Fuel cell health diagnosis method and device under plateau'four-low 'condition and medium

Abstract

The invention discloses a fuel cell health diagnosis method and device under plateau'four-low 'conditions and a medium, and relates to the field of proton exchange membrane fuel cells, the method comprises the following steps: based on a nano sensor, collecting key parameters such as voltage, current, environment temperature, humidity, oxygen concentration and the like of a fuel cell in real time; processing various data by using a BIRCH clustering method, a multi-modal fusion method and a GCN model method to obtain dynamic features, plateau environment features and static features; extracting dynamic features to a Hankel matrix field according to the dynamic feature information; according to the plateau environment feature information and the static feature information, the plateau environment features and the static features are extracted to a co-occurrence matrix field; according to the Hankel matrix field and the co-occurrence matrix field, performing feature standardization and tensorization to a three-dimensional space; and according to the feature three-dimensional space, fusing a multi-layer sensor by using a sample entropy method to obtain a plateau region fuel cell diagnosis result. According to the invention, safe operation of the fuel cell system is guaranteed.

Description

Technical Field

[0001] The present application relates to the field of proton exchange membrane fuel cell technology, and in particular to a fuel cell health diagnosis method, device and medium under plateau "four lows" conditions. Background Art

[0002] The application of fuel cells in plateau areas has always been limited due to harsh environmental conditions. In plateau environments, due to the high altitude, the atmospheric pressure is much lower than that in plains, and the flow and diffusion rate of oxygen may be reduced, resulting in unstable hydrogen supply inside the fuel cell, thus affecting the normal operation of the fuel cell. At the same time, the oxygen content is significantly reduced. This low oxygen environment will affect the efficiency of the redox reaction in the fuel cell, thereby affecting the output power and working efficiency of the fuel cell. Secondly, the plateau environment is generally cold, especially at night when the temperature drops more significantly. The start-up and operation performance of fuel cells at low temperatures will be significantly affected, especially the water management problem is more severe, and water condensation is prone to occur, and even ice will form inside the battery, causing battery damage. The air humidity in plateau areas is usually low, and low humidity will accelerate the drying of the electrolyte membrane, thereby causing the stability and performance of the fuel cell to decline. Existing fuel cell monitoring and diagnostic systems are difficult to achieve timely monitoring and diagnosis of the operating status under such special conditions. In summary, the low air pressure, low oxygen concentration, low temperature and low humidity in plateau areas have many negative effects on the normal operation of fuel cells, and put forward higher requirements on the health diagnosis system of fuel cells. However, general fuel cells often find it difficult to complete the tasks of high-precision monitoring and timely diagnosis due to the lack of adaptation to plateau conditions.

[0003] Therefore, proposing a fuel cell health diagnosis method, device and medium under the "four lows" conditions in the plateau is a technical problem that needs to be solved urgently in this field. Summary of the invention

[0004] The purpose of this application is to provide a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau, so as to solve the problem that the plateau environment diagnosis of the fuel cell system is relatively difficult.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In the first aspect, the present application provides a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau, including:

[0007] Obtain different types of key data such as fuel cell voltage and current, ambient temperature and humidity, membrane status, etc.;

[0008] The various types of data are processed using the BIRCH clustering method, the multimodal fusion method, and the GCN model method to obtain dynamic features, plateau environmental features, and static feature information;

[0009] Extracting features into Hankel matrix field and co-occurrence matrix field respectively according to the dynamic features, plateau environment features and static feature information;

[0010] According to the feature field, the features are normalized and tensorized into a three-dimensional space;

[0011] According to the characteristic three-dimensional space, the multi-layer perceptron is fused by using the sample entropy method to obtain the diagnosis result of the fuel cell in the plateau area;

[0012] Optionally, the various types of data are processed using a BIRCH clustering method, a multimodal fusion method, or a GCN model method to obtain dynamic features, plateau environment features, and static feature information, specifically including:

[0013] The data are subjected to BIRCH cluster analysis to capture the relationship between different variables;

[0014] Determining dynamic feature information according to the relationship between the different variables;

[0015] The data are standardized for each mode to obtain transformed data;

[0016] Dynamically adjust the weight of each modality according to the data transformed by the attention fusion mechanism to determine the characteristic information of the plateau environment;

[0017] aggregating the characteristic information of the data node and the information of the neighboring nodes to obtain converted data;

[0018] Calculate the symbiotic array according to the converted data to determine static characteristic information;

