Adaptive identification method and system of battery stack flow data based on tensor-supported train machine

Through the improved least squares method and gradient descent method combined with tensor train decomposition technology, the estimation inaccuracy and complexity of tensor train machines in high-dimensional tensor data processing is solved, and high-precision time-varying parameters are achieved Online interactive estimation meets real-time requirements and improves estimation accuracy.

CN118627396BActive Publication Date: 2024-11-15GUANGDONG UNIV OF TECH
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411081089.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-11-15
Estimated Expiration
2044-08-08

AI Technical Summary

Technical Problem

Existing tensor-supporting train machines have problems of estimation inaccuracy and high computational complexity when processing high-dimensional tensor data, especially when dealing with large-scale parameters of high-dimensional tensors with inherent complexity, which cannot be met. Real-time requirements.

Method used

By improving the least squares method and gradient descent method, combined with tensor train decomposition technology, online interactive estimation of time-varying weight tensors and bias terms in the support vector regression model is realized, and by compressing the weight of the TT format of STM, it reduces the weight of the STM Computational complexity.

Benefits of technology

It significantly reduces the calculation complexity and parameter estimation error, improves the estimation accuracy of time-varying parameters, meets the real-time requirements, and performs well in high-dimensional models, and can effectively process high-dimensional flow data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118627396B_ABST
    Figure CN118627396B_ABST
Patent Text Reader

Abstract

The present disclosure provides a method and system for adaptively identifying stack flow data based on a tensor train machine, which relates to the technical field of machine learning parameter identification, including: acquiring stack flow data at a set sampling time interval, characterizing a solid oxide fuel cell stack system with a support vector regression model, acquiring a weight tensor in the support vector regression model, performing a tensor train decomposition on the weight tensor, and constructing a tensor train machine model; solving the tensor train machine model, estimating unknown time-varying parameter states, converting the solution to the tensor train machine model into an optimization problem, constructing an objective function, and using a least squares algorithm combined with a gradient descent method to update and solve. When the error meets the requirement, an estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at a future moment is obtained, and online identification of the parameters is completed for subsequent fault detection and identification.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of machine learning parameter identification, and in particular to a method and system for adaptively identifying battery stack flow data based on a tensor train machine. Background Art

[0002] One of the important prerequisites for the smooth and efficient operation of the SOFC power generation system is to strictly monitor the temperature distribution in the stack space. In order to accurately control the spatial temperature of the SOFC system, it is necessary to monitor the operating status of the system, detect and handle various anomalies in a timely manner, and supervise the various characteristics in the stack.

[0003] Mathematical models are the expression of physical relationships and the mapping of input-output relationships of industrial processes. They play a very important role in the automatic control and state estimation of battery stacks. There are generally two methods to establish a mathematical model of a system: one is to derive it through understanding the internal structure of the system and some basic theorems and basic principles. This method is called analytical method. However, in most actual systems, it is impossible to understand its internal structure, and it is difficult to use this method to establish a model. Another method can usually be used: system identification, which uses appropriate optimization strategies to extract mathematical models from the output and input of the system containing errors. According to different execution methods, system identification methods can be divided into online identification methods and offline identification methods. Offline identification is an identification method that performs parameter estimation only after all experimental data are collected. In online identification, data collection and parameter estimation are carried out simultaneously, which can meet the real-time requirements of automatic control.

