A method for predicting the remaining life of an aluminum electrolytic cell

Abstract

The present invention provides a method for predicting the remaining life of an aluminum electrolysis cell, belonging to the technical field of aluminum electrolysis. The method includes: obtaining historical data that can characterize the degradation process of the aluminum electrolysis cell; using the obtained historical data to train an HSMM model with all degradation states and an HSMM model for each degradation state; wherein, HSMM represents a hidden semi-Markov model, and the probability distribution of the state residence time in the HSMM adopts an Erlang distribution; using the trained HSMM models of all single degradation states to identify the current degradation state of the aluminum electrolysis cell, and according to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with the added state residence time to predict the remaining life of the aluminum electrolysis cell. By adopting the present invention, the prediction accuracy of the remaining life of the aluminum electrolysis cell can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of aluminum electrolysis, and in particular to a method for predicting the remaining life of an aluminum electrolysis cell. Background Art

[0002] As aluminum reduction cells continue to operate during production, the complex electrolysis environment causes them to corrode continuously, eventually leading to damage and failure. This is known as leakage, marking the end of the cell's service life. Leakage not only seriously impacts the safety of production personnel and equipment, but also results in significant economic losses for the aluminum smelter due to the need to shut down the cell for major repairs. Furthermore, some cells may still have a remaining useful life (RUL) when scheduled for overhaul, and premature overhauls resulting in production stoppages can also result in economic losses. Therefore, the lifespan of aluminum reduction cells is closely linked to the economic profitability of aluminum smelters and is currently a key concern within the aluminum electrolysis industry.

[0003] Current mainstream life prediction methods are categorized into two main categories: mechanism-based and data-based. While mechanism-based models offer higher accuracy, their modeling requires extensive knowledge of the mechanisms and relevant data, making them often difficult to implement. Data-driven prediction methods, which do not rely on the failure mechanism of the equipment, require monitoring the equipment's operating process and collecting valid failure or performance degradation data. With the rapid advancement of prognostics and health management (PHM) technology, data-driven life prediction is gaining increasing attention. However, current research on data-based analysis of aluminum electrolytic cell life is relatively limited, resulting in low prediction accuracy.

[0004] The Hidden Markov Model (HMM), a probabilistic statistical method suitable for statistical modeling of random process time series, has garnered widespread attention in speech recognition, lifespan prediction, and equipment degradation identification. Carey first applied the HMM from speech recognition to equipment fault diagnosis, pioneering a new research direction in fault diagnosis. Ma et al. deduced that the HMM's Markov property is irrational, requiring it to follow an exponential distribution. Therefore, it cannot be directly used for equipment lifespan prediction. Summary of the Invention

[0005] The embodiment of the present invention provides a method for predicting the remaining life of an aluminum electrolysis cell, which can improve the prediction accuracy of the remaining life of the aluminum electrolysis cell. The technical solution is as follows:

[0006] An embodiment of the present invention provides a method for predicting the remaining life of an aluminum electrolysis cell, comprising:

[0007] Obtain historical data that can characterize the degradation process of an aluminum electrolysis cell;

[0008] Use the obtained historical data to train an HSMM model with all degradation states and HSMM models for each degradation state; wherein, HSMM represents a hidden semi-Markov model, and the probability distribution of the state residence time in the HSMM adopts an Erlang distribution;

[0009] Use the trained HSMM models of all single degradation states to identify the current degradation state of the aluminum electrolysis cell. According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with the added state residence time to predict the remaining life of the aluminum electrolysis cell.

[0010] Furthermore, the obtaining of historical data that can characterize the degradation process of the aluminum electrolysis cell includes:

[0011] Analyze the degradation process of the aluminum electrolysis cell, determine the parameters that can characterize the degradation process of the aluminum electrolysis cell, and obtain the historical data corresponding to the parameters; wherein, the parameters include: the working voltage of the aluminum electrolysis cell, the iron content in the electrolyte, and the silicon content in the electrolyte.

[0012] Furthermore, before using the obtained historical data to train an HSMM model with all degradation states and HSMM models for each degradation state, the method includes:

[0013] Based on queuing theory, change the probability distribution of the state residence time in the traditional HSMM from an exponential distribution form to an Erlang distribution.