[0019] Optionally, according to the dynamic features, extracting features into a Hankel matrix field specifically includes:

[0020] According to the voltage subsequence and the current subsequence, a voltage Hankel matrix and a current Hankel matrix are constructed and mapped to a two-dimensional matrix field Hankel matrix field;

[0021] Optionally, according to the plateau environmental characteristics, extracting the characteristics into a symbiotic array specifically includes:

[0022] Based on the plateau environmental characteristic data of temperature, air pressure, oxygen concentration, and humidity with N time steps, the co-occurrence matrix of a single pair of features and the set of co-occurrence matrices of all feature pairs are constructed;

[0023] Optionally, extracting features into a symbiotic array according to the static features specifically includes:

[0024] According to the type of static feature information, the discrete static features are extracted into the co-occurrence matrix field using the One-Hot encoding method, and the numerical static features are extracted into the co-occurrence matrix field using the standardized method;

[0025] The present invention also provides a fuel cell health diagnosis system under the "four lows" conditions of the plateau, comprising:

[0026] The acquisition module is used to obtain the voltage and current of the fuel cell system, membrane status and other data, as well as the ambient pressure, oxygen concentration, humidity and temperature data of the "four lows" of the plateau environment;

[0027] A conversion module is used to extract features from the data to obtain dynamic features, plateau environment features, and static features, convert them into Hankel matrix fields and co-occurrence matrix fields, and standardize and tensorize them into three-dimensional space;

[0028] Optionally, the conversion module specifically includes:

[0029] A normalization processing unit, used for normalizing the data to obtain converted data;

[0030] A calculation unit, used for calculating a Hankel matrix field and a co-occurrence matrix field according to the converted data;

[0031] A dynamic feature information determination unit, used for determining the dynamic feature information according to the Hankel matrix field;

[0032] A plateau environment characteristic information determination unit, used to determine the plateau environment characteristic information according to the symbiosis matrix field;

[0033] A static feature information determination unit, used to determine the static feature information according to the co-occurrence matrix field;

[0034] A first identification module is used to identify the fuel cell by using a sample entropy method according to the characteristic information to obtain a sample entropy value of the fuel cell;

[0035] The second recognition module is used to use the multi-layer perceptron method to perform recognition based on the feature information to obtain a problem score value of the fuel cell;

[0036] A fusion module is used to fuse the sample entropy value output and the problem score value output to obtain the fuel cell system diagnosis result;

[0037] In the second aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned fuel cell health diagnosis methods, devices, and media under the "four lows" conditions of high altitude.

[0038] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned fuel cell health diagnosis methods, devices and media under the "four lows" conditions in the plateau.

[0039] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0040] The present application discloses a method, device and medium for health diagnosis of fuel cells under the "four lows" conditions of the plateau. The method collects key data such as voltage, current, temperature and humidity, and oxygen concentration of the fuel cell in real time through nanosensors; the data are respectively extracted by BIRCH clustering, multimodal fusion and graph convolution network (GCN) and other methods to obtain dynamic feature information, plateau environmental feature information and static feature information; a Hankel matrix field is constructed according to the dynamic feature information, and a symbiosis matrix field is constructed according to the plateau environmental feature information and the static feature information; according to the Hankel matrix field and the symbiosis matrix field, the features are standardized and tensorized to three-dimensional space; according to the characteristic three-dimensional space, the sample entropy method is used to fuse the multilayer perceptron to obtain the diagnosis results of fuel cells in the plateau area. The present invention comprehensively considers the dynamic, plateau environmental and static characteristics, improves the health diagnosis accuracy of fuel cells under the harsh conditions of the plateau, has the characteristics of real-time, high efficiency and intelligence, and provides comprehensive state monitoring and fault diagnosis capabilities for fuel cell systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 This is a general diagram of a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau according to the present invention;

[0043] Figure 2 This is a flow chart of identification and judgment of the diagnostic process of the present invention;

[0044] Figure 3 The schematic diagram of the Hankel matrix field of the fuel cell current and voltage;

[0045] Figure 4 The symbiosis matrix field map of the plateau environment characteristics of fuel cells;

[0046] Figure 5 A fuel cell health diagnosis method, device and medium flow chart provided by the present invention;

[0047] Figure 6 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0049] The purpose of this application is to provide a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau, aiming to enhance the system's ability to recognize abnormal conditions under harsh plateau environmental conditions of air pressure, low oxygen concentration, low temperature and low humidity, and ensure the accuracy and timeliness of diagnosis.

