CN-121981021-A - Proxy model-based pile multi-scale multi-working-condition optimizing diagnosis method and system

CN121981021ACN 121981021 ACN121981021 ACN 121981021ACN-121981021-A

Abstract

The invention provides a stack multi-scale multi-working condition optimizing diagnosis method and system based on a proxy model, and relates to the technical field of fuel cells; the method comprises the steps of establishing a flow resistance network model with multiple physical fields dynamically coupled, solving a balance equation set of fluid distribution in a galvanic pile to obtain an original data set reflecting a flow distribution uniformity index and a total pressure drop of a system, training and constructing a nonlinear global performance proxy model based on a response surface method, wherein the performance proxy model comprises a multi-order interaction operator, executing a design optimizing mode or a performance diagnosis mode based on the performance proxy model, reversely screening out a structural parameter combination with the highest robustness score in the interval in the design optimizing mode, and outputting a diagnosis report containing performance bottlenecks in the performance diagnosis mode. The invention can realize the acceleration of multi-scale-full-working condition coupling optimizing of the galvanic pile, and has the second-level accurate design and diagnosis capability.

Inventors

FANG MING
HUANG JU
LEI XIANZHANG
LI YIXING
MA XIAOYU
ZHANG QI

Assignees

天府永兴实验室

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (10)