[0014] Furthermore, the obtained historical data that can characterize the degradation process of the aluminum electrolysis cell includes: the historical data of all degradation states of the aluminum electrolysis cell;

[0015] The using of the obtained historical data to train an HSMM model with all degradation states and HSMM models for each degradation state includes:

[0016] Use the obtained historical data of all degradation states of the aluminum electrolysis cell to train an HSMM model with all degradation states;

[0017] Use the obtained historical data of each degradation state of the aluminum electrolysis cell to train the HSMM model of the corresponding degradation state.

[0018] Furthermore, the using of the trained HSMM models of all single degradation states to identify the current degradation state of the aluminum electrolysis cell, and according to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with the added state residence time to predict the remaining life of the aluminum electrolysis cell includes:

[0019] Input the data to be measured into the HSMM model of each degradation state, and identify the current degradation state of the aluminum electrolytic cell by calculating the maximum likelihood probability;

[0020] According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with state residence time added to determine the probability distribution of the remaining residence time of the aluminum electrolytic cell in the current degradation state;

[0021] Determine 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.

[0022] Furthermore, the recurrence formula of the forward algorithm is:

[0023]

[0024] where α t+1 (i, d) represents the probability that the degradation state at time t + 1 is S i and the remaining residence time of this degradation state is d; p i (d) represents the probability that the duration in the degradation state s i is d; α t (i, d + 1) represents the probability that the degradation state at time t is S i and the remaining residence time of this degradation state is d + 1; represents the probability that the observed value is O i when the degradation state is S t ; S t (i) represents the probability that the remaining residence time τ t of the current degradation state is 1 and the next state is S t under the condition that the observed value is O i .

[0025] Furthermore, the 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 includes:

[0026] According to the probability α i (i, d) that the degradation state at time t + 1 is S t+1 and the remaining residence time of this degradation state is d determined by the forward algorithm, determine the remaining residence time D i (S t ) of the current degradation state S i at time t, which is expressed as:

[0027]

[0028] where D represents the maximum residence time;

[0029] According to the obtained D t (S i ), determine that the aluminum electrolysis cell is in the degradation state S i The remaining useful life RUL after staying at time t is:

[0030]

[0031] where N represents the number of degradation states, and D(S j ) represents the remaining residence time of the degradation state s j .

[0032] Furthermore, after training the HSMM model with all degradation states and the HSMM model of each degradation state by using the obtained historical data, the method includes:

[0033] Verify the trained HSMM model with all degradation states and the HSMM model of each degradation state.

[0034] The beneficial effects brought by the technical solution provided by the embodiment of the present invention at least include:

[0035] In the embodiment of the present invention, historical data capable of characterizing the degradation process of the aluminum electrolysis cell is obtained; based on the obtained historical data, the HSMM model is used to model the degradation process of the aluminum electrolysis cell. To improve the remaining life prediction effect of the HSMM, the probability distribution of the state residence time in the HSMM adopts the Erlang distribution to improve the prediction accuracy of the remaining life of the aluminum electrolysis cell; and an improved forward algorithm with the state residence time added is used during life prediction. In this way, not only the calculation of life prediction is simplified and the calculation efficiency is improved, but also the accuracy of the HSMM prediction result is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a schematic flow chart of the method for predicting the remaining life of the aluminum electrolysis cell provided by the embodiment of the present invention;

[0038] Figure 2 It is a schematic structural diagram of the aluminum electrolysis cell provided by the embodiment of the present invention;

[0039] Figure 3 It is a schematic structural diagram of the HSMM provided by the embodiment of the present invention;

[0040] Figure 4 Schematic diagram of the detailed process of the method for predicting the remaining life of an aluminum electrolysis cell provided by an embodiment of the present invention. Specific embodiments

[0041] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings.