[0050] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0051] like Figure 5 As shown, the present invention provides a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau, including:

[0052] Step 101: Obtain data such as voltage and current, membrane status, etc. of the fuel cell system and real-time data such as ambient pressure, oxygen concentration, humidity, and temperature of the plateau environment through nanosensors.

[0053] Step 102: According to different types of data such as the voltage and current of the battery system, the ambient temperature and humidity, and the membrane state, the dynamic feature, the plateau environment feature, and the static feature information are obtained using BIRCH cluster analysis, multimodal fusion, and the GCN model.

[0054] Step 103: extracting features into a Hankel matrix field according to the dynamic features, and extracting features into a co-occurrence matrix field according to the plateau environment features and static feature information.

[0055] Step 104: Convert different feature fields into a unified tensor structure, and integrate the tensor into three-dimensional space to form three-dimensional feature data.

[0056] Step 105: According to the characteristic three-dimensional space, a sample entropy method is used to fuse the multi-layer perceptron to obtain the diagnosis result of the fuel cell in the plateau area.

[0057] Step 105 specifically includes:

[0058] According to the three-dimensional feature data, the sample entropy method and the multi-layer perceptron method are used to obtain the sample entropy value output and the problem score value output combination data; according to the output combination data, the fuel cell diagnostic rules are used to judge the current state of the fuel cell to obtain the fuel cell diagnostic result.

[0059] Part 1: Conversion of numerical information into feature information

[0060] In the first step, the BIRCH algorithm is used to use clustering feature triples to represent the summary information of each cluster. The clustering feature triples are defined as follows:

[0061] CF=(N,LS,SS)

[0062] Where N is the number of data points in the current cluster (fuel cell operating state cluster), indicating the number of sampling points in this operating state;

[0063] The multimodal fusion method is used to extract plateau environmental features to cope with the complexity and multidimensional influencing factors of the plateau environment. In order to facilitate fusion, the features of different modes need to be standardized. Assume x (i) The mean and standard deviation are μ (i) and σ (i) , the features of each mode can be standardized:

[0064]

[0065] Introduce the attention fusion mechanism to dynamically adjust the weight of each modality and automatically assign weights to highlight key modal features. Set a learnable weight parameter w (i) Weight each modal feature:

[0066]

[0067] The GCN model is based on graph convolution operations, which aggregates the feature information of a node with the information of neighboring nodes. The forward propagation process of each layer of GCN is as follows:

[0068]

[0069] Among them, H (l) is the node feature matrix of the lth layer, and the initial layer H (0) is the input representation of static features, is the adjacency matrix with self-connection, I is the identity matrix, for The degree matrix of W (l) is the weight matrix of the lth layer, which needs to be learned through training, and σ is the ReLU nonlinear activation function.

[0070] Part 2: Extracting feature information into feature fields

[0071] The dynamic characteristic data of the fuel cell (such as voltage and current) are divided into multiple clusters after BIRCH clustering, and each cluster represents a dynamic behavior mode. Voltage time series: {V1, V2, …, V N}, current time series: {I1,I2,…,I N Through BIRCH clustering, these time series are divided into L clusters, each of which contains one or more subsequences. For example, the dynamic characteristics of the lth cluster are: voltage series: {V l,1 ,V l,2 ,…,V l,Nl}, current sequence: {I l,1 ,I l,2 ,…,I l,Nl}. Extract the voltage and current subsequences and construct the Hankel matrix H Vl and H Il :

[0072]

[0073] like Figure 3 As shown, the voltage and current Hankel matrices of each cluster are mapped into a two-dimensional matrix field to form a Hankel matrix field H:

[0074] H={(H V1 ,H I1 ),(H V2 ,H I2 ),…,(H VL ,H IL )}

[0075] With the help of dimensionality reduction and pattern analysis of the co-occurrence matrix, the system can process multi-dimensional feature data more efficiently. i ,X j ), the element C of its co-occurrence matrix p Indicates that the two features fall within the interval R p and R q Co-occurrence count:

[0076]

[0077] Among them, R p and R q It is feature X i and X j The range of values ​​of δ(x k ∈R p ,y k ∈R q ) is the indicator function:

[0078]

