Method, device, vehicle and equipment for determining water content of fuel cell stack

Abstract

The present invention relates to a method, device, vehicle, and equipment for determining the water content of a fuel cell stack, and relates to the field of vehicle technology. The method is intended to determine the water content of a fuel cell stack in a vehicle. The method comprises: determining the performance characteristics of the fuel cell based on the operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell; reconstructing the performance characteristics to obtain a reconstructed characteristic; fusing the performance characteristics with the reconstructed characteristic to obtain a fused characteristic; and determining the water content of the fuel cell stack using the fused characteristic. In this way, the water content of the fuel cell stack can be accurately determined, and further, whether a water fault has occurred in the fuel cell can be accurately determined, so that if a water fault occurs in the fuel cell, it can be quickly repaired, ensuring the normal operation of the fuel cell and increasing the service life of the fuel cell.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a method, device, vehicle and equipment for determining the water content of a fuel cell stack. Background Art

[0002] Proton exchange membrane fuel cell is a power generation device that directly converts chemical energy into electrical energy through electrochemical reactions. It has the advantages of zero emissions, high efficiency, low noise, and high power density, and has shown broad application prospects in the field of new energy vehicles.

[0003] The conductivity of the fuel cell depends on the hydration of the proton exchange membrane. If the proton exchange membrane is too dry, its proton conductivity will drop significantly, which will affect the output power and efficiency of the fuel cell. If the humidity of the proton exchange membrane is too high, water will accumulate inside the battery, forming liquid water, affecting the gas transmission of the gas diffusion layer, and may even cause blockage of the reaction zone, affecting the service life of the fuel cell.

[0004] In a related technology, the electrochemical impedance spectroscopy (EIS) of a fuel cell stack under different water content states is measured, and the collected stack EIS data is preprocessed and the EIS curve features are extracted. The extracted curve features are clustered using fuzzy C-means to obtain curve feature clusters under different water content states. The water content state of the fuel cell stack is classified based on boundary constraints and the Euclidean distance between the data to be classified and the centers of different feature clusters. However, this existing technology classifies the dry and wet states of the fuel cell through the preprocessed EIS curve features, and cannot perform fault judgment on the real-time state of the fuel cell stack well. It has certain limitations and is not very accurate.

[0005] In one existing technology, a fuel cell equivalent circuit model is established based on the fault type of the proton exchange membrane fuel cell, the correspondence between the equivalent circuit model parameters and different fault types is observed, and the fault characteristics of different fault types are determined based on the correspondence between the different fault types. Fault characteristics are selected to characterize the health state of the fuel cell, and fault characteristics of different fault states are extracted, and then the fault characteristics are used as feature inputs of the clustering algorithm. The input fault characteristic data is clustered using a clustering algorithm to obtain a sample data set. The probabilistic neural network model is trained and tested using the sample data set, and the trained probabilistic neural network algorithm is used to implement multi-fault classification of the fault samples, effectively eliminating singular data and improving the accuracy of model classification. However, this method has high requirements for the parameter integrity of the fault characteristics and empirical knowledge of the circuit model, and it is difficult to achieve dynamic analysis of the fuel cell stack state.

[0006] Therefore, how to accurately judge the health status of the fuel cell to ensure the output power and efficiency of the fuel cell and improve the service life of the fuel cell has become a technical problem that needs to be solved urgently. Summary of the Invention

[0007] This application provides a method, device, vehicle, and equipment for determining the water content of a fuel cell stack. By determining the water content of the fuel cell stack, it is possible to accurately identify whether a water fault has occurred in the fuel cell. When a water fault occurs in the fuel cell, water fault treatment can be performed to ensure the output power and efficiency of the fuel cell and increase the service life of the fuel cell. The technical solution of this application is as follows:

[0008] According to the first aspect provided by the present application, a method for determining the water content state of a fuel cell stack is provided, the method comprising: determining the performance characteristics of the fuel cell based on the operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell; reconstructing the performance characteristics to obtain a reconstructed characteristic; fusing the performance characteristics and the reconstruction characteristics to obtain a fused characteristic; and determining the water content state of the fuel cell stack using the fused characteristic.

[0009] According to the above technical means, the present application extracts features of the operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell to obtain the performance characteristics of the fuel cell to accurately represent the current status of the fuel cell; the performance characteristics are then reconstructed to obtain reconstructed features, and the feature reconstruction captures the information in the data more accurately by generally combining or converting the original features; the fusion features of the performance characteristics and the reconstructed features are then used to accurately determine the water content of the fuel cell stack, thereby accurately identifying whether a water fault has occurred in the fuel cell, so that water fault processing can be performed when a water fault occurs in the fuel cell to ensure the output power and efficiency of the fuel cell and improve the service life of the fuel cell.

[0010] In one possible approach, the performance characteristics of the fuel cell are determined based on the operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell, including: representing the operating information, impedance and high-frequency resistance as a first vector; and performing feature extraction on the first vector representation to obtain performance characteristics.

