CN115270601A - FDALM-based method and system for monitoring preparation process of ternary cathode material - Google Patents
FDALM-based method and system for monitoring preparation process of ternary cathode material Download PDFInfo
- Publication number
- CN115270601A CN115270601A CN202210720684.7A CN202210720684A CN115270601A CN 115270601 A CN115270601 A CN 115270601A CN 202210720684 A CN202210720684 A CN 202210720684A CN 115270601 A CN115270601 A CN 115270601A
- Authority
- CN
- China
- Prior art keywords
- probability
- factor
- model
- representing
- hidden variable
- 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
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
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N5/00—Computing arrangements using knowledge-based models
- G06N5/04—Inference or reasoning models
-
- 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
- G06F2119/18—Manufacturability analysis or optimisation for manufacturability
-
- 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
- Y02E—REDUCTION OF GREENHOUSE GAS [GHG] EMISSIONS, RELATED TO ENERGY GENERATION, TRANSMISSION OR DISTRIBUTION
- Y02E60/00—Enabling technologies; Technologies with a potential or indirect contribution to GHG emissions mitigation
- Y02E60/10—Energy storage using batteries
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Physiology (AREA)
- Genetics & Genomics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Molecular Biology (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a method and a system for monitoring a preparation process of a ternary cathode material based on FDALM (fully drawn aluminum standard), which are applied to monitoring a sintering process of the ternary cathode material, wherein the sintering process of the ternary cathode material is a typical process industry, and the production process relates to a plurality of chemical reactions coupled with each other, including chemical combination, hydrolysis and side reactions. Firstly, deducing a factor FDALM modeling method by means of a factor modeling method on the basis of the technology of a dynamic autoregressive hidden variable model, wherein the factor FDALM modeling method models data with dynamic and multi-modal characteristics at the same time, and an improved EM algorithm is utilized to learn model parameters; then, in order to give full play to the process output of each factor model, the statistical values of the sub models are fused into the posterior fault probability of the sample by means of a Bayesian inference technology; finally, simulation results compared with other models show that the proposed monitoring method can track modal fluctuations of the process.
Description
Technical Field
The invention relates to the technical field of ternary cathode material preparation process fault detection, and particularly discloses a Factor Dynamic autoregressive hidden Variable Model (FDALM) -based ternary cathode material preparation process monitoring method and system.
Background
The preparation process of the ternary cathode material is a complex process in which substance transmission, energy exchange, heat convection, heat diffusion and other forms of mutual coupling occur simultaneously in a roller kiln. In order to simplify the system, the sintering process is divided into three temperature sections according to the development trend of temperature, namely a temperature rising section, a constant temperature section and a temperature lowering section, and corresponding oxygen and temperature are added into each temperature section according to the generated chemical reaction. Under the condition of specific temperature and oxygen, the materials can generate established reaction in the corresponding temperature zone. However, if the sintering system deviates from the set process, the product quality does not reach the standard or energy is wasted. Therefore, real-time monitoring is established for the sintering process to adjust the sintering system, and operators can be guided to adjust the operating parameters in real time.
However, chemical reactions occurring inside the roller kiln are related, so that data generated at the current moment are influenced by historical moment data, namely the data show high-order dynamic characteristics; in addition, the roller kiln runs continuously twenty-four hours a day, chemical reaction is particularly sensitive to temperature and raw material mixing ratio, and the process presents a plurality of stable working conditions due to the fact that the external environment temperature periodically changes, so that process data presents certain multi-modal characteristics. Therefore, a method for monitoring the preparation process of the ternary cathode material, which considers the high-order dynamic characteristics and the multi-modal characteristics of the process at the same time, is needed to ensure the stable operation of the preparation system of the ternary cathode material.
Therefore, the high-order dynamic characteristics and the multi-modal characteristics presented in the preparation process of the ternary cathode material cause low fault detection rate and high false alarm rate of the traditional monitoring method, and the method is a technical problem to be solved urgently at present.
Disclosure of Invention
The invention provides a method and a system for monitoring a preparation process of a ternary cathode material based on FDALM (fully drawn aluminum model), and aims to solve the technical problems of low fault detection rate and high false alarm rate of the traditional monitoring method due to high-order dynamic characteristics and multi-modal characteristics presented in the preparation process of the ternary cathode material.
The invention relates to a method for monitoring the preparation process of a ternary cathode material based on FDALM, which comprises the following steps:
data preparation and preprocessing, collecting historical data xh(k) K =1, 2., T as a model training set, each sample was normalized; wherein, the first and the second end of the pipe are connected with each other,the m-dimensional process observation data is obtained, k is a time label, and T is the number of samples;
respectively identifying a time-lag coefficient L and a system factor K of the model by using a time-lag identification and clustering algorithm, establishing a factor dynamic autoregressive hidden variable model, and initializing a model parameter thetaold;
Identifying relevant parameters of a factor dynamic autoregressive hidden variable model by adopting an improved EM algorithm;
defining the factor dynamic autoregressive hidden variable model under each sub-model state based on the obtained factor dynamic autoregressive hidden variable modelStatistics ofMeasuring and determining control thresholds for submodelsAnd a significance level α;
collecting online data x (k), k =1, 2., N as a test set of the model, and performing a standardization process;
detecting the test set based on the obtained factor dynamic autoregressive model, and calculating the test sample in each sub-modeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withAnd comparing with the significance level alpha, and outputting a detection result.
Further, a time lag coefficient L and a system factor K of the model are respectively identified by using a time lag identification algorithm and a clustering algorithm, a factor dynamic autoregressive hidden variable model is established, and a model parameter theta is initializedoldComprises the following steps:
identifying to obtain a time lag coefficient L by a trend similarity algorithm;
the physical interpretation of the system factor K is the number of the types of data division, and the system factor K is obtained by identification through an affine clustering propagation algorithm based on a genetic algorithm;
and constructing a factor dynamic autoregressive hidden variable model by utilizing the preprocessed data set.
Further, the step of identifying relevant parameters of the factor dynamic autoregressive hidden variable model by using the improved EM algorithm comprises the following steps:
in the step E, reasonably estimating the posterior distribution of the expanded dynamic hidden variable expectation and the system factor K based on Bayesian filtering and smoothing;
in the step M, model parameters are updated by means of a method of maximizing a likelihood function, a Lagrange multiplier formula is constructed, and factor coefficients are constrained and updated by using factors in the step M.