[0004] In addition, based on statistical learning theory, support vector machine (SVM) was proposed as a mathematical model with high classification accuracy and good generalization ability. It finds an optimal hyperplane by minimizing structural risk, separates two types of samples with the largest interval, and as many sample points as possible fall into it. The training process of SVM can be in the form of solving quadratic programming problems, and a unique global optimal solution can be obtained, rather than a local minimum solution. However, the standard support vector machine model is only suitable for processing vector input samples. For high-order input tensors, SVM needs to reshape them into vectors before processing, which destroys the data structure, ignores the connection between elements, and makes it difficult to process tensor data efficiently. Therefore, in order to retain the structural information of multidimensional data, tensors should be used to store data, and tensor decomposition methods can be used to reduce storage costs. The standard linear SVM is extended to a general tensor form to obtain the support tensor machine (STM), which replaces the vector input with the tensor input and decomposes the corresponding weight vector into a rank 1 tensor. Although STM solves the overfitting problem, the expressive power of the rank 1 weight tensor is limited, resulting in poor classification accuracy. In order to enhance the expressiveness of the model, the rank-1 weight tensor of STM is generalized to the canonical polyadic (CP) and Turker forms. However, CP decomposition has the problem of being difficult to determine a good rank, and the number of parameters of Turker decomposition is exponential and still affected by the curse of dimensionality. Tensor Train (TT) decomposition can overcome these two problems of CP decomposition and Turker decomposition. Therefore, after combining TT decomposition and STM, a Support Tensor Train Machine (STTM) was proposed.

[0005] Both the TT decomposition method and STTM have good performance, but they have certain limitations when used for STM parameter estimation. For example, when dealing with large-scale parameters of high-dimensional tensors with inherent complexity, the accuracy of the estimation algorithm cannot be guaranteed and has high computational complexity. In addition, time-varying data is often accumulated in the actual process, which poses a challenge to the existing STTM method running in batch mode. Therefore, it is very necessary to develop an efficient online identification method to solve these problems. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a method and system for adaptive identification of battery stack flow data based on supporting tensor train machines, which realizes adaptive online interactive estimation of unknown parameters by improving the least squares method and gradient descent method, calculates unknown time-varying parameters with high precision, and significantly reduces the difficulty and complexity of calculation by compressing the weights in the TT format of STM.

[0007] According to some embodiments, the present disclosure adopts the following technical solutions:

[0008] The method for adaptively identifying stack flow data based on a tensor train machine includes: acquiring stack flow data according to a set sampling time interval, characterizing a solid oxide fuel cell stack system with a support vector regression model, acquiring a weight tensor in the support vector regression model, performing a tensor train decomposition on the weight tensor, and constructing a tensor train machine model;

[0009] Solve the supporting tensor train machine model, estimate the unknown time-varying parameter state, transform the solution of the supporting tensor train machine model into an optimization problem, construct the objective function, and use the least squares algorithm combined with the gradient descent method to update the solution. When the error meets the requirements, the estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at the future moment is obtained, and the online identification of the parameters is completed for subsequent fault detection and identification.

[0010] According to some embodiments, the present disclosure adopts the following technical solutions:

[0011] The battery stack flow data adaptive identification system based on the tensor train machine includes:

[0012] A model building module is used to obtain the stack flow data according to the set sampling time interval, characterize the solid oxide fuel cell stack system with a support vector regression model, obtain the weight tensor in the support vector regression model, perform a tensor train decomposition on the weight tensor, and build a support tensor train machine model;

[0013] The parameter identification and solution module is used to solve the tensor support train machine model, estimate the unknown time-varying parameter state, transform the solution of the tensor support train machine model into an optimization problem, construct the objective function, and use the least squares algorithm combined with the gradient descent method to update the solution. When the error meets the requirements, the estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at the future time is obtained, and the online identification of the parameters is completed for subsequent fault detection and identification.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions:

[0015] A non-transitory computer-readable storage medium, which is used to store computer instructions. When the computer instructions are executed by a processor, the method for adaptively identifying battery stack flow data based on supporting a tensor train machine is implemented.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions:

[0017] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for adaptively identifying battery stack flow data based on supporting tensor train machines.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The disclosed method for adaptive identification of battery stack flow data based on a tensor train machine can decompose the weight tensor in the tensor machine into multiple TT cores through TT decomposition, thereby avoiding the influence of the curse of dimensionality. Under the assumption that the weight tensor is a slowly changing flow data, when estimating the last TT core, only the time slice of the last TT core is estimated each time instead of the entire TT core, thereby further compressing the TT format of the weight tensor to achieve the effect of reducing the computational complexity of the method. On this basis, combined with the improved least squares method and gradient descent method, online interactive estimation of time-varying weight tensors and bias terms in the support vector regression model is realized, which can meet the real-time requirements.