[0011] Based on the above technical means, this application represents the fuel cell stack's operating information, as well as the fuel cell's impedance and high-frequency resistance, in the form of vectors. Feature extraction is then performed on these vectors to obtain performance characteristics. Fuel cell performance characteristics are generated using multiple fuel cell operating parameters and accurately reflect the fuel cell's current state. Based on this, these performance characteristics can be used to accurately predict the water content of the fuel cell stack.

[0012] In one possible approach, reconstructing the performance characteristics to obtain reconstructed features includes determining the reconstructed features based on the performance characteristics and a feature reconstruction model, wherein the feature reconstruction model is trained using a cross entropy loss between the fuel cell performance characteristics and the reconstructed features as a loss function.

[0013] Based on the above technical means, this application utilizes a pre-built feature reconstruction model to determine the reconstructed features corresponding to the fuel cell's performance characteristics. This feature reconstruction model is trained using the cross-entropy loss between the fuel cell's performance characteristics and the reconstructed features as a loss function. Therefore, the feature reconstruction model ensures that the reconstructed features are not distorted and that the reconstructed features contain information that is distinct from the performance characteristics. Subsequently, the performance characteristics and the reconstructed features are fused to obtain a fused feature, which enriches the fuel cell's characteristic information. The fused feature can be used to accurately predict the water content of the fuel cell stack.

[0014] In one possible approach, the feature reconstruction model includes an encoder and a decoder, and the reconstructed features are determined based on the performance characteristics and the feature reconstruction model, including: inputting the performance characteristics into the encoder for dimensionality reduction encoding to obtain dimensionality reduction features; and inputting the dimensionality reduction features into the decoder for decoding to obtain reconstructed features.

[0015] Based on the above technical means, this application extracts features by encoding the performance characteristics through dimensionality reduction, and then decodes the extracted features to restore the dimensionality. This cleans the noise and redundant data in the performance characteristics to obtain characteristic information that is more representative of the fuel cell, thereby improving the accuracy of subsequent predictions of the fuel cell's water content.

[0016] In one possible approach, fused features are used to determine the water status of a fuel cell stack, including: determining the water category probability distribution vector of the fuel cell based on the fused features and a state prediction model; determining the water status based on the water category probability distribution vector; wherein the state prediction model is trained using the cross entropy loss between the true value and the predicted value of the water category probability distribution vector corresponding to the fused features of the fuel cell as a loss function.

[0017] According to the above technical means, the present application pre-constructs a state prediction model with the fusion characteristics of the fuel cell as input and the water category probability distribution vector of the fuel cell as output. Since the state prediction model is trained with the cross entropy loss between the true value and the predicted value of the water category probability distribution vector corresponding to the fusion characteristics of the fuel cell as the loss function, the water category probability distribution vector of the fuel cell can be accurately predicted, thereby accurately identifying whether the fuel cell has a water failure based on the water category probability distribution vector.

[0018] In one possible approach, the state prediction model includes a self-attention mechanism layer and a state prediction layer; based on the fused features and the state prediction model, the water category probability distribution vector of the fuel cell is determined, including: inputting the fused features into the self-attention mechanism layer for self-attention learning to obtain intermediate features; inputting the intermediate features into the state prediction layer to obtain the water category probability distribution vector output by the state prediction layer.

[0019] Based on the above technical means, this application uses a self-attention mechanism layer to perform self-attention learning on the fused features to increase the focus on key information in the fused features. Using the features after self-attention learning to predict the water content of the fuel cell can improve the prediction accuracy.

[0020] In one possible approach, the feature reconstruction model and the state prediction model are trained collaboratively.

[0021] In one possible approach, the operating information includes: water inlet temperature, cathode flow, output power, and output voltage.

[0022] In one possible approach, the water-containing category probability distribution vector is a vector representing each water-containing category and its corresponding probability of occurrence; wherein the water-containing categories include a first water-deficient category, a second water-deficient category, a water-normal category, a first flooded category, and a second flooded category with increasing water content.

[0023] According to the above technical means, the present application identifies a fuel cell with an excessive membrane dryness fault if the water content falls into the first water shortage category; a mild membrane dryness fault if the water content falls into the second water shortage category; a mild flooding fault if the water content falls into the first flooding category; and a severe flooding fault if the water content falls into the second flooding category. This method accurately identifies whether a fuel cell has a water fault, facilitating subsequent water fault handling.

[0024] According to a second aspect of the present application, a device for determining the water content state of a fuel cell stack is provided, the device comprising:

[0025] a characteristic determination unit, configured to determine a performance characteristic of the fuel cell based on operating information of the fuel cell stack and impedance and high-frequency resistance of the fuel cell;

[0026] A feature reconstruction unit, used for reconstructing the performance feature to obtain a reconstructed feature;

[0027] A feature fusion unit is used to fuse the performance features and the reconstruction features to obtain fused features;

[0028] The determining unit is used to determine the water status of the fuel cell stack by utilizing the fusion feature.

[0029] In one possible manner, the feature determination unit is configured to: represent the operating information, impedance, and high-frequency resistance as a first vector; and perform feature extraction on the first vector representation to obtain performance features.

[0030] In one possible approach, the feature reconstruction unit is configured to determine a reconstruction feature based on the performance feature and a feature reconstruction model, wherein the feature reconstruction model is trained using a cross entropy loss between the fuel cell's performance feature and the reconstruction feature as a loss function.