Further, based on the obtained factor dynamic autoregressive hidden variable model, defining the factor dynamic autoregressive hidden variable model under each sub-modeStatistics and determining control thresholds for submodelsAnd in the step of the significance level alpha, for the trained factor dynamic autoregressive hidden variable model, the hidden variable is a key variable for driving the operation of the dynamic change process, and the hidden variable of each sub-model can establish a corresponding hidden variableThe statistical quantity is calculated by the statistical quantity,the statistic is calculated by the following formula:
wherein the content of the first and second substances,t representing the kth sub-modality2The statistical quantity is calculated by the statistical quantity,representing an implicit variable zt qWith respect to x1:t q,y1:t qCovariance, which represents the hidden variable zt qWith respect to x1:t q,y1:t qCovariance of E (z)t q|x1:t q,yl:t q) Representing hidden variables zt qWith respect to x1:t q,y1:t qThe condition of (2) is desired;
in order to fully utilize the key information of each mode, the monitoring results of each sub-mode are fused into fault probability by means of a Bayesian inference method, and the event probability that an observation sample fails in the kth mode is constructed as follows:
wherein the content of the first and second substances,represents a prior probability of failure of the process data,represents the prior probability that the process data is normal,representing the conditional probability of the observation sample under the k modal fault condition;representing the probability of the observed variable being observed in the k-th modality;representing the conditional probability of the observation sample under the normal condition of the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition in the kth mode;representing a process data failure prior probability;representing the normal prior probability of the process data;
will be provided withAndin combination with the significance level α, the following formula is defined:
wherein α represents a significance level; actual size is a balance between false positives and false negatives, failure probability for obtaining new data samplesConstructing the conditional probability of observing the occurrence of the sample under normal and fault conditions, and defining the following formula:
wherein, the first and the second end of the pipe are connected with each other,indicating the occurrence of the observed sample under normal conditions in the kth modeConditional probability of (d);representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;is a control limit for each modality, the value of which is uniquely determined by the degree of freedom d and the significance level α of the submodel;t representing the kth sub-modality2And (4) statistic amount.
Further, detecting the test set based on the obtained factor dynamic autoregressive model, and calculating the test sample under each sub-modeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withComparing with the significance level alpha, and outputting the detection result, after the probability of the fault of each local model of the observation data is determined, fusing the probability of the fault of each sub-model by using Bayesian inference technology to obtain the fault probability of the sampleProbability of failure of a sampleCalculated by the following formula:
wherein the content of the first and second substances,denotes the failure probability of the sample, P (k | x)t) Factor k with respect to xtA posterior probability of (d);indicating the relation to a new sample x under a factor mode ktThe posterior probability of failure of (1);
whether the system fails or not is judged by comparing the failure probability with the significance level alpha, and the judgment logic is shown as the following formula:
wherein the content of the first and second substances,indicating the probability of failure of the sample and alpha indicating the level of significance.
Another aspect of the invention relates to a system for monitoring the preparation process of a ternary cathode material based on FDALM, comprising:
a preprocessing module for preparing and preprocessing data and collecting historical data xh(k) K =1, 2., T as a model training set, each sample was normalized; wherein, the first and the second end of the pipe are connected with each other,the data are observed in an m-dimensional process, k is a time label, and T is the number of samples;
the establishing module is used for respectively identifying the time lag coefficient L and the system factor K of the model by using time lag identification and clustering algorithms, establishing a factor dynamic autoregressive hidden variable model and initializing model parameters thetaold;
The identification module is used for identifying relevant parameters of the factor dynamic autoregressive hidden variable model by adopting an improved EM algorithm;
a determining module for defining each sub-mode of the factor dynamic autoregressive hidden variable model based on the obtained factor dynamic autoregressive hidden variable modelStatistics and determining control thresholds for submodelsAnd a significance level α;
the collecting module is used for collecting online data x (k), wherein k =1, 2., N is used as a test set of the model and carrying out standardization processing;
a calculation module for detecting the test set based on the obtained factor dynamic autoregressive model and calculating the test sample in each sub-modeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withAnd comparing with the significance level alpha, and outputting a detection result.
Further, the establishing module comprises:
the first acquisition unit is used for identifying and obtaining a time lag coefficient L through a trend similarity algorithm;
the second obtaining unit is used for obtaining a system factor K through affine clustering propagation algorithm identification based on a genetic algorithm, and the system factor K is physically interpreted as the number of types of data division;
and the first construction unit is used for constructing a factor dynamic autoregressive hidden variable model by utilizing the preprocessed data set.
Further, the identification module comprises:
the estimation unit is used for reasonably estimating the posterior distribution of the expansion dynamic hidden variable expectation and the system factor K based on Bayesian filtering and smoothing in the step E;
and the second construction unit is used for updating the model parameters by means of a method of maximizing a likelihood function in the step M, constructing a Lagrange multiplier formula and utilizing the factors in the step M to restrain and update the factor coefficients.
Further, in the determination module, for the trained factor dynamic autoregressive hidden variable model, hidden variables are key variables for driving the operation of the dynamic change process, and the hidden variables of each sub-mode can establish corresponding hidden variablesThe statistical quantity is calculated by the statistical quantity,the statistic is calculated by the following formula:
wherein the content of the first and second substances,t representing the kth sub-modality2Statistic, E (z)t q|x1:t q,y1:t q)TRepresenting hidden variables zt qWith respect to x1:t q,y1:t qCovariance represents the hidden variable zt qWith respect to x1:t q,y1:t qCovariance of E (z)t q|x1:t q,yl:t q) Representing an implicit variable zt qAbout x1:t q,y1:t qThe condition of (2) is desired;
in order to fully utilize the key information of each mode, the monitoring results of each sub-mode are fused into fault probability by means of a Bayesian inference method, and the event probability that an observation sample fails in the kth mode is constructed as follows:
wherein the content of the first and second substances,represents a prior probability of a failure of the process data,represents a prior probability that the process data is normal,representing the conditional probability of the observation sample under the k modal fault condition;representing the probability of the observed variable being observed in the kth modality;representing the conditional probability of the observation sample under the normal condition of the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition in the kth mode;representing a process data failure prior probability;representing the normal prior probability of the process data;
will be provided withAndin combination with the significance level α, the following formula is defined:
wherein α represents a significance level; actual size is a balance between false positives and false negatives, failure probability for obtaining new data samplesConstructing the conditional probability of observing the occurrence of the sample under normal and fault conditions, and defining the following formula:
wherein, the first and the second end of the pipe are connected with each other,representing the conditional probability of observing the sample under normal conditions in the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;is a control limit for each modality, the value of which is uniquely determined by the degree of freedom d and the significance level α of the submodel;t representing the kth sub-modality2Statistics are obtained.
