Transform-based fuel cell system life optimization method

Abstract

The invention discloses a Transform-based fuel cell system life optimization method, and relates to the field of fuel cells, and the method comprises the following steps: S1, collecting multi-dimensional data of a conventional driving behavior; s2, constructing a model for predicting the attenuation condition of the fuel cell system and the residual life of the electric pile; s3, collecting multi-dimensional data of a specific driving behavior, analyzing an energy demand mode and an energy consumption mode of the vehicle, and training the model; s4, collecting operation data in real time, and predicting the attenuation condition of the current fuel cell system and the residual life of the electric pile; s5, an optimization control strategy is generated, and real-time optimization adjustment of the operation parameters is carried out; and S6, continuously collecting operation data, and carrying out updating training on the model. By predicting and optimizing control and decision of the fuel cell system, the performance and the service life of the fuel cell system are improved, dynamic optimization and global optimization of the system are realized through a data-driven optimization strategy, so that the stability and the reliability of the system are improved, the service life of a galvanic pile is prolonged, and the maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel cells, and particularly to measuring electrical variables, specifically a method for optimizing the life of a fuel cell system based on Transformer. Background Art

[0002] As an efficient and environmentally friendly clean energy conversion device, the fuel cell system has the advantages of high energy density, environmental friendliness, quiet operation, etc., and has received extensive attention and application in the fields of transportation, stationary power generation, etc. in recent years. Its core component - the fuel cell stack directly converts the chemical energy of fuel (such as hydrogen) and oxidant (such as oxygen in the air) into electrical energy through an electrochemical reaction. Its basic principle is to promote the reaction of fuel and oxidant through catalysts on the electrodes to generate current and water, with the advantages of high energy conversion efficiency and few emissions.

[0003] However, during the operation of the fuel cell system, its performance will gradually degrade, and the life of the fuel cell stack will also be affected by various factors, such as temperature, pressure, gas flow rate, humidity, and load changes. Traditional control and optimization methods are difficult to accurately and dynamically handle these complex and interrelated factors, resulting in limitations in system degradation control and fuel cell stack life extension.

[0004] Therefore, it is necessary to improve the deficiencies in the prior art to solve the above problems. Summary of the Invention

[0005] The present invention overcomes the deficiencies of the prior art and provides a method for optimizing the life of a fuel cell system based on Transformer.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for optimizing the life of a fuel cell system based on Transformer, comprising the following steps:

[0007] S1. Collect vehicle energy data, driver driving data, fuel cell operation data, and environmental state data under normal driving behaviors, and preprocess the collected multi-dimensional data;

[0008] S2. Based on the preprocessed multi-dimensional data, construct a Transformer model for predicting the degradation status of the fuel cell system and the remaining life of the fuel cell stack;

[0009] S3. Collect vehicle energy data and driver driving data under specific driving behaviors, analyze the energy demand pattern and energy consumption pattern of the vehicle, and train the Transformer model;

[0010] S4. Collect the current operation data of the fuel cell system in real time. After preprocessing, input it into the Transformer model to predict the current attenuation status of the fuel cell system and the remaining life of the stack according to the driving behavior.

[0011] S5. Generate an optimized control strategy according to the prediction result, which acts on the execution components in the fuel cell system to perform real-time optimization and adjustment of the operation parameters.

[0012] In a preferred embodiment of the present invention, in the step of S1, the following sub-steps are included:

[0013] S11. Collect the vehicle energy data, driver driving data, fuel cell operation data and environmental status data under normal driving behavior.

[0014] S12. Preprocess the collected multi-dimensional data, including removing duplicate, incomplete or abnormal data items and performing normalization to ensure that the data is on the same scale.

[0015] In a preferred embodiment of the present invention, in the step of S2, the following sub-steps are included:

[0016] S21. Perform a relevance analysis on the preprocessed multi-dimensional data that affects the attenuation of the fuel cell system.