[0031] In one possible approach, the feature reconstruction unit is configured to: input the performance feature into an encoder for dimensionality reduction encoding to obtain a reduced-dimensionality feature; and input the reduced-dimensionality feature into a decoder for decoding to obtain a reconstructed feature.

[0032] In one possible embodiment, the determination unit is used to: determine the water category probability distribution vector of the fuel cell based on the fusion feature and the state prediction model; determine the water state based on the water category probability distribution vector; wherein the state prediction model is trained using the cross entropy loss between the true value and the predicted value of the water category probability distribution vector corresponding to the fusion feature of the fuel cell as the loss function.

[0033] In one possible approach, a unit is determined to: input the fused features into the self-attention mechanism layer for self-attention learning to obtain intermediate features; input the intermediate features into the state prediction layer to obtain the water category probability distribution vector output by the state prediction layer.

[0034] In one possible approach, the feature reconstruction model and the state prediction model are trained collaboratively.

[0035] In one possible approach, the operating information includes: water inlet temperature, cathode flow, output power, and output voltage.

[0036] In one possible approach, the water-containing category probability distribution vector is a vector representing each water-containing category and its corresponding probability of occurrence; wherein the water-containing categories include a first water-deficient category, a second water-deficient category, a water-normal category, a first flooded category, and a second flooded category with increasing water content.

[0037] According to the third aspect provided by the present application, a vehicle is provided, which includes a fuel cell; the fuel cell adopts the method for determining the water content of the fuel cell stack as described in the first aspect and any possible embodiment thereof to determine the water content; the water content is used to indicate whether a water failure occurs in the fuel cell stack; the water failure includes a water shortage failure and a flooding failure.

[0038] According to the fourth aspect provided by the present application, an electronic device is provided, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute instructions to implement a method for determining the water status of a fuel cell stack as described in the first aspect and any possible embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present application, and together with the specification are used to explain the principles of the present application, and do not constitute an improper limitation on the present application.

[0040] Figure 1 This is a flow chart showing a method for determining the water content state of a fuel cell stack according to an exemplary embodiment;

[0041] Figure 2 is a structural diagram of a feature reconstruction model according to an exemplary embodiment;

[0042] Figure 3 is a structural diagram of a state prediction model according to an exemplary embodiment;

[0043] Figure 4 is a flow chart showing another method for determining the water content state of a fuel cell stack according to an exemplary embodiment;

[0044] Figure 5 is a flow chart showing another method for determining the water content state of a fuel cell stack according to an exemplary embodiment;

[0045] Figure 6 It is a structural diagram of a device for determining the water status of a fuel cell stack according to an exemplary embodiment. DETAILED DESCRIPTION

[0046] In order to enable ordinary people in the art to better understand the technical solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0047] It should be noted that the terms "first," "second," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments of the application described herein can be implemented in an order other than those illustrated or described herein. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.

[0048] In the embodiments of this application, words such as "exemplary," "for example," or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary," "for example," or "for example" in the embodiments of this application should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary," "for example," or "for example" is intended to present the relevant concepts in a concrete manner.

[0049] First, the relevant technologies involved in this application are explained to facilitate understanding by those skilled in the art.

[0050] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments.

[0051] Embodiments of the present application provide a method for determining the water content of a fuel cell stack, which can be applied in a vehicle. A vehicle may also be referred to as a transportation vehicle, mobile carrier, electric vehicle (EV), hybrid electric vehicle (HEV), plug-in hybrid electric vehicle (PHEV), fuel cell vehicle (FCV), autonomous vehicle, intelligent and connected vehicle (ICV), driverless vehicle, etc.

[0052] In the embodiments of this application, the vehicle may be a sedan, a sport utility vehicle (SUV), a truck, an electric vehicle, a motorcycle, a tricycle, a special vehicle (such as an ambulance, fire truck, or police car), a driverless taxi, an intelligent connected bus, an autonomous logistics vehicle, an electric truck, etc. Furthermore, this method is also applicable to various specialized vehicles, such as agricultural vehicles, mining vehicles, forestry vehicles, airport vehicles, and port vehicles. This application does not impose any specific restrictions on this.

[0053] like Figure 1 As shown, the present application provides a method for determining the water status of a fuel cell stack, specifically including S101-S104.

[0054] S101 : Determine performance characteristics of the fuel cell based on operating information of the fuel cell stack and impedance and high-frequency resistance of the fuel cell.

[0055] A fuel cell refers to a proton exchange membrane fuel cell (PEMFC), a type of fuel cell that uses a proton exchange membrane as a proton conducting medium. A fuel cell stack is the core power generation unit of a fuel cell system, consisting of multiple single cells stacked in series. Operating information for a fuel cell stack includes the water inlet temperature, cathode flow rate, power, and voltage. The high-frequency resistance of a fuel cell is a key parameter measured by electrochemical impedance spectroscopy, which directly reflects the humidity and conductivity of the fuel cell's proton exchange membrane.