[0020] The disclosed method for adaptive identification of battery stack flow data based on tensor train machine supports the parameter estimation error of the proposed method over time. decreases significantly with the increase of Next, the time-varying factor The smaller the corresponding estimation error, the better the method is for STTM models with higher dimensions. The estimation result is better as the value increases, which shows that reusing part of the past sampling data can effectively improve the estimation accuracy of time-varying parameters. In addition, the noise level has an impact on the convergence speed and estimation accuracy, that is, the lower the noise level, the better the estimation effect of the proposed method. However, even in In the case of , this method can still provide relatively good estimation results. The simulation results prove that the proposed method is effective for STTM regression model with high-dimensional flow data parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0022] Figure 1 The overall process of parameter estimation of the embodiment of the present disclosure;

[0023] Figure 2 is the weight tensor of the embodiment of the present disclosure The dimension of Time varying factors Parameter error variation diagram under ;

[0024] Figure 3 is the weight tensor of the embodiment of the present disclosure The dimension of Time varying factors Parameter error variation diagram under ;

[0025] Figure 4 is the weight tensor of the embodiment of the present disclosure The dimension of At different times Parameter error variation diagram under different values;

[0026] Figure 5 is the weight tensor of the embodiment of the present disclosure The dimension of At different noise variances The parameter error variation diagram below. DETAILED DESCRIPTION

[0027] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0028] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0029] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0030] Example 1

[0031] In one embodiment of the present disclosure, a method for adaptively identifying stack flow data based on a support tensor train machine is provided, comprising: step 1: acquiring stack flow data according to a set sampling time interval, characterizing a solid oxide fuel cell stack system with a support vector regression model, acquiring a weight tensor in the support vector regression model, performing a tensor train decomposition on the weight tensor, and constructing a support tensor train machine model;

[0032] Step 2: Solve the supporting tensor train machine model, estimate the unknown time-varying parameter state, transform the solution of the supporting tensor train machine model into an optimization problem, construct the objective function, and use the least squares algorithm combined with the gradient descent method to update the solution. When the error meets the requirements, the estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at the future time is obtained, and the online identification of the parameters is completed for subsequent fault detection and identification.

[0033] As an embodiment, the disclosed method for adaptive identification of battery stack flow data based on tensor train machine supports the method. The method combines the improved least square method and the gradient descent method to realize adaptive online interactive estimation of unknown parameters, and can calculate unknown time-varying parameters with high precision. By compressing the weights of the TT format of STM, the difficulty and complexity of calculation are significantly reduced. The specific implementation process is as follows:

[0034] Step 1: Decompose the weight tensor in the support vector regression model by Tensor Train (TT) and build the Support Tensor Train Machine (STTM) model.

[0035] Specifically, for a solid oxide fuel cell (SOFC) stack system, the stack can be divided into Each node is equipped with a corresponding control unit according to the characteristics of the controlled object. The control variables include the flow rate of the fluid, the amount of the substance, the time and space distribution of the temperature control, etc. variables, and on this basis, considering the autocorrelation of variables in time and space, the time window is introduced ( ), spatial window ( Finally, sampling is performed at a certain sampling interval to obtain a stream data .

[0036] The SOFC stack system can be characterized by a support vector regression model:

[0037]

[0038] in, Is the input data At the moment slices, is the weight tensor to be estimated At the moment slices, is the bias term, It is the content of fuel, air or battery stack, which can be used for subsequent fault detection and identification.