[0017] Calculate the Pearson correlation coefficient r for each influencing factor x and the attenuation status y:

[0018]

[0019] where x i is the multi-dimensional data value of the i-th sample; y i is the attenuation status of the i-th sample corresponding to x i ; is the mean value of the multi-dimensional data, that is is the mean value of the attenuation status, that is n is the number of samples, that is, the total number of observations;

[0020] S22. Assign weights to the multi-dimensional data that affects the attenuation of the fuel cell system.

[0021] Calculate the weight w i :

[0022]

[0023] where d′ j = d j ·|r j | is the adjusted coefficient of variation;

[0024] S23. Construct a Transformer model for outputting and predicting the degradation status of the fuel cell system and the remaining life of the stack;

[0025] S24. Divide a large amount of fuel cell system operation data into a training set and a validation set. Use the training set to train the Transformer model and the validation set to evaluate the performance of the Transformer model to ensure that the model has good generalization ability.

[0026] In a preferred embodiment of the present invention, in step S23, the construction of the Transformer model specifically includes the following steps:

[0027] S231. The input layer converts multi-dimensional data into a form that the model can process;

[0028] S232. The encoder consists of multiple encoder layers, and each encoder layer contains a multi-head self-attention mechanism and a feed-forward neural network;

[0029] S233. The decoder consists of multiple decoder layers, and each decoder layer contains a masked multi-head self-attention mechanism, a multi-head self-attention mechanism, and a feed-forward neural network;

[0030] S234. The output layer converts the output of the decoder into a prediction result, including a regression task: predicting the remaining life of the stack, and a classification task: predicting the degradation status of the fuel cell system.

[0031] In a preferred embodiment of the present invention, in step S3, the following sub-steps are included:

[0032] S31. Collect the vehicle's total energy data and the driver's driving data under different specific driving behaviors to obtain the vehicle's energy demand pattern and energy consumption pattern;

[0033] S32. Analyze the influence of different environmental state data on the vehicle's total energy data and the driver's driving data, and enhance the characteristics of the fuel cell operation data according to the analysis results;

[0034] S33. Use the enhanced characteristics of the fuel cell operation data to train the Transformer model to obtain a Transformer model for predicting the degradation status of the fuel cell system and the remaining life of the stack under specific driving behaviors.

[0035] In a preferred embodiment of the present invention, in step S31, the specific driving behaviors include: rapid acceleration, rapid deceleration, and frequent start and stop;

[0036] Analysis of the energy demand pattern: For the specific driving behavior, calculate the average value P of the vehicle's total demand power avg:

[0037]

[0038] Among them, T P is the time period of a specific driving behavior; P(t) is the change of the vehicle's demand power over time; dt is the infinitesimal change of time, i.e., the differential of time;

[0039] Analysis of the energy consumption pattern: For the specific driving behavior, calculate the total energy consumption E total :

[0040]

[0041] Among them, T E is the duration of the specific driving behavior.

[0042] In a preferred embodiment of the present invention, in the step of S32, the enhancement of the characteristics of the fuel cell operation data includes: construction of interaction item characteristics, environmental dynamic weighting characteristics, and polynomial feature expansion;

[0043] Construction of interaction item characteristics:

[0044] Z = X env × X fc ;

[0045] Among them, Z is the enhanced interaction feature; X env is the environmental parameter; X fc is the fuel cell operation parameter;

[0046] Environmental dynamic weighting characteristics:

[0047] X enhanced = w env · Y fc ;

[0048] Among them, X enhanced is the enhanced feature; w env is the environmental weight coefficient; Y fc is the original fuel cell parameter;

[0049] Polynomial feature expansion:

[0050]

[0051] Among them, β i is the regression coefficient, which is fitted by the least squares method.

[0052] In a preferred embodiment of the present invention, in the step of S5, the following sub-steps are included:

[0053] S51. According to the prediction result in step S4, an optimization algorithm is used to generate an optimized control strategy for the fuel cell system, and the adjusted parameters and ranges are determined;

[0054] S52. The optimized control strategy is converted into corresponding control signals, and the control signals are sent to the execution components of the fuel cell system to adjust the operating parameters of the fuel cell in real time to achieve optimized control.