[0056] The stack water inlet temperature refers to the temperature of the coolant entering the fuel cell stack. The stack cathode flow rate refers to the flow rate of the oxidant gas supplied to the cathode side of the stack.

[0057] In one possible implementation, an integrated temperature sensor is used to measure the water inlet temperature of the fuel cell stack in real time; an air flow meter sensor is used to measure the cathode flow of the fuel cell stack; a current sensor and a voltage sensor are used to measure the power and voltage of the fuel cell stack; and an electrochemical impedance spectroscopy is performed on the fuel cell to measure the impedance of the fuel cell and the high-frequency resistance of the fuel cell stack.

[0058] In another possible implementation, the operating information of the fuel cell stack and each type of data in the impedance and high-frequency resistance of the fuel cell are used as a dimension in the first vector to obtain the first vector, and feature extraction is performed on the first vector representation to obtain performance characteristics.

[0059] S102: Reconstruct the performance characteristics to obtain reconstructed characteristics.

[0060] Among them, the above-mentioned reconstruction of the performance characteristics to obtain the reconstructed characteristics includes inputting the performance characteristics into the feature reconstruction model to obtain the reconstructed characteristics output by the feature reconstruction model. Each dimension of the reconstructed characteristics is compared with the dimension in the performance characteristics, which can more accurately represent the characteristics of the fuel cell stack.

[0061] Among them, the feature reconstruction model refers to a neural network model based on unsupervised learning, whose core goal is to learn the potential features of the input data through the encoding-decoding process and make the output as close to the original input as possible. Figure 2 As shown, the present application provides a structural diagram of a feature reconstruction model, which includes: an input layer 201, an encoding layer 202, a latent space layer 203, a decoding layer 204, and an output layer 205.

[0062] 1-1. Input layer 201.

[0063] The input layer 201 is used to input performance characteristics;

[0064] 1-2. Coding layer 202.

[0065] The encoding layer 202 includes one or more hidden layers, which are used to gradually compress the feature dimensions of the performance feature to obtain a first low-dimensional feature.

[0066] Among them, the encoding layer is used to compress the input data into a low-dimensional latent space representation, that is, to reduce the dimension of the input data and map it to the low-dimensional latent space.

[0067] Among them, the dimensionality reduction feature refers to the dimensionality reduction feature obtained after the performance feature is coded through dimensionality reduction.

[0068] In one possible implementation, the encoding maps the performance characteristics from the input layer to the low-dimensional hidden layer through an encoding formula that satisfies the following formula 1:

[0069] Formula 1.

[0070] in, Represents the hidden layer The dimensionality reduction features of the output of the reconstructed neurons, The input layer neurons, represents the number of neurons in the input layer, Represents the first neurons to the hidden layer The weight of a neuron, Indicates bias, Represents the activation function.

[0071] in, Used to define the mapping relationship between performance feature dimensionality reduction and dimensionality reduction features, Data bias for dynamic calibration of dimension reduction features.

[0072] It's important to note that weights represent the strength of connections between neurons. In a neural network, each input is multiplied by a corresponding weight to measure its contribution to the output. For example, in an image recognition task, the larger the weight of a pixel, the more significant its impact on the classification result.

[0073] It should be noted that the bias is a constant term that each neuron in the neural network has independently. When the input of the neural network is 0, the bias can still trigger the neuron to output a non-zero signal.

[0074] In one possible implementation, the fully connected layers of the encoding layer gradually compress the input performance features, reducing their dimensionality and mapping them to a low-dimensional latent space. During this dimensionality reduction process, key features are captured and redundancy or noise is removed to produce the reduced-dimensional features.

[0075] 1-3, latent space layer 203.

[0076] The latent space layer 203 is used to store the first low-dimensional features output by the encoding layer 202 and perform feature extraction on the first low-dimensional features to obtain second low-dimensional features.

[0077] The latent space layer 203 is the core feature extraction layer of the feature reconstruction model.

[0078] 1-4. Decoding layer 204.

[0079] The decoding layer 204 is used to gradually restore the original feature dimension of the second low-dimensional feature to obtain a reconstructed feature. The decoding layer 204 is symmetrical to the encoding layer 202.

[0080] Among them, the decoding layer is used to map the dimensionality reduction features generated by the encoder back to the original data dimension and reconstruct the output close to the performance characteristics, that is, the reconstruction of the performance characteristics.

[0081] In one possible implementation, the decoding layer maps the reduced-dimensional features from the low-dimensional hidden layer to the output layer through a decoding formula that satisfies the following formula 2:

[0082] Formula 2.

[0083] in, Represents the output layer The reconstructed features of the output of neurons, represents the number of neurons in the output layer, Indicates that from the hidden layer neurons to the output layer The weight of a neuron, Indicates bias, Represents the activation function.

[0084] in, Defines the mapping rules for mapping the reduced dimensionality features back to the original data dimensions, Used to compensate for systematic deviations and align the reconstructed features with the performance characteristic data distribution.

[0085] 1-5. Output layer 205.

[0086] The output layer 205 is used to output the reconstructed features, and the dimensions of the reconstructed features are the same as those of the performance features.