[0079] There are N time steps of plateau environmental characteristic data, including: temperature T = {T1, T2, ..., T N}, air pressure P = {P1, P2, ..., P N}, oxygen concentration O={O1,O2,…,O N}, humidity H={H1,H2,…,H N}. For two features (such as temperature and air pressure), the co-occurrence matrix C is constructed ij Discretize the two features into intervals R p and R q . Traverse each time step k=1,2,...,N and count the number of times the two features co-occur in the corresponding interval. For example, Figure 4 , for the characteristic pair (temperature T and pressure P), the co-occurrence matrix elements are:

[0080]

[0081] For four features T, P, O, H, construct the co-occurrence matrix set of all feature pairs:

[0082] C={C TP ,C TO ,C TH ,C PO ,C PH ,C OH}

[0083] For discrete static features (such as material type), One-Hot encoding is used to convert the category features into continuous vector form and extract them into the co-occurrence matrix field; for numerical static features, standardization is used to ensure that different features are in the same numerical range and are extracted into the co-occurrence matrix field.

[0084] Part 3: Feature information tensorization to three-dimensional space

[0085] Each coordinate axis of the three-dimensional space model represents a different feature type, namely: X-axis for dynamic features, Y-axis for plateau environmental features, and Z-axis for static features. This three-dimensional space will provide a reference for the system to comprehensively analyze the status of fuel cells in multiple dimensions such as time, environment, and design parameters. When constructing the three-dimensional space, first convert different feature data into a unified tensor structure, and then splice the dynamic features, plateau environmental features, and static features according to the feature dimensions to form a comprehensive tensor:

[0086]

[0087] Among them, N is the number of samples, T is the time step, and F is the feature dimension.

[0088] Part 4: Fuel cell health diagnosis based on sample entropy method and multilayer perceptron method

[0089] Sample entropy is used to measure the complexity of the system's dynamic characteristics and evaluate the stability of fuel cell operation:

[0090] (1) Set the embedding dimension m and tolerance r:

[0091] m: Used to define the size of the pattern within the time window.

[0092] r: Set the allowable error range to control the strictness of pattern matching, usually set to 0.1-0.2 times the standard deviation of the data:

[0093] r=k×σ(X)

[0094] Where σ(X) is the standard deviation of the time series and k is a scaling factor (such as 0.1 or 0.2)

[0095] (2) Constructing embedding vector:

[0096] Construct a sequence of embedding vectors of length m from the time series X:

[0097] X i =[x i ,x i+1 ,…,x i+m-1 ]for1≤i≤N-m+1

[0098] (3) Calculate the sample entropy SampEn:

[0099]

[0100] Where A is the number of matched pattern logarithms of length m+1, and B is the number of matched pattern logarithms of length m. When the sample entropy is low, the system runs stably and the pattern repeatability is high; when the sample entropy is high, the system has more fluctuations and the state is more complex or abnormal. Based on the sample entropy calculation results, a preliminary judgment of the abnormal state can be made.

[0101] The multi-layer perceptron method extracts nonlinear relationships between complex dynamic-static features and environmental features based on multi-layer neurons:

[0102] The network layer structure is divided into input layer, hidden layer and output layer. The input layer receives the three-dimensional feature tensor T combined After flattening After the input, if the input tensor is [N, T, F], it will be [N, T×F] after flattening; the ReLU function is often used as the activation function of the hidden layer:

[0103] a (l) =ReLU(z (l) )

[0104] The hidden layer structure is designed as follows:

[0105] Hidden layer 1: 64 neurons, using ReLU activation function.

[0106] Hidden layer 2: 32 neurons, using ReLU activation function.

[0107] Hidden layer 3: 16 neurons, using ReLU activation function.

[0108] During the forward propagation process, data is propagated through the network layer by layer until it reaches the output layer:

[0109] (1) Neuron output of layer l:

[0110] z (l) =W (l) a (l-1) +b (l)

[0111] Among them, W (l) is the weight matrix of the lth layer, b (l) The bias of the lth layer, a (l-1) is the activation value (output) of the previous layer.

[0112] (2) Assume that the output layer of the multi-layer sensor model is a vector Z = [z1, z2, ..., z n ], where n is the number of categories (e.g. “normal”, “slightly abnormal”, “severely abnormal”), and the output layer uses the Softmax activation function:

[0113]

[0114] Among them, P i is the probability score of the ith category, z i is the i-th logit (inactive fraction) of the output layer of the multilayer perceptron model, and n is the total number of categories.