[0039] In the support vector regression model, the weight tensor It is a feature weight. By giving different weights to different features, the model can better learn the relationship between data, thereby improving the accuracy and generalization ability of the model. With input data The same dimension dimensional stream data, where its last dimension Will go on with time increases, while other dimensions is fixed. Therefore, the stream data Can be regarded as the data of the previous moment Add a new slice It can be expressed as an operation: .in It is the concatenation operation of two tensors, which is as follows,

[0040]

[0041] If we estimate the entire weight tensor directly , due to the dimension and time The increase in the weight tensor leads to huge computational complexity. Perform TT tensor decomposition. Through TT decomposition, the weight tensor It can be expressed as,

[0042]

[0043] in, , It is called TT core, corresponding to is the TT rank, Each element in can be calculated by the following formula:

[0044]

[0045] Therefore, it is necessary to estimate that TT cores , rather than the entire weight tensor , to reduce the computational complexity of the algorithm. Define variables ,

[0046]

[0047] in, Indicates time Streaming data Assuming that the TT kernel is slowly changing, it can be approximated , and then the following relationship is derived:

[0048]

[0049] Combining the above formula, we can get the TT core and it in the moment Slices relation: If every moment The estimated TT core , its computational complexity will increase as and by choosing each moment Only estimate slices , which can keep the amount of calculation fixed. Therefore, the parameters that the STTM method needs to estimate are the TT kernel ,slice and the bias term , which is faster than directly estimating the weight tensor and the bias term The amount of calculation has been significantly reduced.

[0050] Step 2: Solve the support tensor train machine model and estimate the unknown time-varying parameter states.

[0051] Specifically, estimating the unknown time-varying The separating hyperplane of STTM can be expressed as,

[0052] (3)

[0053] in and They are weight tensors And the input sample At the moment When the time is calculated Estimated value of , can be calculated according to formula (2) ,

[0054]

[0055] definition , , formula (3) can be restated as,

[0056] (4)

[0057] Here, use the moment Previous Samples to estimate Instead of using this samples, reducing the computational burden. Definition:

[0058]

[0059] We can further obtain

[0060]

[0061] in , and From the time To time this Output data, input data and bias terms A vector of estimates of .

[0062] but The estimated value of can be obtained by solving the following optimization problem,

[0063]

[0064] therefore, Can be based on improved The least squares algorithm of samples is used to update the obtained

[0065]

[0066] in By the time Before Parameter values ​​of the TT cores Calculated The expansion matrix of is a parameter The estimated value at time t.

[0067] Step 3: Estimate the unknown time-varying In order to estimate the TT cores ,

[0068] , we can use STTM about The separating hyperplane of is reformulated as,

[0069]

[0070] in Is After TT decomposition, there is no In the case of, we can get the sum, that is, Therefore, formula (8) can be reconstructed into a vector form,

[0071]

[0072] in and TT core and variables In order to minimize the structural risk, STTM is used to update The optimization problem can be expressed as:

[0073] in is the error variable, which represents the error between the actual output and the predicted output, is the regularization factor, It is TT cores At the moment The vectorization of is the bias term.

[0074] Optimization Problem (10) Substitution Error After that, you can get information about The loss function as follows,

[0075]

[0076] Then the loss function about The gradient of can be calculated as,

[0077]

[0078] Define two variables as,

[0079]

[0080] in For the TT cores The number of parameters contained.

[0081] The following regression relationship can be obtained:

[0082]

[0083] in and Variables and The value at the previous moment.

[0084] Then equation (14) can be reformulated as

[0085]

[0086] Based on the above definition, we can further get The Hessian matrix is

[0087]

[0088] in yes The identity matrix of .

[0089] at last, The estimated value of can be calculated as

[0090]

[0091] in For the TT cores At the moment The estimated value of It's time The Hessian matrix of .