1. The pile multi-scale multi-working condition optimizing diagnosis method based on the proxy model is characterized by comprising the following steps of: Step S1, constructing a parameterized input vector comprising a structural variable subset and an operation variable subset, wherein the structural variable subset comprises macroscopic manifold parameters of a galvanic pile, the number of single cell stacking joints and microscopic flow channel geometric parameters of single cells, and the operation variable subset comprises operating temperature, operating pressure, reaction gas metering ratio, operating current density and inlet relative humidity; Step S2, a flow resistance network model with multiple physical fields dynamically coupled is established, a balance equation set of fluid distribution in a pile is solved based on kirchhoff topology law, and an original data set reflecting a flow distribution uniformity index and a total pressure drop of a system is obtained, wherein in the solving process, a variable property correction kernel which is in real-time linkage with the operation variable subset and a local loss coefficient correction kernel which dynamically changes along with a flow state are introduced; Step S3, training and constructing a nonlinear global performance proxy model based on a response surface method by utilizing the original data set to establish a mapping function which is driven by the structural variable subset and the operation variable subset together and is used for predicting the pile performance, wherein the performance proxy model comprises a multi-order interaction operator used for representing the coupling relation between different scale design variables and between the design variables and the operation variables; S4, executing a design optimizing mode or a performance diagnosis mode based on the performance agent model; In the design optimizing mode, receiving performance constraint of a target application scene in a preset operation interval, and reversely screening out a structural parameter combination with the highest robustness score in the interval; and in the performance diagnosis mode, acquiring the existing structural parameters of the electric pile to be evaluated, scanning the whole domain of the operation variable subset, and outputting a diagnosis report containing the performance bottleneck.
2. The proxy model-based pile multiscale multi-condition optimizing diagnostic method according to claim 1, wherein in the step S2, the flow resistance network model uses the following formula when calculating the cell branch pressure drop: In the formula (I), in the formula (II), In the event of a single cell voltage drop, In order to achieve a coefficient of resistance along the way, L is the total length of the flow channel, Is of the hydraulic diameter and is of the same type, In order to achieve a fluid density, As a result of the average flow rate of the fluid, Is a local loss pressure drop; Wherein the calculation formula of the partial loss pressure drop is as follows In which, in the process, Is a dynamic local loss coefficient.
3. The proxy model-based pile multiscale, operating condition optimizing diagnostic method of claim 2, wherein in step S2, the flow resistance network model uses the following formula when calculating manifold segment pressure drop: In the formula (I), in the formula (II), Is the first Pressure drop across the segment manifold, Is the first The on-way drag coefficient of the segment manifold, For the segment length to be a segment length, For the hydraulic diameter of the manifold, Is the first The average flow rate within the segment manifold, Is a dynamic pressure correction term.
4. The proxy model-based pile multiscale multi-condition optimizing diagnostic method according to claim 2, wherein in said step S2, said dynamic local loss coefficient Different calculation formulas are adopted according to the node type: when fluid is shunted from the manifold into the single cell flow channel, the calculation formula of the dynamic local loss coefficient is as follows: ; When fluid is collected into the manifold from the single cell flow channel, the calculation formula of the dynamic local loss system is as follows: ; The dynamic local loss coefficient when the fluid turns inside the flow channel The calculation formula of (2) is as follows: ; In the formula, As a local drag loss coefficient for fluid as it continues to flow along the manifold trunk at the manifold split point, As a local drag loss coefficient when fluid is diverted from the manifold trunk into the cell flow channel branches, To provide a local drag loss coefficient of the original fluid in the manifold after being disturbed by the lateral branch fluid at the confluence node, Is the total local resistance loss coefficient after the branch fluid of the single cell flow channel is converged into the main manifold channel, To be a local drag loss coefficient of the fluid turning in the flow channel, For a characteristic flow rate of fluid at the manifold main road node, For the characteristic flow rate of the fluid at the cell flow path branching node, For a characteristic flow rate of fluid at a continuation section or section upstream of a junction at the manifold main road node, As the reynolds number of the fluid at the turns inside the cell channels, For corresponding manifold main road characteristic flow rate The reynolds number of the fluid in the region, Characterised by flow rates for continuation or disturbance of the corresponding manifold The reynolds number of the fluid in the region, For corresponding to the characteristic flow velocity of the single cell flow channel branch The reynolds number of the fluid in the region.
5. The proxy model-based pile multiscale multi-condition optimizing diagnostic method according to claim 1, wherein the variational property correction kernel in step S2 comprises: Aiming at a high-temperature operation condition, introducing a variable coupling correction formula to calculate the offset between the actual operation pressure drop and the rated reference condition pressure drop: ; In the formula, For the actual operating condition pressure drop, For a nominal reference operating pressure drop, In order to run the average temperature of the product, As a reference to the standard temperature of the sample, As a reference to the standard pressure of the pressure, Is the running average pressure.
6. The proxy model-based pile multiscale multi-condition optimizing diagnosis method according to claim 1, wherein in the step S3, the multi-order interaction operator is specifically configured to: constructing a quadratic regression equation containing interaction influence terms between the number N of single cell stacking sections and single cell level micro geometric variables, wherein the general expression of the performance agent model is as follows: wherein Y is a target evaluation value, Is a constant term which is used to determine the degree of freedom, Is a coefficient of linearity which is a function of the coefficient of linearity, For the coefficients of the quadratic term, For the interaction term coefficient, the method is used for quantifying the sensitivity evolution of the cell flow channel internal resistance requirement of the cell pile scale expansion, 、 The input vectors containing the structural and operational variables, respectively.
7. The proxy model-based pile multiscale multi-condition optimizing diagnosis method according to claim 1, wherein the specific process of designing the optimizing mode in step S4 comprises: In a preset operation interval, calculating the performance fluctuation rate of the candidate structure parameter combination in the variable temperature and variable load process; Selecting the flow distribution uniformity index Unif to be always higher than a preset threshold value and the total pressure drop of the system in the operation interval The structural scheme which is always lower than a preset threshold value is used as an optimal scheme; Wherein, the flow distribution uniformity index Unif is defined as: In the formula (I), in the formula (II), Is the first The mass flow rate of the flow channel of the single cell, The arithmetic average value of the flow channels of the whole pile is adopted, and N is the total number of sections of the pile; Total pressure drop of the system Is defined as: In the formula (I), in the formula (II), For the total pressure drop along the intake manifold, Is the voltage drop of the single cell unit branch, Is the total pressure drop along the outlet manifold.
8. The proxy model-based pile multiscale, operating condition optimizing diagnostic method according to claim 1, wherein in step S4, the performance diagnostic mode output diagnostic report comprises: calculating a flow resistance matching health degree curve of the existing design in a full-working-condition operation interval, and identifying a core contribution factor causing flow distribution mismatch by calculating a partial derivative matrix of each design variable to the flow distribution uniformity index Unif, and positioning a physical bottleneck caused by geometric parameter misalignment or working condition mismatch.
9. The proxy model-based pile multiscale multi-condition optimizing diagnostic method according to claim 1, wherein the step S4 further comprises: And carrying out probability scanning in the space where the operation variable subset is located by utilizing the performance agent model, and outputting an operation dessert region coordinate graph meeting preset performance constraint under the current structural parameters for guiding the operation strategy setting of the electric pile control system.
10. A proxy model-based stack multiscale, optimization diagnostic system for performing the method of any one of claims 1-9, comprising: The full space definition module is used for acquiring and storing definition thresholds of the structure variable subset and the operation variable subset; a dynamic physical calculation kernel which is internally provided with a program based on a flow resistance network method and integrated with a variable property correction and local loss dynamic correction algorithm and is used for generating an original data set; The performance agent model engine is used for storing and running a nonlinear regression equation obtained based on the training of the response surface method; And the decision terminal is used for providing a scene constraint input interface, an optimal parameter set output interface and a performance diagnosis analysis interface so as to execute a design optimizing or performance diagnosis mode.