[0087] In summary, the embodiment of the present application inputs performance characteristics into a feature reconstruction model. The encoding layer of the feature reconstruction model performs dimensionality reduction processing on the structured data in the performance characteristics and maps the reduced dimensionality data into a latent space layer. Key features are extracted from the data in the low-dimensional latent space layer. After the key feature data is extracted, the structured data in the key feature data is upgraded to the initial dimension through the encoding layer. The upgraded key feature data is then combined to obtain reconstructed features.

[0088] In one possible implementation, the feature reconstruction model is trained using the cross entropy loss between the performance characteristics of the fuel cell and the reconstructed characteristics as a loss function.

[0089] In another possible implementation, the feature reconstruction model uses a leaky linear rectifier unit (LeakyRelu) activation function as the activation function. The LeakyRelu activation function satisfies the following formula 3:

[0090] Formula 3.

[0091] Among them, x represents the performance characteristics of the output of the previous layer, Represents the introduced weight parameters.

[0092] Specifically, compared with the traditional Rectified Linear Unit (Relu) activation function, the LeakyRelu activation function introduces a weight parameter in the negative input interval on the basis of the Relu activation function, so that the output retains some negative information, that is, when x<0, the weight parameter is introduced. Multiplying the performance characteristics of the previous layer of data solves the problem that the traditional ReLU function maps all values ​​to 0 in the negative range, and avoids the problem of gradient disappearance during the training of the feature reconstruction model.

[0093] It should be noted that the vanishing gradient in the feature reconstruction model means that during the backpropagation process, the gradient decays exponentially with the increase in the number of network layers, causing the weight update of the encoder's input layer or the decoder's initial layer to almost stagnate, seriously affecting the model training effect.

[0094] S103: Fuse the performance feature and the reconstruction feature to obtain a fused feature.

[0095] The fusion feature refers to the feature obtained by fusing the performance feature and the reconstruction feature.

[0096] It should be noted that the fusion feature is intended to ensure a more accurate judgment of the water content state of the fuel cell stack.

[0097] In one possible implementation, the performance features and the reconstructed features output by the feature reconstruction model are fused through a fusion function to obtain fused features.

[0098] S104: Determine the water content of the fuel cell stack using the fusion feature.

[0099] The water content status of the fuel cell stack includes a first water shortage category, a second water shortage category, a water normal category, a first water flooding category and a second water flooding category with water content increasing in sequence.

[0100] Among them, the first water shortage category is a severe membrane dry state, the second water shortage category is a mild membrane dry state, the normal water category is a normal state, the first flooding category is a mild flooding state, and the second flooding category is a severe flooding state.

[0101] Among them, the core feature of the state prediction model is that all neurons in adjacent layers are connected, that is, the input of each neuron comes from the output of all neurons in the previous layer, and nonlinear transformation is achieved through weights and activation functions, which is often used in classification tasks. Figure 3 As shown, the structure of the state prediction model includes: input layer 301, self-attention mechanism layer 302, hidden layer 303, state prediction layer 304, and output layer 305.

[0102] 2-1. Input layer 301.

[0103] The input layer 301 is used to receive fused features, and the number of neurons is equal to the number of dimensions of the fused features.

[0104] 2-2. Self-attention mechanism layer 302.

[0105] Among them, the self-attention mechanism layer 302 is used to weight the data in the fusion feature according to the importance.

[0106] 2-3. Hidden layer 303.

[0107] The hidden layer 303 includes one or more fully connected layers, and extracts deep features of the fused features through a nonlinear activation function.

[0108] Among them, the fully connected layer is a network layer composed of multiple neurons.

[0109] 2-4. State prediction layer 304.

[0110] The state prediction layer 304 predicts the water state of the fuel cell stack based on the deep features extracted by the hidden layer 303 using the Softmax activation function.

[0111] 2-5. Output layer 305.

[0112] The output layer outputs the probability of the water-containing category of the fuel cell based on the prediction result calculated by the state prediction layer 304, and the number of neurons is equal to the number of water-containing categories of the fuel cell stack.

[0113] In another possible implementation, dynamic weights are assigned to the filtered feature data through a self-attention mechanism layer.

[0114] In one example, when a sudden drop in cathode voltage is detected, the self-attention mechanism layer assigns higher weights to voltage-related features while reducing the weights of irrelevant features (such as constant ambient temperature), allowing the state prediction model to focus on analyzing key data.

[0115] After assigning weights to the feature data, the state prediction layer calculates the probabilities of the five water-containing states of the fuel cell stack through the Softmax function based on the fusion features.

[0116] Furthermore, the output layer outputs the probability of the water content state of the fuel cell stack based on the results calculated by the state prediction layer. The five water content state probabilities of the fuel cell stack correspond to the five neurons in the output layer of the state prediction model.

[0117] according to Figure 1 According to the technical means in the invention, the present invention extracts the characteristics of the operation information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell to obtain the performance characteristics of the fuel cell to accurately represent the current status of the fuel cell; then the performance characteristics are reconstructed to obtain reconstructed characteristics, and the feature reconstruction captures the information in the data more accurately by generally combining or converting the original characteristics; then the fusion characteristics of the performance characteristics and the reconstructed characteristics are used to accurately determine the water status of the fuel cell stack, so as to accurately identify whether a water fault occurs in the fuel cell, so as to perform water fault processing when a water fault occurs in the fuel cell, so as to ensure the output power and efficiency of the fuel cell and improve the service life of the fuel cell.