[0042] Embodiment 1

[0043] As Figure 1 shown, an embodiment of the present invention provides a method for predicting the remaining life of an aluminum electrolysis cell, including:

[0044] S101, obtaining historical data that can characterize the degradation process of the aluminum electrolysis cell;

[0045] In this embodiment, the structural schematic diagram of the aluminum electrolysis cell is as Figure 2 shown. By analyzing the influence of the aluminum electrolysis production process on the aluminum electrolysis cell, and integrating various literatures and expert experiences, the degradation of the aluminum electrolysis cell can be summarized into the following main reasons:

[0046] (1) Sodium penetration: The main signals of the breakage of the bottom lining of the electrolysis cell are the formation of thick fluoride crystals and the breakage of carbon blocks. The formation of this breaking force is mainly due to the expansion of osmotic crystallization and the reaction of sodium and electrolyte when the electrolysis cell is just started.

[0047] (2) Air infiltration causes oxidation and breakage of the lining: Due to poor sealing, air enters the lining, and then a sodium-carbon-air reaction directly occurs under the cathode lining, continuously causing oxidation and corrosion of the lining.

[0048] (3) Electrolyte leakage: The bottom refractory bricks are eroded by the melt, and the electrolyte leakage causes the steel bars to melt. If the crack depth of the cathode carbon block reaches the top of the steel bar, high-temperature aluminum liquid or electrolyte liquid will continuously infiltrate, resulting in rapid melting and corrosion of the anode steel bar, and a leakage accident will occur soon.

[0049] (4) Bulging of the cathode carbon block: During baking and start-up, the cathode bulges, and the electrolyte infiltrates into the lining of the cathode carbon block, resulting in volume expansion of the cathode carbon block. The electrolyte seeps down along the carbon seams and cracks between the blocks until it seeps to the bottom of the cathode carbon block, intensifying the bulge and ultimately causing the breakage of the carbon block.

[0050] (5) Engineering construction and baking start-up control: The construction technology of the pre-baked cell furnace is unqualified, resulting in uneven current distribution after power-on, forming local overheating, and the cathode carbon block overheats and deforms to generate cracks. Or during the baking and start-up process of the pre-baked cell, the heating is uneven, resulting in premature generation of large-gap cracks and breakage of individual cathode carbon blocks.

[0051] In terms of the data trend, the characteristics of the parameters related to the degradation process of the electrolytic cell are as follows: the temperature of the cathode steel bar gradually increases, the operating voltage of the electrolytic cell gradually rises, the cathode voltage drop becomes larger, the electrolytic cell shows needle vibration phenomenon, the temperature of the cell shell gradually increases, the iron content and silicon content in the electrolyte increase, and the aluminum tapping volume decreases, etc.

[0052] It can be seen that by analyzing the degradation process of the aluminum electrolytic cell, the parameters that can characterize the degradation process of the aluminum electrolytic cell can be determined, and the historical data corresponding to the parameters can be obtained; among them, the parameters include: the operating voltage of the aluminum electrolytic cell, the iron content in the electrolyte, and the silicon content in the electrolyte.

[0053] In this embodiment, according to the expert experience and the actual statistical results, the failure range indexes of the key parameters for the degradation modeling of the above aluminum electrolytic cell are shown in Table 1.

[0054] Table 1 Failure range indexes of key parameters for aluminum electrolytic cell degradation modeling

[0055]

[0056] S102, using the obtained historical data to train a hidden semi-Markov model (HSMM) with all degradation states and the HSMM model of each degradation state;

[0057] In this embodiment, in order to predict the remaining life of the aluminum electrolytic cell, first, it is necessary to model the degradation process of the aluminum electrolytic cell. Specifically, an improved HSMM model is used to model the degradation process of the aluminum electrolytic cell, including: using the historical data of all degradation states of the aluminum electrolytic cell to train an HSMM model with all degradation states; using the historical data of each degradation state of the aluminum electrolytic cell to train the HSMM model of the corresponding degradation state; then, through a certain prediction algorithm, the remaining life of the aluminum electrolytic cell is predicted.