[0055] In a preferred embodiment of the present invention, it further includes step S6: continuously collect the operating data of the fuel cell system, and update and train the Transformer model to keep the model with high prediction accuracy and optimization effect.

[0056] In a preferred embodiment of the present invention, in the step S6, it includes the following sub-steps:

[0057] S61. Continuously collect the operating data of the fuel cell system. Every certain period of time or when the newly collected data volume reaches a certain scale, the new data is merged with the historical data;

[0058] S62. Re-divide the training set and the validation set, and update and train the Transformer model to adapt to the possible component aging and operating environment changes of the system, and ensure that the prediction accuracy and optimization effect of the model remain at a high level.

[0059] The present invention solves the defects in the background technology, and the present invention has the following beneficial effects:

[0060] (1) The present invention provides a method for optimizing the life of a fuel cell system based on Transformer. By using multi-dimensional data to provide rich basic information, applying the Transformer model for prediction and optimizing the control and decision-making of the fuel cell system, accurate prediction and optimized control of the performance decay and the remaining life of the fuel cell stack of the fuel cell system are realized. It not only improves the performance and life of the fuel cell system, but also realizes the dynamic optimization and global optimization of the system through a data-driven optimization strategy, thereby contributing to improving the stability and reliability of the system, extending the service life of the fuel cell stack, and reducing the maintenance cost.

[0061] (2) In the present invention, by obtaining multi-dimensional data during the operation of the fuel cell system, various influencing factors of the fuel cell system in actual operation can be comprehensively captured, providing rich input information for model training, and at the same time ensuring the quality of the data, avoiding interference with model training, thereby improving the stability and accuracy of the model.

[0062] (3) In the present invention, by performing a relevance analysis on multi-dimensional data that affects the degradation of the fuel cell system and assigning weights to important features, the sensitivity of the model to key features is enhanced, the prediction accuracy of the model is improved. At the same time, the constructed Transformer model can capture the complex relationships and dynamic changes among various factors in the fuel cell system, which helps to more accurately understand the influencing factors of system performance degradation and stack life. Combining the multi-head self-attention mechanism and the feed-forward neural network to achieve global optimization of the system. Compared with traditional prediction methods, it has higher prediction accuracy and stronger generalization ability, thus effectively improving the performance and life of the stack.

[0063] (4) In the present invention, by collecting data under specific driving behaviors and analyzing the energy demand and consumption patterns, the prediction ability of the model for the performance degradation of the fuel cell system and the remaining life of the stack under different driving behaviors is enhanced. This not only improves the prediction accuracy and generalization ability of the model, but also provides more accurate data support for the generation of optimized control strategies, thereby realizing real-time optimization and adjustment of the fuel cell system, extending the service life of the stack, reducing maintenance costs, and improving the stability and reliability of the system.

[0064] (5) In the present invention, an optimized control strategy is generated in real time according to the prediction results, which can dynamically adjust the operating parameters of the fuel cell system, perform more precise control and optimization of the fuel cell system, adapt to load changes and environmental condition changes, improve the dynamic adaptability of the system. By continuously collecting new operating data and regularly updating the model, it is ensured that the prediction accuracy and optimization effect of the model always remain at a high level, extending the service life of the model and enhancing the robustness and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings;

[0066] Figure 1 It is a flowchart of a method for optimizing the life of a fuel cell system based on Transformer according to a preferred embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0068] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0069] As Figure 1 shown, a method for optimizing the lifespan of a fuel cell system based on Transformer includes the following steps:

[0070] S1. Collect the vehicle energy data, driver driving data, fuel cell operation data, and environmental status data under normal driving behaviors, and preprocess the collected multi-dimensional data;

[0071] S2. Based on the preprocessed multi-dimensional data, construct a Transformer model for predicting the degradation status of the fuel cell system and the remaining lifespan of the stack;

[0072] S3. Collect the vehicle energy data and driver driving data under specific driving behaviors, analyze the energy demand pattern and energy consumption pattern of the vehicle, and train the Transformer model;

[0073] S4. Real-time collect the current operation data of the fuel cell system, preprocess it, and input it into the Transformer model to predict the current degradation status of the fuel cell system and the remaining lifespan of the stack according to the driving behavior;

[0074] S5. According to the prediction results, generate an optimization control strategy, which acts on the execution components in the fuel cell system to perform real-time optimization adjustment of the operation parameters;

[0075] S6. Continuously collect the operation data of the fuel cell system, update and train the Transformer model to keep the model with high prediction accuracy and optimization effect.