[0092] Step 4: Estimate the unknown bias term To estimate the bias term , the loss function of STTM can be defined as,

[0093]

[0094] Then the loss function about The gradient of is,

[0095]

[0096] definition , formula (20) can be re-expressed as,

[0097]

[0098] Finally, at the moment Estimated value of By setting equation (21) equal to 0, we can get

[0099]

[0100] When the time is calculated Estimated value of , and , we can get the SOFC system at time Estimated value of fuel, air or stack content , which can be used for subsequent fault detection and identification.

[0101] Simulation experiment

[0102] Consider the following STTM model with streaming data:

[0103]

[0104] The estimated It is streaming data No. slices, is the bias term. Slice Generated by the following method,

[0105]

[0106] in is the time-varying factor, Yes and Gaussian distribution with zero mean and unit variance in the same dimension ,and and Initialized to a Gaussian distribution . Randomly generate input samples Gaussian distribution ,set up , The variance is The corresponding output sample can be calculated . The time-varying factors Set to , and Generating streaming data After that, set , and TT rank is , using the proposed method to estimate the STTM model (21), we can get the parameters At the moment Estimated value of and estimation error , over time The error change diagram of the changing parameters is shown in Figure 2 In order to test the performance of the proposed method in identifying the STTM model with higher-order flow data, it is assumed that the weight tensor in the STTM model (21) It is also generated according to formula (22), while ensuring that other conditions remain unchanged, and the following is also obtained: Figure 3 Parameter error variation diagram shown.

[0107] For testing The value and noise level, assuming that the weight tensor is a time-varying factor In the case of , it is generated according to formula (21): , respectively set and , while keeping other conditions unchanged, using the proposed method to estimate the STTM model (21) can be obtained as follows Figure 4 and Figure 5 Shown over time Error variation plot for varying parameters.

[0108] Depend on Figure 2 , Figure 3 It can be seen that the parameter estimation error of the proposed method decreases with time. decreases significantly with the increase of Next, the time-varying factor The smaller the corresponding estimation error, the better the method also works for STTM models with higher dimensions. Figure 4 It can be seen that the proposed method will The larger the value, the better the estimation result, which shows that reusing part of the past sampling data can effectively improve the estimation accuracy of time-varying parameters. In addition, Figure 5 It can be seen that the noise level has an impact on both the convergence speed and the estimation accuracy, that is, the lower the noise level, the better the estimation effect of the proposed method. In the case of , this method can still provide relatively good estimation results. Therefore, the simulation results prove that the proposed method is effective for STTM regression models with high-dimensional flow data parameters.

[0109] Example 2

[0110] In one embodiment of the present disclosure, a battery stack flow data adaptive identification system based on a tensor train machine is provided, comprising:

[0111] A model building module is used to obtain the stack flow data according to the set sampling time interval, characterize the solid oxide fuel cell stack system with a support vector regression model, obtain the weight tensor in the support vector regression model, perform a tensor train decomposition on the weight tensor, and build a support tensor train machine model;

[0112] The parameter identification and solution module is used to solve the tensor support train machine model, estimate the unknown time-varying parameter state, transform the solution of the tensor support train machine model into an optimization problem, construct the objective function, and use the least squares algorithm combined with the gradient descent method to update the solution. When the error meets the requirements, the estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at the future time is obtained, and the online identification of the parameters is completed for subsequent fault detection and identification.

[0113] Example 3

[0114] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the method for adaptively identifying battery stack flow data based on supporting a tensor train machine is implemented.

[0115] Example 4

[0116] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the method for adaptively identifying battery stack flow data based on supporting tensor train machines.