Furthermore, in the calculation module, after the probability of the fault of each local model of the observation data is determined, the probability of the fault of each sub-model is further fused by using a Bayesian inference technology to obtain the fault probability of the sampleProbability of failure of a sampleCalculated by the following equation:
wherein the content of the first and second substances,denotes the failure probability of the sample, P (k | x)t) Factor k with respect to xtThe posterior probability of (d);indicating the relation to a new sample x under a factor modality ktThe posterior probability of failure of (1);
whether the system fails or not is judged by comparing the failure probability with the significance level alpha, and the judgment logic is shown as the following formula:
wherein the content of the first and second substances,indicating the probability of failure of the sample and alpha indicating the level of significance.
The beneficial effects obtained by the invention are as follows:
the invention provides a method and a system for monitoring the preparation process of a ternary cathode material based on FDALM (fully drawn alkaline metal-insulator-metal), which adopt data preparation and pretreatment to collect historical data xh(k) K =1, 2., T as a model training set, each sample was normalized; respectively identifying a time lag coefficient L and a system factor K of the model by using a time lag identification and clustering algorithm, establishing a factor dynamic autoregressive hidden variable model, and initializing a model parameter thetaold(ii) a Identifying relevant parameters of the dynamic autoregressive hidden variable model of the factor by adopting an improved EM (effective electromagnetic) algorithm; defining the factor dynamic autoregressive hidden variable model under each sub-model state based on the obtained factor dynamic autoregressive hidden variable modelStatistics and determining control thresholds for submodelsAnd a significance level α; collecting online data x (k), k =1, 2., N as a test set of models and performing a normalization process; detecting the test set based on the obtained factor dynamic autoregressive model, and calculating the test sample under each submodeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withAnd comparing with the significance level alpha, and outputting a detection result.
The FDALM-based method and system for monitoring the preparation process of the ternary cathode material are applied to monitoring the sintering process of the ternary cathode material, the sintering process of the ternary cathode material is a typical process industry, and the production process relates to a plurality of chemical reactions coupled with each other, including chemical combination, hydrolysis and side reactions. These reactions can lead to side reactions if the temperature is not well controlled, which are reversible, and thus, fluctuating changes in the external environment and materials can lead to the process being in a different steady state process. Leading to certain multimodal properties of the process. Meanwhile, the coupling of different temperature zones enables the data to be related before and after, namely the dynamic property of the process data is not negligible, so that the process data has complex characteristics. Firstly, deducing a factor FDALM modeling method by means of a factor modeling method on the basis of the technology of a dynamic autoregressive hidden variable model, wherein the factor FDALM modeling method models data with dynamic and multi-modal characteristics at the same time, and learns model parameters by utilizing an improved EM algorithm; then, in order to give full play to the process output of each factor model, the statistical values of the sub models are fused into the posterior fault probability of the sample by means of a Bayesian inference technology; finally, simulation results compared with other models show that the monitoring method can track modal fluctuations of the process.
Drawings
Fig. 1 is a schematic flow chart of an embodiment of a method for monitoring a preparation process of a ternary cathode material based on FDALM according to the present invention;
FIG. 2 is a schematic flow diagram of an embodiment of process monitoring incorporating an FDALM model and Bayesian inference in the FDALM-based ternary positive electrode material preparation process monitoring method provided by the present invention;
FIG. 3 is a process data classification diagram of an embodiment of the FDALM-based method for monitoring the manufacturing process of a ternary positive electrode material provided in the present invention;
FIG. 4 is a schematic diagram of monitoring results of abnormal temperature rise faults of a temperature zone by different methods in the method for monitoring the preparation process of the FDALM-based ternary cathode material provided by the invention;
FIG. 5 is a schematic diagram of monitoring results of temperature abnormal drop faults of a temperature zone by different methods in the FDALM-based ternary cathode material preparation process monitoring method provided by the invention;
FIG. 6 is a schematic diagram of the monitoring results of the shutdown faults of the roller kiln according to different methods in the method for monitoring the preparation process of the FDALM-based ternary cathode material provided by the invention;
FIG. 7 is a functional block diagram of an embodiment of a FDALM-based ternary positive electrode material preparation process monitoring system provided in the present disclosure;
FIG. 8 is a functional block diagram of one embodiment of the setup module shown in FIG. 7;
FIG. 9 is a functional block diagram of the identification module shown in FIG. 7 according to an embodiment.
The reference numbers indicate:
10. a pre-processing module; 20. establishing a module; 30. an identification module; 40. a determining module; 50. A collection module; 60. a calculation module; 21. a first acquisition unit; 22. a second acquisition unit; 23. a first building element; 31. an estimation unit; 32. a second building element.
Detailed Description
In order to better understand the technical scheme, the technical scheme is described in detail in the following with reference to the attached drawings of the specification and specific embodiments.
As shown in fig. 1 to 6, a first embodiment of the present invention provides a method for monitoring a preparation process of a ternary cathode material based on FDALM, comprising the following steps:
step S100, data preparation and preprocessing, and historical data x collectedh(k) K =1, 2., T as a model training set, each sample was normalized; wherein, the first and the second end of the pipe are connected with each other,and k is the time label and T is the sample number of the m-dimensional process observation data.
The data set used for model training and testing was collected from a production plant of some typical ternary cathode material. The sintering link of the ternary material is crucial to the performance of the battery material, so that the sintering process of the material needs to be accurately monitored, and the aims of ensuring the product quality and reducing the energy consumption are fulfilled. A total of 7 observation variables are selected, the first 6 variables are important temperature measurement data of the temperature rising region, and the last data is observation data representing the quality of the product and residual lithium. A total of 2200 consecutive time data are selected, and the sequence data factory always produces the same batch of battery materials, and the sampling period is 30min. The data set is preprocessed or standardized, and the standardization operation is as follows: the average value of the sample of the variable to which each element in the sample set belongs is subtracted, and then the average value is divided by the standard deviation of the sample, so that the average value of the data corresponding to each process variable and each key quality variable is 0, and the variance is 1.
Step S200, respectively identifying a time lag coefficient L and a system factor K of the model by using a time lag identification and clustering algorithm, establishing a factor dynamic autoregressive hidden variable model, and initializing a model parameter thetaold。
The step S200 includes:
and step S210, identifying through a trend similarity algorithm to obtain a time lag coefficient L.
And S220, physically interpreting the system factor K into the number of the types of data division, and identifying the types of the data division through an affine clustering propagation algorithm based on a genetic algorithm.
And step S230, constructing a factor dynamic autoregressive hidden variable model by utilizing the preprocessed data set.
The time lag coefficient L is obtained by identifying through a trend similarity algorithm; the physical interpretation of the factor K is the number of the types of data division, and the factor K is obtained by affine clustering propagation algorithm identification based on a genetic algorithm.