[0118] like Figure 4 As shown, the above method S104 may further include: S401-S402.

[0119] S401: Input the fused features into the self-attention mechanism layer for self-attention learning to obtain intermediate features.

[0120] Among them, the self-attention mechanism refers to an adjustment mechanism that enables the neural network model to automatically focus on the part most relevant to the current task when processing input data through weight distribution, while ignoring secondary information.

[0121] Among them, the intermediate features refer to the features obtained after the self-attention mechanism assigns weights to the data in the fusion features.

[0122] In one possible implementation, a self-attention mechanism layer is inserted between the input layer and the hidden layer. The self-attention mechanism layer generates three vectors, namely "query", "key information" and "actual content" (value), through three small neural networks. The weight distribution of the fused features is generated by calculating the similarity between the fused features and these three vectors and then normalizing them through the Softmax function.

[0123] Furthermore, the weights are weighted and summed with the fusion features to generate intermediate features containing contextual relationships.

[0124] In one possible implementation, the intermediate feature satisfies the following formula 4:

[0125] Formula 4.

[0126] Among them, Q represents the question vector, K represents the key information vector, and V represents the actual content vector. Represents the scaling factor.

[0127] In one example, when the question vector is the water content at the cathode inlet of a fuel cell, the self-attention mechanism layer can use the high-frequency resistance feature as the key information vector and the fluctuations in the fuel cell's voltage and current as the actual content vector. The self-attention mechanism calculates the similarity between the data in the fused feature and the question vector, key information vector, and actual content vector, and then normalizes it using the Softmax function to generate a weighted distribution of the data in the fused feature.

[0128] It should be noted that Scaling factors are used to prevent exploding gradients in neural networks.

[0129] S402: Input the intermediate features into the state prediction layer to obtain the water content category probability distribution vector output by the state prediction layer.

[0130] The hidden layer and the output layer of the state prediction model are unified as a state prediction layer for the water state of the fuel cell stack.

[0131] Among them, the water content categories include the first water shortage category, the second water shortage category, the normal water category, the first flooding category and the second flooding category, where the water content increases in order from low to high.

[0132] Among them, the first water shortage category is a severe membrane dry state, the second water shortage category is a mild membrane dry state, the normal water category is a normal state, the first flooding category is a mild flooding state, and the second flooding category is a severe flooding state.

[0133] The water-containing category probability distribution vector refers to the vector of each water-containing category and its corresponding probability of occurrence.

[0134] In one possible implementation, the intermediate features are input into the state prediction layer and batch normalized. The normalized features satisfy the following formula 5:

[0135] Formula 5.

[0136] in, represents the normalized features, Represents the input layer The intermediate features of the input; Indicates the The mean of the intermediate features; Indicates the The variance of the intermediate features.

[0137] Furthermore, the hidden layer receives the outputs of all neurons in the previous layer as input data for the neurons in this layer. Each neuron performs a linear transformation on the input data. The result of this linear transformation is then passed through an activation function to introduce nonlinear relationships, enabling the network to fit complex data relationships. Feature extraction is performed on nonlinear data. Shallow hidden layers capture basic features, while deeper hidden layers learn complex features through layer-by-layer combination. Each neuron in the hidden layer integrates the global input to form a holistic understanding of the data distribution.

[0138] Furthermore, the Softmax function is applied in the state prediction layer to process each element of the input vector of the previous layer and perform an exponential operation. The generated new vector is then normalized, that is, mapped to the interval (0, 1), to form the probability of the water-containing category of each fuel cell stack, and the sum of the probabilities of each category is 1, thus obtaining the water-containing category probability distribution vector.

[0139] In one example, in an output water-containing category probability distribution vector: the probability of severe membrane drying is 0.8, the probability of mild membrane drying is 0.05, the probability of normal is 0.05, the probability of mild flooding is 0.05, and the probability of severe flooding is 0.05.

[0140] The Softmax function satisfies the following formula six:

[0141] Formula 6.

[0142] in, Represents the output layer The water category probability vector of the output fuel cell stack of each node, represents the number of neurons in the output layer, e represents the natural base, Represents the combined score of all water-containing categories.

[0143] In another possible implementation, the output water-containing category probability distribution vector is obtained through the Softmax function. Further, a loss function needs to be applied to quantify the error between the water-containing category probability prediction value and the true value, thereby measuring the performance of the model and guiding subsequent optimization.

[0144] according to Figure 4 In this application, a self-attention mechanism layer is used to perform self-attention learning on the fused features, thereby increasing the focus on key information in the fused features. Using the features learned after self-attention learning can improve the accuracy of the prediction of the water content of the fuel cell.

[0145] In one embodiment, Figure 5 As shown, the method provided in the embodiment of the present application also includes: S501-S502.

[0146] S501 , training a state prediction model using the cross entropy loss between the true value and the predicted value of the water-containing category probability distribution vector corresponding to the fusion feature of the fuel cell as a loss function.