[0115] (3) Output layer results:

[0116]

[0117] Output the probabilities of each category to complete the classification task, and output the question scores to complete the regression task.

[0118] In back propagation, the loss function is selected according to the task type:

[0119] (1) Classification task: Cross entropy loss function:

[0120]

[0121] (2) Regression task: Mean Square Error (MSE):

[0122]

[0123] The gradient of loss to weight is calculated through back propagation, the Adam optimizer is used to optimize the algorithm, the learning rate is dynamically adjusted to increase the convergence speed, multimodal data and sample entropy values ​​are used as input features, forward propagation and back propagation training are performed, the error is calculated according to the loss function, and the model parameters and model weights are updated using the gradient descent method:

[0124]

[0125] where η is the learning rate.

[0126] The sample entropy result is combined with other features as the input vector and input into the multilayer perceptron model. After the multilayer perceptron model is inferred, it can be known whether the output state is abnormal and the problem score is given. If the system state is abnormal or the problem score exceeds the set threshold, a warning is triggered. Figure 2 First, the feature data is processed by the sample entropy method and input into the multilayer perceptron model for state inference. According to the different ranges of sample entropy value (SampEn) and problem score (P), namely SampEn≥1.0 or P≥0.7, 0.5≤SampEn<1.0 or 0.3≤P<0.7, SampEn<0.5 and P<0.3, the system triggers a serious abnormality warning, a minor abnormality warning, and judges that there is no abnormality. In the case of serious abnormalities, the system sends an emergency notification, automatically generates a fault report, and reminds emergency troubleshooting; in the case of minor abnormalities, the system sends an information notification to remind the monitoring of abnormal status; if there is no abnormality, the system status is recorded and continuously monitored. In addition, the process also includes a data correction and verification mechanism to ensure data accuracy, and the model can be continuously optimized through feedback learning to improve the accuracy of diagnosis and system responsiveness.

[0127] The present invention also provides a fuel cell health diagnosis system under the "four lows" conditions of the plateau, comprising:

[0128] The acquisition module is used to obtain the voltage and current of the fuel cell system, membrane status and other data, as well as the ambient pressure, oxygen concentration, humidity and temperature data of the "four lows" of the plateau environment;

[0129] A conversion module is used to extract features from the data to obtain dynamic features, plateau environment features, and static features, convert them into Hankel matrix fields and co-occurrence matrix fields, and standardize and tensorize them into three-dimensional space;

[0130] A first identification module is used to identify the fuel cell by using a sample entropy method according to the characteristic information to obtain a sample entropy value of the fuel cell;

[0131] The second recognition module is used to use the multi-layer perceptron method to perform recognition based on the feature information to obtain a problem score value of the fuel cell;

[0132] A fusion module is used to fuse the sample entropy value output and the problem score value output to obtain the fuel cell system diagnosis result;

[0133] In an exemplary embodiment, a computer device is provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program to implement a fuel cell health diagnosis method, device, and medium under plateau "four lows" conditions.

[0134] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a fuel cell health diagnosis method, device and medium under high-altitude "four lows" conditions are implemented.

[0135] In an exemplary embodiment, a computer program product is provided, including a computer program, which, when executed by a processor, implements a fuel cell health diagnosis method, device and medium under high-altitude "four lows" conditions.

[0136] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 6 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau are realized.

[0137] Those skilled in the art will understand that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0138] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0139] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0140] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0141] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0142] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau, characterized in that: include: Obtain different types of key data such as fuel cell voltage and current, ambient temperature and humidity, membrane status, etc.; The various types of data are processed using the BIRCH clustering method, the multimodal fusion method, and the GCN model method to obtain dynamic features, plateau environmental features, and static feature information; Extracting features into Hankel matrix field and co-occurrence matrix field respectively according to the dynamic features, plateau environment features and static feature information; According to the feature field, the features are normalized and tensorized into a three-dimensional space; According to the characteristic three-dimensional space, the multi-layer perceptron is fused by using the sample entropy method to obtain the diagnosis result of the fuel cell in the plateau area; According to the information of each feature in the three-dimensional space, the multilayer perceptron method is combined with the sample entropy method to diagnose the information, and the diagnostic results shown by the sample entropy value (SampEn) and the problem score (P) are obtained. SampEn is the complexity evaluation of the fuel cell system output by the sample entropy method, and P is the score result of whether the fuel cell has abnormalities output by the multilayer perceptron method. The first step is to calculate the sample entropy of the fuel cell system: r=k×σ(X) X i =[x i ,x i+1 ,...,x i+m-1 ]for 1≤i≤N-m+1 Among them, m is the size of the pattern in the defined time window, r is the allowed error range, which controls the strictness of pattern matching and is usually set to 0.1-0.2 times the standard deviation of the data, σ(X) is the standard deviation of the time series, k is the scaling factor (such as 0.1 or 0.2), A is the number of matched pattern logarithms of length m+1, and B is the number of matched pattern logarithms of length m. The second step is to calculate the problem score of the fuel cell system: and (l) =ReLU(from (l) ) z (l) =W (l) a (l-1) +b (l) Among them, the ReLU function is used as the activation function, W (l) is the weight matrix of the lth layer, b (l) The bias of the lth layer, a (l-1) is the activation value (output) of the previous layer, P i is the probability score of the ith category, z i is the i-th logit (inactive fraction) of the output layer of the multilayer perceptron model, and n is the total number of categories. The fuel cell diagnosis results based on the sample entropy method and multi-layer perceptron can be obtained:

2. According to the method, device and medium for fuel cell health diagnosis under the "four lows" conditions of plateau in claim 1, it is characterized in that: The various types of data are processed using the BIRCH clustering method, the multimodal fusion method, and the GCN model method to obtain dynamic features, plateau environmental features, and static feature information, specifically including: The data are subjected to BIRCH cluster analysis to capture the relationship between different variables; Determining dynamic feature information according to the relationship between the different variables; The data are standardized for each mode to obtain transformed data; Dynamically adjust the weight of each modality according to the data transformed by the attention fusion mechanism to determine the characteristic information of the plateau environment; aggregating the characteristic information of the data node and the information of the neighboring nodes to obtain converted data; The symbiotic array is calculated based on the converted data to determine static feature information.

3. According to the method, device and medium for fuel cell health diagnosis under the "four lows" conditions of plateau in claim 1, it is characterized in that: According to the dynamic features, extracting features into a Hankel matrix field specifically includes: According to the voltage subsequence {V l,1 ,V l,2 ,…,V l,Nl } and the current subsequence {I l,1 ,I l,2 ,…,I l,Nl }, construct the Hankel matrix H Vl and H Il : According to the Hankel matrix H of voltage and current Vl and H Il , mapped to the two-dimensional Hankel matrix field H: H={(H V1 ,H I1 ),(H V2 ,H I2 ),…,(H VL ,H IL )} Optionally, based on the Hankel matrix field, for each Hankel matrix H l Perform singular value decomposition (SVD) to extract principal components and analyze the timing patterns of dynamic features, including: Among them, Σ Vl and Σ Il It is a singular value matrix, which contains the main change mode of the matrix. The maximum singular value can reflect the main dynamic behavior of the voltage or current characteristics within the cluster.

4. According to claim 1, a fuel cell health diagnosis method, device and medium under the "four lows" conditions of the plateau, characterized in that: According to the plateau environmental characteristics, the characteristics are extracted into the symbiotic array, specifically including: There are N time steps of temperature T = {T1, T2, ..., T N }, air pressure P = {P1, P2, ..., P N }, oxygen concentration O={O1,O2,…,O N }, humidity H={H1,H2,…,H N }The plateau environmental feature data is composed of a single pair of feature co-occurrence matrices and a set of co-occurrence matrices of all feature pairs, for example: C={C TP ,C TO ,C TH ,C PO ,C PH ,C OH } in, is the co-occurrence matrix of temperature and air pressure, and C is the co-occurrence matrix of four characteristic temperature T, pressure P, oxygen concentration O, and humidity H.

5. According to the method, device and medium for fuel cell health diagnosis under the "four lows" conditions of plateau in claim 1, it is characterized in that: According to the static features, extracting features into the symbiotic array specifically includes: According to the type of static feature information, the One-Hot encoding method is used to extract discrete static features into the co-occurrence matrix field, and the standardized method is used to extract numerical static features into the co-occurrence matrix field.

6. A fuel cell health diagnosis method, device and medium under plateau "four lows" conditions according to claim 1, characterized in that: According to the feature field, the features are normalized and tensorized into three-dimensional space: T combined =[T dynamic ,T environment ,T static ] Among them, x (i) The mean and standard deviation are μ (i) and σ (i) , T combined is a three-dimensional tensor, T dynamic , T environment , T static They are dynamic characteristics, plateau environment characteristics and static characteristics respectively.