Constructing a factor dynamic autoregressive hidden variable model by utilizing the preprocessed data set, and assuming that the data set meets the following relational expression:
in the formula (1), zt,k∈RdRepresenting an implicit variable describing the state of the system at the current time t under the mode k, ht,kH, an augmented hidden variable representing the hidden state composition of the states at time t and L previous times in the mode kt,k=[zt, k Tzt-1,k T…zt-L+1,k T]T∈RdLAnd L is a time lag coefficient identified by a time lag identification algorithm. x is the number oft,k∈RvAnd d and v respectively represent dimensions of a hidden variable and an observed variable. A. Thek∈Rd×dLRepresenting a hidden variable h from augmentationt-1,kShifting to hidden variable z at presentt,kThe state transition matrix of (2). B isk∈Rv×dRepresenting an implicit variable zt,kDivergence to the observed variable xt,kThe divergence matrix of (1).Respectively, the hidden variable of the mode k and the Gaussian noise of the observed variable. Assuming that the noises are independent of each other, the distributions obeyed are respectivelyZero mean gaussian distribution.
The model parameters Θ of the whole dynamic autoregressive hidden variable model are as follows:
Θ={Ak,Bk,μπ,k,∑π,k,∑z,k,∑x,k,P(k)|k=1,2,…,K} (2)
in the formula (2), μπ,kSum sigmaπ,kRespectively for increasing hidden variable h at initial moment0,kMean and variance of gaussian distribution; sigmaz,k,∑x,kThe variance of Gaussian noise of an implicit variable and an observed variable respectively expressed as a mode k; p (k) is a factor coefficient in different modes.
And step S300, identifying relevant parameters of the factor dynamic autoregressive hidden variable model by adopting an improved EM algorithm.
Step S300 includes:
and S310, reasonably estimating the posterior distribution of the expanded dynamic hidden variable expectation and the system factor K based on Bayesian filtering and smoothing in the step E.
And S320, in the step M, updating the model parameters by means of a method of maximizing a likelihood function, constructing a Lagrange multiplier formula, and updating factor coefficients by using factor constraint in the step M.
The FDALM structure not only has unobservable hidden variables, but also needs to identify a system factor K and a corresponding factor coefficient. The traditional EM algorithm cannot meet the requirements, so by solving the model parameters with an improved EM algorithm: and E, reasonably estimating the posterior distribution of the expanded dynamic hidden variable expectation and the factor K based on Bayesian filtering and smoothing. In the step M, model parameters are updated by means of a method of maximizing a likelihood function, in addition, a Lagrange multiplier formula is constructed, and the factor number is updated by utilizing factor constraint in the step M.
In step E, model parameters { A }k,Bk,μπ,k,∑π,k,∑z,k,∑x,kP (K) | K =1,2, \8230;, K } is randomly initialized, where Ak=[A1,k,A2,k,…,AL,k]. Expanding dynamic hidden variable expectation E by adopting Bayesian filtering and smoothingz(zt,k)、And a posterior distribution P (k | x) of factor kt) An accurate estimation is performed, which is mainly disclosed as follows:
in the formula (3), Ez(zt,k) Indicating that the sample information of the 1: T time is utilized to T timeEstimating the expectation of the hidden variable of the moment; x is a radical of a fluorine atom1:TRepresents 1: all samples at time T; thetaoldRepresenting a previous iteration parameter set;respectively representing the k-th sub-modal hidden variable ht,kWith respect to observation sequence x1:TThe mean and variance of the posterior probability distribution of (a);represents the k-th sub-modal hidden variable ht,kWith respect to observation sequence x1:TTransposing the mean of the posterior probability distribution of (1);representing a k-th sub-modal lag i time hidden variable ht,kWith respect to observation sequence x1:TTransposing the mean of the posterior probability distribution of (1); a. Thei,kA state transition matrix representing hidden variables under the kth sub-modality at the lag i moment;representing the covariance of the ith variable at time t-1 and the lag of the variable by time L;representing the k-th sub-modal lag at time t-1, i, and the hidden variable ht,kWith respect to observation sequence x1:TThe mean of the posterior probability distribution of (1);represents the k-th sub-modal lag L time hidden variable h at the time t-1t,kWith respect to observation sequence x1:TTransposing the mean of the posterior probability distribution;represents the utilization of 1: estimating the covariance of the hidden variables at the time T by using the sample information at the time T;indicates that with 1: estimating the covariance of hidden variables at the T moment and the T-i moment by the T moment sample information;indicates that with 1: estimating covariance of hidden variables at T moment and T-L moment by sample information at T moment; m is a unit oft,kAnd Mt,kRespectively, the kth sub-modal hidden variable ht,kWith respect to observation sequence x1:TThe mean and variance of the posterior probability distribution of (a).
Another type of hidden variable is the posterior probability P (k | x) for factor kt) Wherein T =0,1, \8230;, T, the mathematical solution is shown below:
in the formula (4), P (k | x)t) Represents the posterior probability of factor k; p (x)t| k) represents the conditional probability of the observed variable in the kth modality; p (k) represents a factor coefficient; p (x)t) Representing an observed variable xtThe probability of occurrence.
In the step M, model parameters { A ] are carried out by utilizing the hidden variable expectation obtained in the step E and the posterior probability of the factor kk,Bk,μπ,k,∑π,k,∑z,k,∑x,kUpdate of P (K) | K =1,2, \8230;, K }:
in the formulas (5) to (10),respectively representing the updated augmentation hidden variables h in the kth mode0Mean and variance of;representing a state transition matrix, an observation matrix, an updating value of hidden variable variance and a corresponding transpose under the kth mode;representing covariance estimation of an augmentation hidden variable at the t-1 moment in the kth mode;representing covariance estimation of a t-moment hidden variable and a t-1 moment amplification hidden variable under a kth mode;representing a covariance estimate of an initial augmented hidden variable in a kth mode;representing covariance estimation of an implicit variable at lag i time in a kth mode; x is the number oftRepresenting an observed variable;Ez(h0,k) Respectively representing the expectation of the hidden variable at the t moment and the initial augmentation hidden variable in the kth mode.
In order to update the factor coefficient P (k), since the factor involves more constraints, an optimization condition needs to be constructed separately for the factor. First, all terms related to P (k) are separated and expressed as follows:
in formula (11), g (k) represents a log-likelihood function with respect to the factor coefficient P (k); p (k | x)t) Represents the posterior probability of factor k; p (k) represents a factor coefficient.
Due to the presence of constraintsIntroducing a Lagrange multiplier lambda and constructing a Lagrange function form as follows:
in equations (12) to (13), f (k) represents a lagrangian function constructed for updating the factor coefficient P (k); p (k | x)t) Represents the posterior probability of factor k; g (k) represents a log-likelihood function with respect to the factor coefficient P (k), and λ represents a Lagrangian multiplier.