[0058] In this embodiment, in order to better understand the HSMM model, a brief description is given first:

[0059] Different from the fact that one state of the hidden Markov model corresponds to one observation value, in the HSMM model, one state can correspond to multiple observation values, and these multiple observation values can be regarded as an observation unit. Suppose there are N observation units, and q n is the time index of a single observation point in the nth (1≤n≤N) observation unit, and N represents the number of degradation states. Let the hidden state at time t be S t , O is the observation value sequence corresponding to this state. For the nth observation unit, these observation values are and they have the same microscopic state flag h n represents the nth degradation state, and its structure diagram is asFigure 3 as shown

[0060] The simplified hidden semi - Markov model can be described by four probability matrices, and its model can be expressed as:

[0061] λ=(π, A, B, P) (1)

[0062] Where π represents the initial state probability distribution vector of the aluminum electrolytic cell, π=(π1, π2, π3,..., π N ), π i represents the probability that the model is in the degradation state S t at time q i , that is:

[0063] π i =P(q t =S i ), 1≤i≤N,

[0064]

[0065] A represents the probability matrix of the mutual transfer between the various degradation states of the aluminum electrolytic cell (abbreviation: state transition matrix), A = {a ij}, 1≤i, j≤N. a ij represents the probability of transferring from the degradation state s i to the degradation state s j , that is:

[0066] a ij =P(q t =S j |q t-1 =S i ),

[0067]

[0068] B represents the probability matrix of the observed values of the aluminum electrolytic cell (abbreviation: observed value probability matrix), B = {b i (V m )}, 1≤i≤N, 1≤m≤M. b i (V m ) represents the probability that the observed value is V i when the degradation state is s m , that is:

[0069] b i (V m ) = P(O t =V m |q t =S i ), 1≤i≤N,

[0070]

[0071] Among them, M represents the number of observation values, and O t represents an observation value, and P(O t =V m |q t =S i ) represents the probability that the observation value is V under the condition that the model is in the degraded state s at the q t moment; i ; m ;

[0072] P represents the probability distribution matrix of the duration of the aluminum electrolysis cell state (abbreviation: state residence time probability distribution matrix):

[0073] P i (d)=P(d|q t =S i )

[0074] , 1 ≤ i ≤ N, 1 ≤ d ≤ D (5)

[0075] Among them, D represents the longest possible duration of any degraded state (abbreviation: maximum residence time), and P i (d) represents the probability that the duration in the degraded state s i is d (i.e., the state residence time d), and P(d|q t =i) represents the probability that the duration in this degraded state is d under the condition that the model is in the degraded state S at the q t moment. i ;

[0076] In the hidden semi-Markov model, usually the distribution of the state residence time generally selects the form of the exponential distribution, but from the actual effect, the effect is not ideal in some cases. In order to improve the prediction effect of HSMM, in this embodiment, based on queuing theory, the probability distribution of the state residence time in the traditional HSMM is changed from the exponential distribution form to the Erlang distribution to improve the accuracy of the remaining life prediction.

[0077] In probability statistics, the Erlang distribution is a continuous probability distribution, which, like the exponential distribution, can be used to represent the time interval between the occurrence of independent random events. The Erlang distribution is widely used in queuing theory. Compared with the exponential distribution, changing the parameters of the Erlang distribution can approximate almost all continuous distributions, so it can better fit the real data.

[0078] If k random variables X i , i = 1, 2,..., k, respectively follow the exponential distribution, then the random variable X = X1 + X2 +... + X kSubject to the Erlang distribution. That is, a random variable with a k - order Erlang distribution can be regarded as the sum of k independent random variables with the same exponential distribution.

[0079] The probability density f(x; k, μ) of the Erlang distribution is:

[0080]

[0081] where k represents the order number, which is a positive integer; μ represents 1 / λ, λ represents the exponential distribution parameter, μ > 0, and the mean and variance are kμ and kμ² respectively 2 ; x is the independent variable.

[0082] Define ξ(i, J, d) as the probability ξ that the observed values O1, O2,..., O are observed when the degradation state of the aluminum electrolytic cell at time t - 1 is Si and the degradation state at time t is Sj and the residence time is d i , t is Sj j and the residence time is d T of ξ t (i, j, d):

[0083] ξ t (i, j, d) = P(q t = Si i , q t+1 = Sj j |O, C i = d) (7)

[0084] where C i represents the duration of the degradation state Si i lasting;

[0085] Then, when performing parameter estimation on the data, the mean μ(i) and variance σ(i) can be calculated through the following two formulas respectively:

[0086]

[0087]

[0088] where T represents the total time;

[0089] Then the average residence time of the state is:

[0090]