[0147] In one possible implementation, the state prediction model is continuously trained by calculating the cross entropy value based on the cross entropy loss function, and the loss function is back-propagated to the input layer of the state prediction model, and the method in S501-S502 is repeated until the cross entropy value is reduced to an allowed range.

[0148] Specifically, the cross-entropy loss function is used to measure the difference between the predicted value and the true value of the probability distribution vector of the water category. Entropy can be seen as a measure of a system's degree of uncertainty and certainty, used to measure the degree of disorder in a system. Relative entropy, also known as the KL divergence (Kullback-Leibler divergence), is a metric used in information theory to measure the difference between two probability distributions. It is used to compare the similarity or difference between two probability distributions. Entropy satisfies the following formula 7:

[0149] Formula 7.

[0150] in, represents the true value of the probability distribution vector of the water content state category of the fuel cell stack, Representative event Y= The amount of information.

[0151] The relative entropy satisfies the following formula eight:

[0152] Formula 8.

[0153] in, Represents the predicted value of the probability distribution of the water content state category of the fuel cell stack.

[0154] The above formula shows that entropy is the expected value of the amount of information and is a measure of the uncertainty of a random variable. The larger the entropy value, the more difficult it is to determine the value of the random variable and the more unstable the system; the smaller the entropy value, the easier it is to determine the value of the random variable and the more stable the system. In practical applications, if is the true distribution of the target, and is the predicted distribution. In order to make the two as close as possible, it is necessary to minimize the relative entropy. The cross entropy is the mutual information between the two and is often used to reflect their degree of correlation. In deep learning classification tasks, is a certain target true probability distribution, so its entropy is a constant. Therefore, minimizing the relative entropy is equivalent to minimizing the cross entropy. The cross entropy function is used to measure the difference between the predicted value and the true value, which satisfies the following formula 9:

[0155] Formula nine.

[0156] in, represents the cross entropy value, represents the true value of the probability distribution vector of the water content state category of the fuel cell stack, Represents the predicted value of the probability distribution of the water content state category of the fuel cell stack.

[0157] It should be noted that the cross entropy value represents the inconsistency between the true value and the predicted value of the probability distribution vector of the water content state category of the fuel cell stack. The smaller the cross entropy value, the closer the two are, and the larger the cross entropy value, the greater the deviation between the two.

[0158] Furthermore, the loss function is back-propagated to the input layer of the state prediction model to repeat the above S501-S502 method until the cross entropy value is reduced to an allowed range, thereby obtaining a state prediction model that can accurately predict the probability of the water state category of the fuel cell stack.

[0159] S502 , training a feature reconstruction model using the cross entropy loss between the performance characteristics of the fuel cell and the reconstructed characteristics as a loss function.

[0160] In one possible implementation, the cross entropy loss function is used to train the feature reconstruction model.

[0161] Specifically, the degree of fit between the reconstructed features and the fuel cell's performance characteristics is used as the true value, and the degree of fit between the output value of the feature reconstruction model and the fuel cell's performance characteristics is used as the predicted value. When the cross-entropy value exceeds a preset standard value, the model parameters are optimized through backpropagation. Once the cross-entropy value falls below the preset standard value, the feature reconstruction model is successfully trained and its output feature reconstruction vector can well represent the fuel cell's performance characteristics.

[0162] like Figure 6 As shown, the present application provides a structural diagram of a device for determining the water content state of a fuel cell stack. Figure 6 As shown, the fuel cell stack water state determination device 60 can be used to execute the fuel cell stack water state determination method of any of the above embodiments. The fuel cell stack water state determination device 60 includes: a feature determination unit 601, a feature reconstruction unit 602, a feature fusion unit 603, and a determination unit 604.

[0163] A characteristic determination unit 601 is configured to determine a performance characteristic of the fuel cell based on operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell;

[0164] A feature reconstruction unit 602 is used to reconstruct the performance feature to obtain a reconstructed feature;

[0165] A feature fusion unit 603 is used to fuse the performance feature and the reconstruction feature to obtain a fused feature;

[0166] The determination unit 604 is configured to determine the water content of the fuel cell stack by utilizing the fusion feature.

[0167] In one possible embodiment, the feature determination unit 601 determines the performance characteristics of the fuel cell based on the operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell, including: representing the operating information, impedance and high-frequency resistance as a first vector; and performing feature extraction on the first vector representation to obtain performance characteristics.

[0168] In one possible approach, the feature reconstruction unit 602 reconstructs the performance feature to obtain the reconstructed feature, including determining the reconstructed feature based on the performance feature and a feature reconstruction model. The feature reconstruction model is trained using a cross entropy loss between the fuel cell's performance feature and the reconstructed feature as a loss function.

[0169] In one possible embodiment, the feature reconstruction unit 602, the feature reconstruction model includes an encoder and a decoder, and determines the reconstructed features based on the performance characteristics and the feature reconstruction model, including: inputting the performance characteristics into the encoder for dimensionality reduction encoding to obtain the reduced dimensionality features; inputting the reduced dimensionality features into the decoder for decoding to obtain the reconstructed features.