And the two above two equations are accumulated for k terms (for k), resulting in:
in equation (14), P (k) represents a factor coefficient, and λ represents a lagrange multiplier.
Substituting the result back to equation 13 yields the result of the factor coefficient, as shown in the following equation:
in formula (15), P (k) represents a factor coefficient, P (k | x)t) Representing the posterior probability of factor k.
In the process of constructing the model, the maximum likelihood value obtained by calculating the new model parameter is compared with the corresponding maximum likelihood value of the original model parameter, if the error threshold value is met, the step S400 is entered, otherwise, the model parameter is continuously updated in an iterative manner. For example, after each M step, the new resulting model parameters ΘnewAnd old model parameter thetaoldComparing, if | | Θ is satisfiedold-ΘnewIf the | | is less than or equal to the sigma, the model training is finished and the step S400 is entered, otherwise, the model parameters are continuously updated according to the EM algorithm strategy of the step S300. Wherein σ is a threshold value of model convergence, and a model complete log-likelihood function is as follows:
in formula (16), Q (Θ | Θ)old) Log-likelihood function InP (x) representing equivalent conversion of log-likelihood function to computational complete data1:T,z1-L:T| Θ) about a hidden variable z1-L:TThe condition of (2) is desired; in { P (k) P (x)1:T,z1-L:TI k) is to be In (x)1:T,z1-L:T,k|Θold) Displaying according to conditional probability; inP (k) is the log probability of modal factors; cons represents a constant;Bkzt,kand others are part of a gaussian distribution after logarithmic expansion.
Step S400, based on the obtained factor dynamicsThe autoregressive hidden variable model is defined in each sub-model state of the factor dynamic autoregressive hidden variable modelStatistics and determining control thresholds for submodelsAnd significance level a.
It is necessary to define the model in each sub-modalityStatistics and determining control thresholds for submodelsAnd significance level a.
For the trained factor dynamic autoregressive hidden variable model, the hidden variables are key variables for driving the operation of the dynamic variation process, and the hidden variables of each sub-mode can establish correspondingThe statistical quantity is calculated according to the data,the statistic is calculated by the following formula:
in the case of the formula (17),t representing the kth sub-modality2Statistic, E (z)t q|x1:t q,y1:t q)TRepresenting an implicit variable zt qAbout x1:t q,y1:t qCovariance represents the hidden variable zt qAbout x1:t q,y1:t qCovariance of E (z)t q|xl:t q,yl:t q) Representing hidden variables zt qWith respect to x1:t q,y1:t qThe conditions of (a) are expected.
In order to fully utilize the key information of each mode, the monitoring results of each sub-mode are fused into fault probability by means of a Bayesian inference method, and the event probability that an observation sample fails in the kth mode is constructed as follows:
in the formula (18), the first and second groups,represents a prior probability of failure of the process data,represents a prior probability that the process data is normal,representing the conditional probability of the occurrence of the observation sample under the k-th modal fault condition;representing the probability of the observed variable being observed in the kth modality;representing the condition probability of the observation sample under the normal condition of the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition in the kth mode;representing a process data failure prior probability;representing the normal prior probability of the process data.
Will be provided withAndin combination with the significance level α, the following formula is defined:
in formula (19), α represents the significance level; the actual size is a balance between false positives and false negatives, the probability of failure in order to obtain a new data sampleConstructing the conditional probability of the observation sample under normal and fault conditions, and defining the following formula:
in the formula (20), in the following formula,representing the conditional probability of observing the sample under normal conditions in the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;is the control limit of each mode, whose value is given by the sum of the degrees of freedom d of the submodelsThe significance level alpha is uniquely determined;t representing the kth sub-modality2Statistics are obtained.
Step S500, collecting online data x (k), k =1, 2., N as a test set of models, and performing a normalization process.
The data set used for model training and testing was collected from a production plant of some typical ternary positive electrode material. The sintering link of the ternary material is crucial to the performance of the battery material, so that the sintering process of the material needs to be accurately monitored, and the aims of ensuring the product quality and reducing the energy consumption are fulfilled. A total of 7 observation variables are selected, the first 6 variables are important temperature measurement data of the heating area, and the last data are observation data representing the residual lithium in the product quality. A total of 2200 consecutive moments of data were selected, and the sequence data factory produced the same batch of battery material all the time, with a sampling period of 30min. The data set is preprocessed or standardized, and the standardization operation is as follows: the average value of the sample of the variable to which each element in the sample set belongs is subtracted, and then the average value is divided by the standard deviation of the sample, so that the average value of the data corresponding to each process variable and each key quality variable is 0, and the variance is 1.
Step S600, detecting a test set based on the obtained factor dynamic autoregressive model, and calculating the test sample under each submodeAnd merging each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withComparison with significance level αAnd outputs the detection result.
After the probability of the fault of each local model of the observation data is determined, the probability of the fault of each sub-model is further fused by using a Bayesian inference technology to obtain the fault probability of the sampleProbability of failure of a sampleCalculated by the following formula:
in the case of the formula (21),denotes the failure probability of the sample, P (k | x)t) Factor k is related to xtA posterior probability of (d);indicating the relation to a new sample x under a factor modality ktThe posterior probability of failure.
Whether the system fails or not is judged by comparing the failure probability with the significance level alpha, and the judgment logic is shown as the following formula:
in the formula (22), the first and second groups,indicating the probability of failure of the sample and alpha indicating the level of significance.
The method for monitoring the preparation process of the ternary cathode material based on the FDALM, provided by the embodiment, is applied to monitoring the sintering process of the ternary cathode material, the sintering process of the ternary cathode material is a typical process industry, and the production process involves a plurality of chemical reactions coupled with each other, including chemical combination, hydrolysis and side reactions. These reactions, if poorly controlled, lead to the occurrence of side reactions which are reversible and therefore fluctuating changes in the external environment and materials lead to the process being in a different steady state. Leading to certain multimodal properties of the process. Meanwhile, the coupling of different temperature zones enables the data to be related front and back, namely the dynamic property of the process data is not negligible, so that the process data has complex characteristics. Firstly, deducing a factor FDALM modeling method by means of a factor modeling method on the basis of the technology of a dynamic autoregressive hidden variable model, wherein the factor FDALM modeling method models data with dynamic and multi-modal characteristics at the same time, and learns model parameters by utilizing an improved EM algorithm; then, in order to give full play to the process output of each factor model, the statistical values of the sub models are fused into the posterior fault probability of the sample by means of a Bayesian inference technology; finally, simulation results compared with other models show that the proposed monitoring method can track modal fluctuations of the process.