[0091] Define the probability that the degradation state of the model at time t is Si i can be denoted as γ t (i), that is:

[0092]

[0093] Among them, ξ t (i, j) represents the probability that the state of the aluminum electrolysis cell at time t - 1 is S i , and the state at time t is S j ; according to the definition of the model parameters, other parameters of the model are updated using intermediate variables, and the update formulas are as follows:

[0094] π i = the expected probability that the model is in state S at the initial time t = 1 i = γ1(i) (12)

[0095]

[0096]

[0097] Among them, represents the probability that the observed value is V when in the degradation state S j ; the form δ(x, y) is a judgment function, which outputs 1 when x = y, and 0 otherwise. m The probability that the observed value is V when in the degradation state S

[0098] S103. Use the HSMM models of all single degradation states obtained by training to identify the current degradation state of the aluminum electrolysis cell. According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with state residence time added to predict the remaining life of the aluminum electrolysis cell, which can specifically include the following steps:

[0099] Input the data to be measured into the HSMM models of each degradation state, and identify the current degradation state of the aluminum electrolysis cell by calculating the maximum likelihood probability;

[0100] According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with state residence time added to determine the probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state;

[0101] According to the determined probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state, determine the remaining life of the aluminum electrolysis cell.

[0102] In this embodiment, when the hidden semi - Markov model is used for life prediction, three problems also need to be solved, namely, the identification problem, the decoding problem, and the learning problem. To solve the above three problems, relatively classical algorithms are usually used for solution, which are the Forward - backward algorithm, the Viterbi algorithm, and the Baum - Welch algorithm respectively. Among them, the derivation of the Forward - backward algorithm is the basis of the following two algorithms.

[0103] In this embodiment, the remaining useful life is predicted by using an improved forward algorithm. Among them, the improved forward variable is defined as:

[0104]

[0105] The forward variable α t (i, d) represents the probability that, under the condition of the known observation sequence O t , the degradation state at time t is S i and the remaining sojourn time τ of this degradation state t is d;

[0106] The recurrence formula of the forward algorithm is:

[0107]

[0108] where α t+1 (i, d) represents the probability that the degradation state at time t + 1 is S i and the remaining sojourn time of this degradation state is d; p i (d) represents the probability that the duration in the degradation state S i is d; α t (i, d + 1) represents the probability that the degradation state at time t is S i and the remaining sojourn time of this degradation state is d + 1; represents the probability that the observed value is O i when the degradation state is S t ; S t (i) represents the probability that, under the condition that the observed value is O t , the remaining sojourn time τ of the current degradation state t is 1 and the next state is S i ; S t (i) and are intermediate calculation values of the forward algorithm. Therefore, at any time t, based on the model parameters and the known observation sequence, the distribution probability of the remaining sojourn time of the aluminum electrolysis cell in the current state S t can be solved.

[0109] At time t, the maximum likelihood probability is calculated according to the Viterbi algorithm, and then the current degradation state q t = S i of the aluminum electrolysis cell can be estimated; then, according to the forward algorithm, the probability α i (i, d) that the degradation state at time t + 1 is S t+1 and the remaining sojourn time of this degradation state is d is determined, and the remaining sojourn time D t (S i ) of the current degradation state at time t is represented as:

[0110]

[0111] According to the obtained D t (S i ), determine the degradation state S of the aluminum electrolysis cell i The remaining useful life RUL after staying at time t is as follows:

[0112]

[0113] Among them, N represents the number of degradation states, and D(S j ) represents the remaining residence time of the degradation state S j .

[0114] As Figure 4 shown, the steps of the remaining life prediction method of the aluminum electrolysis cell based on HSMM are as follows:

[0115]

[0116]

[0117] In this embodiment, after training the HSMM model with all degradation states and the HSMM model of each degradation state by using the obtained historical data, the method includes:

[0118] Verify the trained HSMM model with all degradation states and the HSMM model of each degradation state.

[0119] In this embodiment, the actual data of a certain aluminum electrolysis plant is used for verification. Specifically:

[0120] Select the daily report data and chemical analysis data actually recorded in the production of the aluminum electrolysis workshop. Among these data, select two electrolytic cells from each of the two work areas that have undergone major repairs (No. 2605 and No. 2616 in Work Area 6 and No. 2720 and No. 2732 in Work Area 7). Because there are differences in the set voltage, working mode, etc. of different work areas, it is necessary to separately model and analyze the aluminum electrolysis cells in different work areas.