[0076] It should be noted that by utilizing multi-dimensional data to provide rich basic information, applying the Transformer model for prediction and optimizing fuel cell system control and decision-making, accurate prediction and optimized control of the performance degradation of the fuel cell system and the remaining life of the stack are achieved. This not only improves the performance and life of the fuel cell system, but also realizes the dynamic optimization and global optimization of the system through data-driven optimization strategies, thereby contributing to improving the stability and reliability of the system, extending the service life of the stack, and reducing maintenance costs.

[0077] In some specific implementation cases, in the step of S1, it includes the following sub-steps:

[0078] S11. Collect the vehicle energy data, driver driving data, fuel cell operation data, and environmental status data under normal driving behaviors;

[0079] S12. Preprocess the collected multi-dimensional data; including removing duplicate, incomplete, or abnormal data items, and performing normalization processing to ensure that the data is on the same scale.

[0080] It should be noted that by obtaining multi-dimensional data during the operation of the fuel cell system, various influencing factors of the fuel cell system during actual operation can be comprehensively captured, providing rich input information for model training. At the same time, the quality of the data is ensured, interference with model training is avoided, and thus the stability and accuracy of the model are improved.

[0081] In this embodiment, in the step of S11, normal driving behaviors include: constant-speed cruising (stable throttle opening, constant vehicle speed, and stable power demand), gentle acceleration (gradually increasing throttle opening, gradually increasing vehicle speed, and linearly increasing power demand), and gentle deceleration (gradually decreasing throttle opening or gently stepping on the brake, gradually decreasing vehicle speed, and linearly decreasing power demand); the vehicle energy data includes: vehicle demand power, voltage, current, and power of the power battery; the driver driving data includes: driving behaviors of the driver (throttle and brake operations); the fuel cell operation data includes: voltage, current, temperature, power output, efficiency, and hydrogen consumption of the fuel cell; the environmental status data includes: temperature, humidity, and air pressure of the environment.

[0082] In this embodiment, in the step of S12, it is necessary to remove duplicate data from the collected multi-dimensional data, identify and process outliers: Outliers may be caused by data entry errors or special situations of patients, and need to be processed according to the actual situation, such as replacing with the mean, median, or deleting; the specific normalization process is to scale the data proportionally so that it falls into a small specific interval, such as [0,1] or [-1,1], which helps to eliminate the influence of the dimension on the result. The calculation formula is:

[0083]

[0084] Among them, X is the original data; X' is the normalized data.

[0085] In some specific embodiments, in the step of S2, the following sub-steps are included:

[0086] S21. Conduct a relevance analysis on the preprocessed multi-dimensional data that affects the degradation of the fuel cell system;

[0087] S22. Assign weights to the multi-dimensional data that affects the degradation of the fuel cell system;

[0088] S23. Construct a Transformer model for outputting and predicting the degradation status of the fuel cell system and the remaining life of the stack;

[0089] S24. Divide a large amount of fuel cell system operation data into a training set and a validation set, use the training set to train the Transformer model, and use the validation set to evaluate the performance of the Transformer model to ensure that the model has good generalization ability.

[0090] It should be noted that by conducting a relevance analysis on multi-dimensional data that affects the degradation of the fuel cell system and assigning weights to important features, the sensitivity of the model to key features is enhanced, the prediction accuracy of the model is improved. At the same time, the constructed Transformer model can capture the complex relationships and dynamic changes among various factors in the fuel cell system, which helps to more accurately understand the influencing factors of system performance degradation and stack life. Combining the multi-head self-attention mechanism and the feed-forward neural network to achieve global optimization of the system. Compared with traditional prediction methods, it has higher prediction accuracy and stronger generalization ability, thus effectively improving the performance and life of the stack.

