradation state.
CN202210199501.1A 2022-03-01 2022-03-01 Method for predicting residual life of aluminum electrolytic cell Pending CN114741838A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210199501.1A CN114741838A (en) 2022-03-01 2022-03-01 Method for predicting residual life of aluminum electrolytic cell

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210199501.1A CN114741838A (en) 2022-03-01 2022-03-01 Method for predicting residual life of aluminum electrolytic cell

Publications (1)

Publication Number Publication Date
CN114741838A true CN114741838A (en) 2022-07-12

Family

ID=82275660

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210199501.1A Pending CN114741838A (en) 2022-03-01 2022-03-01 Method for predicting residual life of aluminum electrolytic cell

Country Status (1)

Country Link
CN (1) CN114741838A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115310373A (en) * 2022-10-11 2022-11-08 国网浙江省电力有限公司电力科学研究院 Method for predicting residual life of hydrogen production electrolytic cell

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115310373A (en) * 2022-10-11 2022-11-08 国网浙江省电力有限公司电力科学研究院 Method for predicting residual life of hydrogen production electrolytic cell
CN115310373B (en) * 2022-10-11 2023-05-23 国网浙江省电力有限公司电力科学研究院 Hydrogen production electrolytic tank residual life prediction method

Similar Documents

Publication Publication Date Title
CN109814527B (en) Industrial equipment fault prediction method and device based on LSTM recurrent neural network
CN109086804B (en) Hydraulic equipment early failure prediction method based on fusion of multi-source state monitoring information and reliability characteristics
CN114114049B (en) Lithium ion battery life prediction method based on sample migration
WO2021073343A1 (en) Method and apparatus for analyzing root cause of failure of communication system, system and computer storage medium
CN109117566B (en) Mixed maintenance planning method based on product life prediction model
CN107133400A (en) A kind of aircraft structure fatigue reliability Bayes's combination forecasting method
CN106372299B (en) A kind of pipeline method for predicting residual useful life based on inverse Gaussian process and EM-PF
CN109919394B (en) Method for predicting residual life of power transformer
CN113378939B (en) Structure digital twin modeling and parameter identification method based on physical driving neural network
CN114741838A (en) Method for predicting residual life of aluminum electrolytic cell
CN109145373B (en) Residual life prediction method and device based on improved ESGP and prediction interval
CN113205233A (en) Lithium battery life prediction method based on wolf algorithm and multi-core support vector regression
CN116862079B (en) Enterprise pollutant emission prediction method and prediction system
CN114462305A (en) Argo buoy long-endurance energy consumption optimization and life estimation method
Hatherall et al. Load prediction based remaining discharge energy estimation using a combined online and offline prediction framework
CN113269356B (en) Missing data-oriented equipment residual life prediction method and system
CN114943281A (en) Intelligent decision-making method and system for heat pipe cooling reactor
Cui et al. The prediction of remaining useful life of aluminum reduction cells based on improved hidden semi-Markov model
CN115101136A (en) Large-scale aluminum electrolysis cell global anode effect prediction method
Chen et al. Anode effect prediction based on collaborative two-dimensional forecast model in aluminum electrolysis production.
CN112836380A (en) Fault estimation and maintenance method for intelligent manufacturing production line
Liao et al. An improved prediction model for equipment performance degradation based on Fuzzy-Markov Chain
CN112485676B (en) Battery energy storage system state estimation early warning method under digital mirror image
Zhang et al. A Hidden Markov Model based online system reliability estimating of fluorochemical engineering processes
CN112801398B (en) Cooling device failure prediction method and device, electronic equipment and storage medium

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