Description

Proxy model-based pile multi-scale multi-working-condition optimizing diagnosis method and system Technical Field The invention relates to the technical field of fuel cells, in particular to a stack multi-scale multi-working-condition optimizing diagnosis method and system based on a proxy model. Background At present, in the design and development of fuel cells and electrolytic cells, three-dimensional CFD simulation is generally adopted to analyze the flow characteristics inside a pile, or a topology optimization technology is utilized to carry out microscopic-level generation type design on the form of a single cell flow channel. Although the three-dimensional CFD method can obtain a relatively accurate flow field distribution result, the calculation period is up to a plurality of weeks, and the large-scale parameter scanning and multi-objective optimization are difficult to support, the topology optimization technology is mainly focused on the innovation of the flow channel configuration of the single cell scale, the generated complex shape often faces the manufacturing process limitation, and the problem of system-level matching between a macroscopic manifold and a plurality of sections of galvanic piles cannot be solved. However, in the related art, there are the following technical problems: Firstly, the microscopic flow channel design and the macroscopic manifold design are mutually split, the impedance matching relationship between the microscopic flow channel design and the macroscopic manifold design is ignored, so that uneven flow distribution frequently occurs in the actual application of the high-power galvanic pile, secondly, the traditional static structure design fails to fully consider the dynamic influence of the changes of the operating temperature, the pressure, the humidity and the working current on the physical properties of fluid, the stability of the flow field of the galvanic pile in a full-power variable working condition interval is difficult to ensure, thirdly, the flow distribution uniformity and the system pressure drop have a physical mutual exclusion relationship, the optimal balance point under the full working condition is difficult to find in a short time by the traditional method, and fourthly, the built stock galvanic pile is lack of diagnostic tools capable of rapidly identifying the causes of flow resistance mismatch and giving quantitative improvement suggestions. Disclosure of Invention In order to solve at least part of the technical problems in the related art, the invention provides a pile multi-scale multi-working-condition optimizing diagnosis method and system based on a proxy model. In order to achieve the above purpose, the technical scheme adopted by the invention comprises the following steps: according to a first aspect of the invention, a proxy model-based pile multi-scale multi-working condition optimizing diagnosis method is provided, comprising the following steps: Step S1, constructing a parameterized input vector comprising a structural variable subset and an operation variable subset, wherein the structural variable subset comprises macroscopic manifold parameters of a galvanic pile, the number of single cell stacking joints and microscopic flow channel geometric parameters of single cells, and the operation variable subset comprises operating temperature, operating pressure, reaction gas metering ratio, operating current density and inlet relative humidity; Step S2, a flow resistance network model with multiple physical fields dynamically coupled is established, a balance equation set of fluid distribution in a pile is solved based on kirchhoff topology law, and an original data set reflecting a flow distribution uniformity index and a total pressure drop of a system is obtained, wherein in the solving process, a variable property correction kernel which is in real-time linkage with the operation variable subset and a local loss coefficient correction kernel which dynamically changes along with a flow state are introduced; Step S3, training and constructing a nonlinear global performance proxy model based on a response surface method by utilizing the original data set to establish a mapping function which is driven by the structural variable subset and the operation variable subset together and is used for predicting the pile performance, wherein the performance proxy model comprises a multi-order interaction operator used for representing the coupling relation between different scale design variables and between the design variables and the operation variables; S4, executing a design optimizing mode or a performance diagnosis mode based on the performance agent model; In the design optimizing mode, receiving performance constraint of a target application scene in a preset operation interval, and reversely screening out a structural parameter combination with the highest robustness score in the interval; and in the performance diagnosis mode, acq