[0091] In this embodiment, in the step of S21, the multi-dimensional data is used as the influencing factor for establishing the Transformer model, and the Pearson correlation coefficient r is calculated for each influencing factor x and the degradation status y:

[0092]

[0093] where x i is the multi-dimensional data value of the i-th sample; y i is the degradation status of the i-th sample corresponding to x i ; is the mean value of the multi-dimensional data, that is is the mean value of the degradation status, that is n is the number of samples, that is, the total number of observed values;

[0094] When the Pearson correlation coefficient r = 1, it indicates that the influencing factor x and the attenuation condition y are completely positively correlated, that is, the increase of one variable is always accompanied by the increase of the other variable; when r = -1, it indicates that the influencing factor x and the attenuation condition y are completely negatively correlated, that is, the increase of one variable is always accompanied by the decrease of the other variable; when r = 0, it indicates that there is no linear correlation between the influencing factor x and the attenuation condition y; the closer the absolute value of r is to 1, the stronger the linear relationship between the influencing factor x and the attenuation condition y, and the closer it is to 0, the weaker the linear relationship.

[0095] In this embodiment, in the step of S22, the correlation coefficient is combined with the entropy weight method to calculate the weight of the influencing factor. The entropy weight method is an objective weight allocation method that determines the weight based on the degree of data dispersion rather than subjective judgment. The combination of the two ensures that the weight allocation not only considers the degree of data dispersion but also the actual correlation strength between the influencing factor and the pain degree.

[0096] Calculate the entropy value e of each influencing factor j :

[0097]

[0098] Among them, x′ ij is the value of the i-th sample on the j-th influencing factor, and p ij represents the proportion of the i-th sample on the j-th influencing factor, and m is the total number of evaluation objects;

[0099] The difference coefficient d reflecting the degree of index dispersion j :

[0100] d j = 1 - e j ;

[0101] Combined with the correlation adjustment, the difference coefficient is d j ′ = d j ·|r j |;

[0102] Calculate the weight w of each influencing factor using the adjusted difference coefficient i :

[0103]

[0104] In this embodiment, in the step of S23, the construction of the Transformer model specifically includes the following steps:

[0105] S231. The input layer converts the multi-dimensional data into a form that the model can process;

[0106] Input linear transformation:

[0107] src = Linear(X');

[0108] Wherein, X' is the input data; Linear is a linear transformation layer for converting the feature dimension of the input data from feature_dim to d_model;

[0109] S232. The encoder is composed of multiple encoder layers, and each encoder layer includes a multi-head self-attention mechanism and a feed-forward neural network;

[0110] Multi-head self-attention mechanism MSA:

[0111] MSA(Q, K, V) = Concat(head1, head2,..., head h )W O

[0112]

[0113] Wherein, Q, K, and V are the query, key, and value matrices respectively; and are the linear transformation weight matrices respectively; W O is the output linear transformation weight matrix; d k is the dimension of the key; h is the number of heads; QK T is the matrix multiplication operation of the query matrix Q and the transpose K of the key matrix K T ;

[0114] Feed-forward neural network FFN:

[0115] FFN(s) = max(0, sW1 + b1)W2 + b2;

[0116] Wherein, s is the matrix input to this layer; W1 and W2 are linear transformation weight matrices; b1 and b2 are bias terms;

[0117] Layer normalization and residual connection:

[0118] LayerNorm(s + Sublayer(s));

[0119] Wherein, Sublayer(s) is the output of the sublayer; LayerNorm is the layer normalization operation;

[0120] S233. The decoder is composed of multiple decoder layers, and each decoder layer includes a masked multi-head self-attention mechanism, a multi-head self-attention mechanism, and a feed-forward neural network;