Referring to fig. 7 to 9, the present invention further provides a system for monitoring the preparation process of the FDALM-based ternary cathode material, which includes a preprocessing module 10, an establishing module 20, an identifying module 30, a determining module 40, a collecting module 50 and a calculating module 60, wherein the preprocessing module 10 is used for preparing and preprocessing data and collecting historical data xh(k) K =1, 2., T as a model training set, each sample was normalized; wherein the content of the first and second substances,the data are observed in the m-dimensional process, k is a time label, and T is the number of samples; the establishing module 20 is used for respectively identifying the time lag coefficient L and the system factor K of the model by using time lag identification and clustering algorithm, establishing a factor dynamic autoregressive hidden variable model and initializing model parameters thetaold(ii) a An identification module 30 for identifying relevant parameters of the factor-dynamic autoregressive hidden variable model using the improved EM algorithm(ii) a A determining module 40, configured to define, based on the obtained factor dynamic autoregressive hidden variable model, a factor dynamic autoregressive hidden variable model in each sub-modeStatistics and determining control thresholds for submodelsAnd a significance level α; a collecting module 50, configured to collect online data x (k), k =1,2, ·, N as a test set of the model, and perform a normalization process; a calculating module 60, configured to detect the test set based on the obtained factor dynamic autoregressive model, and calculate the test sample in each sub-modeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withAnd comparing with the significance level alpha, and outputting a detection result.
Further, please refer to fig. 8, fig. 8 is a schematic functional module diagram of an embodiment of the establishing module shown in fig. 7, in this embodiment, the establishing module 20 includes a first obtaining unit 21, a second obtaining unit 22 and a first constructing unit 23, where the first obtaining unit 21 is configured to obtain the time lag coefficient L through a trend similarity algorithm; the second obtaining unit 22 is configured to obtain a system factor K through identification by an affine clustering propagation algorithm based on a genetic algorithm, where the system factor K is physically interpreted as a number of types of data division; the first constructing unit 23 is configured to construct a factor dynamic autoregressive hidden variable model by using the preprocessed data set.
Preferably, referring to fig. 9, fig. 9 is a functional module schematic diagram of an embodiment of the identification module shown in fig. 7, in this embodiment, the identification module 30 includes an estimation unit 31 and a second construction unit 32, where the estimation unit 31 is configured to reasonably estimate, at step E, the posterior distribution of the expanded dynamic hidden variable expectation and the system factor K based on bayesian filtering and smoothing; and the second construction unit 32 is configured to update the model parameters by using a method of maximizing a likelihood function in the step M, construct a lagrangian multiplier formula, and constrain and update the factor coefficients by using the factors in the step M.
Further, in the determining module 40, for the trained factor dynamic autoregressive hidden variable model, the hidden variables are key variables for driving the operation of the dynamic change process, and the hidden variables of each sub-model can establish correspondingThe statistical quantity is calculated by the statistical quantity,the statistic is calculated by the following formula:
in the formula (23), the first and second groups,t representing the kth sub-modality2Statistic, E (z)t q|x1:t q,y1:t q)TRepresenting hidden variables zt qAbout x1:t q,y1:t qCovariance, which represents the hidden variable zt qWith respect to x1:t q,y1:t qCovariance of E (z)t q|xl:t q,yl:t q) Representing hidden variables zt qAbout xl:t q,yl:t qThe conditions of (a) are expected.
In order to fully utilize the key information of each mode, the monitoring results of each sub-mode are fused into fault probability by means of a Bayesian inference method, and the event probability that an observation sample fails in the kth mode is constructed as follows:
in the case of the formula (24),represents a prior probability of a failure of the process data,represents a prior probability that the process data is normal,representing the conditional probability of the occurrence of the observation sample under the k-th modal fault condition;representing the probability of the observed variable being observed in the kth modality;representing the condition probability of the observation sample under the normal condition of the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition in the kth mode;representing a process data failure prior probability;representing the normal prior probability of the process data;
will be provided withAndin combination with the significance level α, the following formula is defined:
in formula (25), α represents the significance level; the actual size is a balance between false positives and false negatives, the probability of failure in order to obtain a new data sampleConstructing the conditional probability of the observation sample under normal and fault conditions, and defining the following formula:
in the formula (26), the first and second groups,representing the conditional probability of observing the sample under normal conditions in the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;is a control limit for each modality, the value of which is uniquely determined by the degree of freedom d and the significance level α of the submodel;t representing the kth sub-modality2Statistics are obtained.
Further, in the calculation module 60, after the probability of the fault occurrence of each local model of the observation data is determined, the probability of the fault occurrence of each sub-model is further fused by using the bayesian inference technology to obtain the fault probability of the sampleProbability of failure of a sampleCalculated by the following formula:
in the formula (27), the first and second groups,denotes the failure probability of the sample, P (k | x)t) Factor k is related to xtA posterior probability of (d);indicating the relation to a new sample x under a factor modality ktThe posterior probability of failure.
Whether the system fails or not is judged by comparing the failure probability with the significance level alpha, and the judgment logic is shown as the following formula:
in the formula (28), the first and second groups,indicating the probability of failure of the sample and alpha indicating the level of significance.
The FDALM-based ternary cathode material preparation process monitoring system provided in this embodiment is applied to monitoring of a sintering process of a ternary cathode material, the sintering process of the ternary cathode material is a typical process industry, and a production process involves numerous chemical reactions coupled to each other, including chemical combination, hydrolysis, and side reactions. These reactions can lead to side reactions if the temperature is not well controlled, which are reversible, and thus, fluctuating changes in the external environment and materials can lead to the process being in a different steady state process. Leading to certain multimodal properties of the process. Meanwhile, the coupling of different temperature zones enables the data to be related front and back, namely the dynamic property of the process data is not negligible, so that the process data has complex characteristics. Firstly, deducing a factor FDALM modeling method by means of a factor modeling method on the basis of the technology of a dynamic autoregressive hidden variable model, wherein the factor FDALM modeling method models data with dynamic and multi-modal characteristics at the same time, and learns model parameters by utilizing an improved EM algorithm; then, in order to give full play to the process output of each factor model, the statistical values of the sub models are fused into the posterior fault probability of the sample by means of a Bayesian inference technology; finally, simulation results compared with other models show that the proposed monitoring system can track modal fluctuations of the process.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including the preferred embodiment and all changes and modifications that fall within the scope of the invention. It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (10)
1. A method for monitoring the preparation process of a ternary cathode material based on FDALM is characterized by comprising the following steps:
data preparation and preprocessing, collecting historical data xh(k) K =1, 2., T as a model training set, each sample was normalized; wherein, the first and the second end of the pipe are connected with each other,the data are observed in an m-dimensional process, k is a time label, and T is the number of samples;
respectively identifying a time lag coefficient L and a system factor K of the model by using a time lag identification and clustering algorithm, establishing a factor dynamic autoregressive hidden variable model, and initializing a model parameter thetaold;
Identifying relevant parameters of the factor dynamic autoregressive hidden variable model by adopting an improved EM algorithm;
defining T of the factor dynamic autoregressive hidden variable model under each sub-model state based on the obtained factor dynamic autoregressive hidden variable modelk 2Statistics and determining control thresholds for submodelsAnd a significance level α;
collecting online data x (k), k =1, 2., N as a test set of models and performing a normalization process;
detecting the test set based on the obtained factor dynamic autoregressive model, and calculating the test sample under each submodeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withAnd comparing with the significance level alpha, and outputting a detection result.