[0121] 3.1 Modeling of the degradation process of the aluminum electrolysis cell

[0122] For the daily report data and chemical analysis data, select three parameters, namely, the working voltage of the aluminum electrolysis cell, the iron content in the electrolyte, and the silicon content in the electrolyte, which can characterize the degradation process of the aluminum electrolysis cell, as the model training data, and perform preprocessing; among them, the data of No. 2720 and No. 2605 from the start of recording to 837 days and 771 days when the aluminum electrolysis cell is deactivated. Perform change point detection, segment clustering, and processing using the maximum-minimum redundancy algorithm on the preprocessed data, and finally determine that the aluminum electrolysis cell has experienced a total of 4 degradation states.

[0123] Before modeling, the initial parameters of the model need to be determined first. Under normal circumstances, since the aluminum electrolytic cell is in an undegraded state at the initial moment, the initial state probability distribution vector π = [1, 0, 0, 0]. Also, because the degradation of the aluminum electrolytic cell can only transfer from the current state to the next degraded state or remain in the current state, and will not transfer to the previous degraded state, that is, a left-right type HSMM model. Therefore, the initial state transition matrix can be specified in the following form:

[0124]

[0125] Generally, the determination of the observation value probability matrix B is rather tricky. In this embodiment, the K-means clustering method is used to estimate the initial value. The idea of this method is to cluster the data of one state divided into F classes using the K-means clustering algorithm, and then calculate the mean vector and covariance matrix of the data of these F classes respectively, so as to obtain the parameters of F normal distributions. Then, by dividing the number of data contained in each class by the total number of data in this state, the mixing coefficient of each class density function can be obtained. Finally, the probability density function of a certain state can be obtained by linearly adding these F normal distribution functions. The state residence time probability distribution matrix P adopts the Erlang distribution and is estimated using equations (6) to (10).

[0126] The data of cell No. 2720 and cell No. 2605 are respectively input into the HSMM for training, and the output parameters of the model are obtained respectively, so as to obtain the degradation model of the complete data cycle of the electrolytic cell. The state transition paths of the two cells go through a total of 4 states, namely degradation state 2, degradation state 3, and degradation state 4, from the initial stable operation stage (referred to as: degradation state 1), and reach the end of the life when the aluminum electrolytic cell stops operating at the last data point. At the same time, the HSMM models corresponding to each degradation state are trained with the data of the four degradation states of each electrolytic cell respectively, so as to facilitate subsequent determination of the degradation state of the verification data / test data by inputting the verification data / test data into each degradation model. Furthermore, the remaining life of the aluminum electrolytic cell can be predicted by using the forward algorithm with the state residence time through the degradation state and the state residence time.

[0127] 3.2 Prediction of the remaining life of the aluminum electrolytic cell

[0128] Select the data of slot No. 2732 and slot No. 2616 as the validation set, input the validation data into the trained HSMM model, and judge the degradation state of the aluminum electrolysis cell by comparing the maximum likelihood probability value. Subsequently, according to the current degradation state and the state residence time, the remaining life of the aluminum electrolysis cell in the current state is obtained through equations (17) and (18). Through verification, it is obtained that the prediction accuracy of the aluminum electrolysis cell remaining life prediction method provided by the embodiment of the present invention is more accurate than the existing aluminum electrolysis cell life prediction results.

[0129] For the aluminum electrolysis cell remaining life prediction method described in the embodiment of the present invention, historical data capable of characterizing the degradation process of the aluminum electrolysis cell is obtained; based on the obtained historical data, an HSMM model is used to model the degradation process of the aluminum electrolysis cell. To improve the remaining life prediction effect of HSMM, the probability distribution of the state residence time in HSMM adopts the Erlang distribution to improve the prediction accuracy of the remaining life of the aluminum electrolysis cell; and an improved forward algorithm with the state residence time added is used during life prediction. In this way, not only the calculation of life prediction is simplified and the calculation efficiency is improved, but also the accuracy of the HSMM prediction result is further improved.