[0170] In one possible manner, the determination unit 604 uses the fusion features to determine the water status of the fuel cell stack, including: determining the water category probability distribution vector of the fuel cell based on the fusion features and the state prediction model; determining the water status based on the water category probability distribution vector; wherein the state prediction model is trained using the cross entropy loss between the true value and the predicted value of the water category probability distribution vector corresponding to the fusion features of the fuel cell as the loss function.

[0171] In one possible manner, in determination unit 604, the state prediction model includes a self-attention mechanism layer and a state prediction layer; based on the fused features and the state prediction model, the water-containing category probability distribution vector of the fuel cell is determined, including: inputting the fused features into the self-attention mechanism layer for self-attention learning to obtain intermediate features; inputting the intermediate features into the state prediction layer to obtain the water-containing category probability distribution vector output by the state prediction layer.

[0172] In one possible approach, the feature reconstruction model and the state prediction model are trained collaboratively.

[0173] In one possible approach, the operating information includes: water inlet temperature, cathode flow, output power, and output voltage.

[0174] In one possible approach, the water-containing category probability distribution vector is a vector representing each water-containing category and its corresponding probability of occurrence; wherein the water-containing categories include a first water-deficient category, a second water-deficient category, a water-normal category, a first flooded category, and a second flooded category with increasing water content.

[0175] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete the full classification or partial functions described above.

[0176] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0177] The units described as separate components may or may not be physically separate, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0178] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0179] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application, or the part that contributes to the prior art, or the entire classification part or part of the technical solution can be embodied in the form of a software product. The software product is stored in a storage medium and includes a number of instructions for enabling a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute the entire classification part or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, ROM, RAM, disk or optical disk, etc., various media that can store program code.

[0180] The above are only specific embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or replacements within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for determining the water content state of a fuel cell stack, characterized in that: The method comprises: determining a performance characteristic of the fuel cell based on operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell; Determining the reconstructed features based on the performance features and a feature reconstruction model; wherein the feature reconstruction model is trained using a cross entropy loss between the performance features of the fuel cell and the reconstructed features as a loss function; Fusing the performance feature and the reconstruction feature to obtain a fused feature; The hydration state of the fuel cell stack is determined using the fused features.

2. The method according to claim 1, characterized in that Determining the performance characteristics of the fuel cell based on the operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell includes: Representing the operation information, the impedance, and the high-frequency resistance as a first vector; Feature extraction is performed on the first vector representation to obtain the performance feature.

3. The method according to claim 1, characterized in that The feature reconstruction model includes an encoder and a decoder, and determining the reconstruction feature based on the performance feature and the feature reconstruction model includes: Inputting the performance characteristics into the encoder for dimensionality reduction encoding to obtain dimensionality reduction characteristics; The dimensionality reduction features are input into the decoder for decoding to obtain the reconstructed features.

4. The method according to claim 1, wherein The determining the water status of the fuel cell stack by utilizing the fusion feature includes: Determining a water-containing category probability distribution vector of the fuel cell based on the fusion feature and the state prediction model; determining the water content state based on the water content category probability distribution vector; The state prediction model is trained using the cross entropy loss between the true value and the predicted value of the water-containing category probability distribution vector corresponding to the fusion feature of the fuel cell as a loss function.

5. The method according to claim 4, characterized in that The state prediction model includes a self-attention mechanism layer and a state prediction layer; The determining of the water-containing category probability distribution vector of the fuel cell based on the fusion feature and the state prediction model includes: Inputting the fused features into the self-attention mechanism layer for self-attention learning to obtain intermediate features; The intermediate features are input into the state prediction layer to obtain the water-containing category probability distribution vector output by the state prediction layer.

6. The method according to claim 4, characterized in that The feature reconstruction model and the state prediction model are trained collaboratively.

7. The method according to claim 1, characterized in that The operation information includes: water inlet temperature, cathode flow, output power and output voltage.

8. The method according to claim 4 or 5, characterized in that The water-containing category probability distribution vector is a vector representing each water-containing category and its corresponding probability of occurrence; The water content categories include a first water shortage category, a second water shortage category, a normal water category, a first flooding category and a second flooding category, with the water content increasing in sequence.

9. A device for determining the water content state of a fuel cell stack, characterized in that: The device comprises: a characteristic determination unit, configured to determine a performance characteristic of the fuel cell based on operating information of the fuel cell stack and the impedance and high-frequency resistance of the fuel cell; A feature reconstruction unit, configured to determine the reconstructed feature based on the performance feature and a feature reconstruction model; wherein the feature reconstruction model is trained using a cross entropy loss between the performance feature of the fuel cell and the reconstructed feature as a loss function; a feature fusion unit, configured to fuse the performance feature and the reconstruction feature to obtain a fused feature; A determination unit is used to determine the water status of the fuel cell stack by using the fusion feature.

10. A vehicle, characterized in that: The vehicle includes a fuel cell; the fuel cell determines the water content state using the method according to any one of claims 1 to 8; The water content state is used to indicate whether a water failure occurs in the fuel cell stack; The water fault includes a water shortage fault and a flooding fault.

11. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method according to any one of claims 1 to 8.

Images