2. The FDALM-based triad of claim 1The method for monitoring the preparation process of the meta-anode material is characterized in that a time lag coefficient L and a system factor K of a model are respectively identified by using a time lag identification and clustering algorithm, a factor dynamic autoregressive hidden variable model is established, and a model parameter theta is initializedoldComprises the following steps:
identifying and obtaining the time-lag coefficient L by a trend similarity algorithm;
identifying and obtaining a system factor K through an affine clustering propagation algorithm based on a genetic algorithm, wherein the system factor K is physically interpreted as the number of types of data division;
and constructing a factor dynamic autoregressive hidden variable model by utilizing the preprocessed data set.
3. The FDALM-based ternary positive electrode material manufacturing process monitoring method of claim 2, wherein said step of identifying relevant parameters of said factorial dynamic autoregressive hidden variable model using a modified EM algorithm comprises:
in the step E, reasonably estimating the posterior distribution of the expanded dynamic hidden variable expectation and the system factor K based on Bayesian filtering and smoothing;
in the step M, model parameters are updated by means of a method of maximizing a likelihood function, a Lagrange multiplier formula is constructed, and factor constraint updating factor coefficients in the step M are utilized.
4. The FDALM-based ternary positive electrode material preparation process monitoring method of claim 1, wherein the factor dynamic autoregressive hidden variable model is defined based on the obtained factor dynamic autoregressive hidden variable model in each sub-modeStatistics and determining control thresholds for submodelsAnd in the step of the significance level alpha, for the trained factor dynamic autoregressive hidden variable model, the hidden variable is a driving motionThe key variable of the operation of the state change process, namely the hidden variable of each sub-mode can establish the corresponding hidden variableThe statistical quantity is calculated by the statistical quantity,the statistic is calculated by the following formula:
wherein the content of the first and second substances,t representing the kth sub-modality2Statistic, E (z)t q|x1:t q,y1:t q)TRepresenting an implicit variable zt qWith respect to x1:t q,y1:t qCovariance, which represents the hidden variable zt qWith respect to x1:t q,y1:t qCovariance of E (z)t q|x1:t q,yl:t q) Representing hidden variables zt qWith respect to x1:t q,y1:t qThe conditions of (a);
in order to fully utilize the key information of each mode, the monitoring results of all the sub-modes are fused into fault probability by means of a Bayesian inference method, and the fault event probability of an observation sample in the kth mode is constructed as follows:
wherein, the first and the second end of the pipe are connected with each other,represents a prior probability of a failure of the process data,represents a prior probability that the process data is normal,representing the conditional probability of the occurrence of the observation sample under the k-th modal fault condition;representing the probability of the observed variable being observed in the k-th modality;representing the conditional probability of the observation sample under the normal condition of the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;representing a process data failure prior probability;representing the normal prior probability of the process data;
wherein α represents a significance level; actual size is a balance between false positives and false negatives, failure probability for obtaining new data samplesConstructing the conditional probability of the observation sample under normal and fault conditions, and defining the following formula:
wherein the content of the first and second substances,representing the conditional probability of the occurrence of the observation sample under normal conditions in the kth modality;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;is a control limit for each modality, the value of which is uniquely determined by the degree of freedom d and the significance level α of the submodel;t representing the kth sub-modality2Statistics are obtained.
5. The FDALM-based ternary positive electrode material preparation process monitoring method of claim 1, wherein the FDALM-based ternary positive electrode material preparation process monitoring method is characterized in that a test set is detected based on the obtained factor dynamic autoregressive model, and the value of a test sample in each sub-mode is calculatedAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withIn the step of comparing with the significance level alpha and outputting the detection result, after the probability of the fault of each local model of the observation data is determined, the probability of the fault of each sub-model is further fused by using the Bayesian inference technology to obtain the fault probability of the sampleProbability of failure of a sampleCalculated by the following formula:
wherein, the first and the second end of the pipe are connected with each other,denotes the failure probability of the sample, P (k | x)t) Factor k with respect to xtA posterior probability of (d);indicating the relation to a new sample x under a factor modality ktThe posterior probability of failure of (1);
whether the system fails or not is judged by comparing the failure probability with the significance level alpha, and the judgment logic is shown as the following formula:
6. A FDALM-based ternary positive electrode material preparation process monitoring system is characterized by comprising:
a preprocessing module (10) for data preparation and preprocessing, for collecting historical data xh(k) K =1, 2., T as a model training set, each sample was normalized; wherein, the first and the second end of the pipe are connected with each other,the data are observed in the m-dimensional process, k is a time label, and T is the number of samples;
the establishing module (20) is used for respectively identifying the time lag coefficient L and the system factor K of the model by using time lag identification and clustering algorithm, establishing a factor dynamic autoregressive hidden variable model and initializing model parameters thetaold;
An identification module (30) for identifying relevant parameters of the factor dynamic autoregressive hidden variable model by adopting an improved EM algorithm;
a determination module (40) for dynamically self-determining based on the derived factorsA regression hidden variable model for defining the factor dynamic autoregressive hidden variable model in each sub-model stateStatistics and determining control thresholds for submodelsAnd a significance level α;
a collection module (50) for collecting online data x (k), k =1, 2., N as a test set of models and performing a normalization process;
a calculation module (60) for detecting the test set based on the obtained factor dynamic autoregressive model and calculating the test sample in each sub-modeAnd fusing each sub-modality using Bayesian inference techniquesStatistics to obtain the posterior fault probability of the sampleWill be provided withAnd comparing with the significance level alpha, and outputting a detection result.