[0130] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for predicting the remaining life of an aluminum electrolysis cell, characterized in that, Including: Obtaining historical data that can characterize the degradation process of an aluminum electrolysis cell; Using the obtained historical data to train an HSMM model with all degradation states and an HSMM model for each degradation state; where HSMM represents a hidden semi-Markov model, and the probability distribution of the state residence time in the HSMM adopts an Erlang distribution; Using the trained HSMM models of all single degradation states to identify the current degradation state of the aluminum electrolysis cell. According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with the state residence time added to predict the remaining life of the aluminum electrolysis cell; Wherein, before using the obtained historical data to train an HSMM model with all degradation states and an HSMM model for each degradation state, the method includes: Based on queuing theory, changing the probability distribution of the state residence time in the traditional HSMM from an exponential distribution form to an Erlang distribution.

2. The method for predicting the remaining life of an aluminum electrolytic cell according to claim 1, wherein The obtaining of historical data that can characterize the degradation process of the aluminum electrolysis cell includes: Analyzing the degradation process of the aluminum electrolysis cell, determining the parameters that can characterize the degradation process of the aluminum electrolysis cell, and obtaining the historical data corresponding to the parameters; wherein, the parameters include: the operating voltage of the aluminum electrolysis cell, the iron content in the electrolyte, and the silicon content in the electrolyte.

3. The method for predicting the remaining life of an aluminum electrolytic cell according to claim 1, characterized in that, The obtained historical data that can characterize the degradation process of the aluminum electrolysis cell includes: historical data of all degradation states of the aluminum electrolysis cell; The using of the obtained historical data to train an HSMM model with all degradation states and an HSMM model for each degradation state includes: Using the historical data of all degradation states of the aluminum electrolysis cell to train an HSMM model with all degradation states; Using the historical data of each degradation state of the aluminum electrolysis cell to train the HSMM model of the corresponding degradation state.

4. The method for predicting the remaining life of an aluminum electrolytic cell according to claim 1, wherein, The using of the trained HSMM models of all single degradation states to identify the current degradation state of the aluminum electrolysis cell. According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with the state residence time added to predict the remaining life of the aluminum electrolysis cell includes: Inputting the data to be measured into the HSMM models of each degradation state, and identifying the current degradation state of the aluminum electrolysis cell by calculating the maximum likelihood probability; According to the degradation state identification result, the HSMM model with all degradation states uses the forward algorithm with the state residence time added to determine the probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state; According to the determined probability distribution of the remaining residence time of the aluminum electrolysis cell in the current degradation state, determining the remaining life of the aluminum electrolysis cell.

5. The method for predicting the remaining life of an aluminum electrolytic cell according to claim 4, wherein, The recurrence formula of the forward algorithm is: where, α t+1 (i, d) represents the probability that the degradation state at time t + 1 is S i and the remaining sojourn time of this degradation state is d; p i (d) represents the probability that the duration in the degradation state S i is d; α t (i, d + 1) represents the probability that the degradation state at time t is S i and the remaining sojourn time of this degradation state is d + 1; represents the probability that the observed value is O i when the degradation state is S t ; S t (i) represents the probability that the remaining sojourn time τ t of the current degradation state is 1 and the next state is S t under the condition that the observed value is O i .

6. The method for predicting the remaining life of an aluminum electrolytic cell according to claim 4, wherein The determining of 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 includes: The degradation state at time t+1 determined according to the forward algorithm is S i and the probability α that the remaining sojourn time of this degradation state is d t+1 (i, d), to determine the current degradation state S at time t i of the remaining sojourn time D t (S i ), is expressed as: Wherein, D represents the maximum residence time; Based on the obtained D t (S i ), determine that the aluminum electrolysis cell is in the degradation state S i The remaining useful life RUL after dwelling at time t is: where N represents the number of degradation states, and D(S j ) represents the remaining residence time of the degradation state S j .

7. The method for predicting the remaining life of an aluminum electrolytic cell according to claim 1, characterized in that, After using the obtained historical data to train an HSMM model with all degradation states and an HSMM model for each degradation state, the method includes: Verifying the trained HSMM model with all degradation states and the HSMM models for each degradation state.

Images