7. A fuel cell health diagnosis method, device and medium under plateau "four lows" conditions according to claim 1, characterized in that: According to the characteristic three-dimensional space, the sample entropy method is used to fuse the multi-layer perceptron to obtain the diagnosis results of fuel cells in plateau areas: According to the information of each feature in the three-dimensional space, the multilayer perceptron method is combined with the sample entropy method to diagnose the information, and the diagnostic results shown in the sample entropy value (SampEn) and the problem score (P) are obtained. SampEn is the complexity of the fuel cell system output by the sample entropy method, and P is the score result of whether the fuel cell has an abnormality output by the multilayer perceptron method. The first step is to calculate the sample entropy of the fuel cell system: r=k×σ(X) X i =[x i ,x i+1 ,...,x i+m-1 ]for 1≤i≤N-m+1 Among them, m is the size of the pattern in the defined time window, r is the allowed error range, which controls the strictness of pattern matching and is usually set to 0.1-0.2 times the standard deviation of the data, σ(X) is the standard deviation of the time series, k is the scaling factor (such as 0.1 or 0.2), A is the number of matched pattern logarithms of length m+1, and B is the number of matched pattern logarithms of length m. The second step is to calculate the problem score of the fuel cell system based on the neural network: and (l) =ReLU(from (l) ) z (l) =W (l) a (l-1) +b (l) Among them, the ReLU function is used as the activation function, W (l) is the weight matrix of the lth layer, b (l) The bias of the lth layer, a (l-1) is the activation value (output) of the previous layer, P i is the probability score of the ith category, z i is the i-th logit (inactive fraction) of the output layer of the multilayer perceptron model, and n is the total number of categories. The fuel cell diagnosis results based on the sample entropy method and multi-layer perceptron can be obtained: Based on the sample entropy value (SampEn) and the problem score (P), SampEn<0.5, P<0.3 determines that the fuel cell state is normal; based on the sample entropy value (SampEn) and the problem score (P), 0.5≤SampEn<1.0, 0.3≤P<0.7 determines that the fuel cell state is slightly abnormal; based on the sample entropy value (SampEn) and the problem score (P), SampEn≥1.0, P≥1.0 determines that the fuel cell state is seriously abnormal; Determine whether the sample entropy value (SampEn) and the problem score (P) belong to the first condition, the first condition is SampEn≥1.0, P≥1.0, determine it as a serious abnormality, generate a fault report, and remind emergency investigation; Determine whether the sample entropy value (SampEn) and the problem score (P) belong to the second condition, the second condition is 0.5≤SampEn<1.0, 0.3≤P<0.7, generate a system information notification, and remind to monitor the abnormal status; Determine whether the sample entropy value (SampEn) and the problem score (P) belong to the third condition, the third condition is SampEn<0.5, P<0.3, record the system status and continuously monitor.

8. A fuel cell health diagnosis system under the "four lows" conditions of the plateau, characterized in that: include: The acquisition module is used to obtain the voltage and current of the fuel cell system, membrane status and other data, as well as the "four lows" of the plateau environment, including ambient pressure, oxygen concentration, humidity and temperature; A conversion module is used to extract features from the data to obtain dynamic features, plateau environment features, and static features, convert them into Hankel matrix fields and co-occurrence matrix fields, and standardize and tensorize them into three-dimensional space; A first identification module is used to identify the fuel cell by using a sample entropy method according to the characteristic information to obtain a sample entropy value of the fuel cell; The second recognition module is used to identify the problem of the fuel cell by using a multi-layer perceptron method according to the feature information to obtain a problem score value of the fuel cell; The fusion module is used to fuse the sample entropy value output and the problem score value output to obtain the fuel cell system diagnosis result.

9. A fuel cell health diagnosis method, device and medium under plateau "four lows" conditions according to claim 1, characterized in that: According to the task type, select the loss function in back propagation and update the model parameters based on the gradient descent method, including: in, The cross entropy loss function selected for the classification task, The mean square error function selected for the regression task, W (l) is the weight matrix in the l-th layer of the neural network, η is the learning rate, is the loss function, is the gradient of the loss function with respect to the weight.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement a fuel cell health diagnosis method, device, and medium under high-altitude "four lows" conditions as described in any one of claims 1-9.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements a fuel cell health diagnosis method, device and medium under high-altitude "four lows" conditions as described in any one of claims 1-9.

Images