7. The FDALM-based ternary positive electrode material production process monitoring system of claim 6, wherein the establishing module (20) comprises:
a first obtaining unit (21) for identifying and obtaining the time lag coefficient L by a trend similarity algorithm;
the second obtaining unit (22) is used for obtaining a system factor K through affine clustering propagation algorithm identification based on a genetic algorithm, and the system factor K is physically interpreted as the number of types of data division;
and a first construction unit (23) for constructing a factor dynamic autoregressive hidden variable model by using the preprocessed data set.
8. The FDALM-based ternary positive electrode material production process monitoring system of claim 7, wherein the identification module (30) comprises:
the estimation unit (31) is used for reasonably estimating the expanded dynamic hidden variable expectation and the posterior distribution of the system factor K based on Bayesian filtering and smoothing in the step E;
and the second construction unit (32) is used for updating the model parameters by means of a method of maximizing the likelihood function in the step M, constructing a Lagrange multiplier formula and utilizing factors in the step M to restrain and update factor coefficients.
9. The FDALM-based ternary positive electrode material preparation process monitoring system of claim 6, wherein in the determining module (40), for the trained factor dynamic autoregressive hidden variable model, the hidden variables are key variables for driving the operation of the dynamic variation process, and the hidden variables of each sub-mode can establish a corresponding hidden variableThe statistical quantity is calculated by the statistical quantity,the statistic is calculated by the following formula:
wherein the content of the first and second substances,t representing the kth sub-modality2Statistic, E (z)t q|x1:t q,y1:t q)TRepresenting hidden variables zt qWith respect to x1:t q,y1:t qCovariance, which represents the hidden variable zt qWith respect to x1:t q,y1:t qCovariance of E (z)t q|x1:t q,yl:t q) Representing hidden variables zt qWith respect to x1:t q,y1:t qThe conditions of (a);
in order to fully utilize the key information of each mode, the monitoring results of all the submodes are fused into fault probability by means of a Bayesian inference method, and the probability of the fault event of the observation sample in the kth mode is constructed as follows:
wherein the content of the first and second substances,represents a prior probability of a failure of the process data,a priori probability of the process data being normal is represented,representing the conditional probability of the occurrence of the observation sample under the k-th modal fault condition;representing the probability of the observed variable being observed in the kth modality;representing the conditional probability of the observation sample under the normal condition of the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;representing a process data failure prior probability;representing the normal prior probability of the process data;
will be provided withAndin combination with the significance level α, the following formula is defined:
wherein α represents a significance level; actual size is a balance between false positives and false negatives, failure probability for obtaining new data samplesConstructing the conditional probability of the observation samples under normal and fault conditions, and defining the following formula:
wherein, the first and the second end of the pipe are connected with each other,representing the conditional probability of the observation sample under normal conditions in the kth mode;representing the conditional probability of the occurrence of the observation sample under the fault condition of the kth mode;is a control limit for each modality, the value of which is uniquely determined by the degree of freedom d and the significance level α of the submodel;t representing the kth sub-modality2And (4) statistic amount.
10. The FDALM-based ternary positive electrode material preparation process monitoring system of claim 6, wherein in the calculation module (60), after the probability of the occurrence of the fault of each local model of the observed data is determined, the probability of the occurrence of the fault of each sub-model is further fused by using a Bayesian inference technology to obtain the fault probability of the sampleProbability of failure of a sampleCalculated by the following formula:
wherein the content of the first and second substances,denotes the failure probability of the sample, P (k | x)t) Factor k with respect to xtA posterior probability of (d);indicating the relation to a new sample x under a factor mode ktThe posterior probability of failure of (a);
and judging whether the system fails or not by comparing the failure probability with the significance level alpha, wherein the judgment logic is shown as the following formula:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210720684.7A CN115270601A (en) | 2022-06-23 | 2022-06-23 | FDALM-based method and system for monitoring preparation process of ternary cathode material |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210720684.7A CN115270601A (en) | 2022-06-23 | 2022-06-23 | FDALM-based method and system for monitoring preparation process of ternary cathode material |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115270601A true CN115270601A (en) | 2022-11-01 |
Family
ID=83761879
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210720684.7A Pending CN115270601A (en) | 2022-06-23 | 2022-06-23 | FDALM-based method and system for monitoring preparation process of ternary cathode material |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115270601A (en) |
-
2022
- 2022-06-23 CN CN202210720684.7A patent/CN115270601A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Zhao et al. | Soft sensor modeling of chemical process based on self-organizing recurrent interval type-2 fuzzy neural network | |
Tian et al. | Identification of abnormal conditions in high-dimensional chemical process based on feature selection and deep learning | |
CN114077876B (en) | Strip steel hot continuous rolling multi-mode process monitoring method and device | |
CN113325721B (en) | Model-free adaptive control method and system for industrial system | |
Wang et al. | Fault detection and diagnosis for multiple faults of VAV terminals using self-adaptive model and layered random forest | |
CN114282443B (en) | Residual service life prediction method based on MLP-LSTM supervised joint model | |
CN115096627B (en) | Method and system for fault diagnosis and operation and maintenance in manufacturing process of hydraulic forming intelligent equipment | |
Tang et al. | A deep learning model for measuring oxygen content of boiler flue gas | |
Liu et al. | Model fusion and multiscale feature learning for fault diagnosis of industrial processes | |
CN114015825A (en) | Method for monitoring abnormal state of blast furnace heat load based on attention mechanism | |
CN114403486B (en) | Intelligent control method of airflow type cut-tobacco drier based on local peak value coding circulation network | |
CN110427019B (en) | Industrial process fault classification method and control device based on multivariate discriminant analysis | |
CN110084301B (en) | Hidden Markov model-based multi-working-condition process working condition identification method | |
CN110245398B (en) | Soft measurement deep learning method for thermal deformation of air preheater rotor | |
CN113705897B (en) | Product quality prediction method and system for industrial copper foil production | |
Li et al. | Data cleaning method for the process of acid production with flue gas based on improved random forest | |
CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
CN115270601A (en) | FDALM-based method and system for monitoring preparation process of ternary cathode material | |
WO2023231374A1 (en) | Semi-supervised fault detection and analysis method and apparatus for mechanical device, terminal, and medium | |
CN116020879A (en) | Technological parameter-oriented strip steel hot continuous rolling space-time multi-scale process monitoring method and device | |
CN115438309A (en) | Unmanned aerial vehicle engine health assessment method based on LSTM-SVR | |
CN113151842B (en) | Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production | |
Liu et al. | STA-APSNFIS: STA-optimized adaptive pre-sparse neuro-fuzzy inference system for online soft sensor modeling | |
CN116391193A (en) | Method and apparatus for energy-based latent variable model based neural networks | |
CN113341926B (en) | Multi-stage intermittent process fault detection method based on sparse weighted neighborhood preserving embedding |
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 |