[0117] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0119] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for adaptively identifying battery stack flow data based on a tensor train machine, characterized in that: include: Acquire the stack flow data at a set sampling time interval, characterize the solid oxide fuel cell stack system with a support vector regression model, obtain the weight tensor in the support vector regression model, perform a tensor train decomposition on the weight tensor, and construct a support tensor train machine model; Solve the tensor train model, estimate the unknown time-varying parameter state, convert the solution of the tensor train model into an optimization problem, construct the objective function, and use the least squares algorithm combined with the gradient descent method to update the solution. When the error meets the requirements, the estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at the future time is obtained, and the online identification of the parameters is completed for subsequent fault detection and identification; The solid oxide fuel cell stack system is characterized by a support vector regression model: in, Is the input data At the moment slices, is the weight tensor to be estimated At the moment slices, is the bias term, is the fuel, air or stack content; The parameters supporting the tensor trainer model estimation are the TT kernel ,slice and the bias term , estimating the unknown time-varying , the separating hyperplane of the tensor train machine model is expressed as, in, and They are weight tensors and input data At the moment Slice of Estimated value of .

2. The method for adaptively identifying battery stack flow data based on a tensor-supported train machine according to claim 1, characterized in that: For the solid oxide fuel cell stack system, the stack is divided into multiple nodes along the airflow direction. Each node is equipped with a corresponding control unit according to the differences in the characteristics of the controlled object. The control variables include the flow rate of the fluid, the amount of substance, and the spatiotemporal distribution of temperature control. Considering the autocorrelation of the variables in time and space, multiple time windows and multiple space windows are introduced, and sampling is performed at the set sampling interval to obtain flow data.

3. The method for adaptively identifying battery stack flow data based on a tensor-supported train machine according to claim 1, characterized in that: In the support vector regression model, the weight tensor Is the same as the input data The same dimension dimensional stream data, where its last dimension Will go on with time increases, while other dimensions Is fixed, streaming data It is the data of the previous moment Add a new slice Get, the weight tensor Perform TT tensor decomposition. Through TT decomposition, the weight tensor It is expressed as: ;in, , It is the TT core.

4. The method for adaptively identifying battery stack flow data based on a tensor-supported train machine according to claim 3, characterized in that: Each element in is calculated by the following formula: Among them, the corresponding It is TT rank.

5. The method for adaptively identifying battery stack flow data based on a tensor-supported train machine according to claim 1, characterized in that: For update The objective function of the optimization problem is expressed as: in, is the error variable, which represents the error between the actual output and the predicted output, is the regularization factor, It is TT cores At the moment The vectorization of is the bias term.

6. The battery stack flow data adaptive identification system based on the tensor train machine is characterized by: include: A model building module is used to obtain the stack flow data according to the set sampling time interval, characterize the solid oxide fuel cell stack system with a support vector regression model, obtain the weight tensor in the support vector regression model, perform a tensor train decomposition on the weight tensor, and build a support tensor train machine model; The parameter identification and solution module is used to solve the tensor train model, estimate the unknown time-varying parameter state, convert the solution of the tensor train model into an optimization problem, construct the objective function, and use the least squares algorithm combined with the gradient descent method to update the solution. When the error meets the requirements, the estimated value of the fuel, air or stack content of the solid oxide fuel cell stack system at the future time is obtained, and the online identification of the parameters is completed for subsequent fault detection and identification; The solid oxide fuel cell stack system is characterized by a support vector regression model: in, Is the input data At the moment slices, is the weight tensor to be estimated At the moment slices, is the bias term, is the fuel, air or stack content; The parameters supporting the tensor trainer model estimation are the TT kernel ,slice and the bias term , estimating the unknown time-varying , the separating hyperplane of the tensor train machine model is expressed as, in, and They are weight tensors and input data At the moment Slice of Estimated value of .

7. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the method for adaptively identifying battery stack flow data based on a tensor train machine as described in any one of claims 1 to 5 is implemented.

8. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the method for adaptively identifying battery stack flow data based on supporting tensor train machines as described in any one of claims 1-5.

Citation Information

Patent Citations

  • Solid oxide fuel cell voltage prediction method, terminal equipment and storage medium

    CN112363060A

  • Nonlinear system identification method based on tensor product network B spline

    CN113779780A

  • Fuel cell failure diagnosis system using variable operating point and the method using it

    KR102418259B1