[0121] Masked multi-head self-attention mechanism MMSA:

[0122] MMSA(Q, K, V) = Concat(head1, head2,..., head h )W O

[0123]

[0124] where mask is a mask matrix used to prevent the current position from seeing information of future positions;

[0125] Multi-Head Self-Attention Mechanism MSA:

[0126] MSA(Q, K, V) = Concat(head1, head2,..., head h )W O

[0127]

[0128] Feed-Forward Neural Network FFN:

[0129] FFN(s) = max(0, sW1 + b1)W2 + b2;

[0130] Layer Normalization and Residual Connection:

[0131] LayerNorm(s + Sublayer(s));

[0132] S234. The output layer converts the output of the decoder into a prediction result, including a regression task (remaining life of the stack) and a classification task (attenuation status of the fuel cell system);

[0133] Regression task: output_regression = Linear(decoder_output);

[0134] Classification task: output_classification = softmax(Linear(decoder_output)).

[0135] In this embodiment, in the step of S24, during the training of the Transformer model, the error between the quantization prediction value and the true value is quantified:

[0136] L = α·MSE(output_regression, true_regression) + (1 - α)·CE(output_classification, ture_classification) Where, L is the loss function; MSE is the mean squared error loss function, which is used for the regression task (prediction of the remaining life of the stack); CE is the cross-entropy loss function, which is used for the classification task (attenuation status of the fuel cell system);

[0137] α is the balance parameter, which is used to balance the losses of the regression and classification tasks;

[0138] Mean squared error loss function MSE:

[0139]

[0140] Where, N is the number of samples; z i is the true value; v i is the predicted value;

[0141] Cross-entropy loss function CE:

[0142]

[0143] Where, C is the number of classes; z ij is the one-hot encoding of the true label; v ij is the predicted probability;

[0144] Continuously adjust the weight parameters of the model through the backpropagation algorithm, and calculate the gradient of the loss function with respect to each parameter:

[0145]

[0146] Where, L is the loss function; is the model output; u is the intermediate variable; θ is the model parameter;

[0147] Update the model parameters using the optimizer:

[0148]

[0149] Where, θ t is the parameter at the t-th iteration; η is the learning rate; m t is the first moment estimate; υ t is the second moment estimate; ∈ is a small constant used to prevent the denominator from being zero;

[0150] After each training epoch ends, calculate the loss on the validation set. If the loss value on the validation set does not decrease for n consecutive training epochs, stop training:

[0151] if val_Lt ≥val_L t-1 For n epochs, stop training.

[0152] In some specific embodiments, in step S3, the following sub-steps are included:

[0153] S31. Collect the vehicle's overall energy data and the driver's driving data under different specific driving behaviors to obtain the vehicle's energy demand pattern and energy consumption pattern;

[0154] S32. Analyze the influence of different environmental state data on the vehicle's overall energy data and the driver's driving data, and enhance the characteristics of the fuel cell operation data according to the analysis results;

[0155] S33. Use the enhanced characteristics of the fuel cell operation data to train the Transformer model to obtain a Transformer model for predicting the attenuation status of the fuel cell system and the remaining life of the fuel cell stack under specific driving behaviors.

[0156] It should be noted that by collecting data under specific driving behaviors, analyzing the energy demand and consumption patterns, and enhancing the model's prediction ability for the performance attenuation of the fuel cell system and the remaining life of the fuel cell stack under different driving behaviors, not only the prediction accuracy and generalization ability of the model are improved, but also more accurate data support is provided for the generation of optimized control strategies, thereby realizing real-time optimization and adjustment of the fuel cell system, extending the service life of the fuel cell stack, reducing the maintenance cost, and improving the stability and reliability of the system.

[0157] In this embodiment, in step S31, the vehicle's overall energy data and the driver's driving data collected under different specific driving behaviors are preprocessed in the same manner as in step S1; specific driving behaviors include: rapid acceleration (the throttle pedal opening increases rapidly, and the power demand surges within a short time), rapid deceleration (the brake pedal is frequently stepped on or deeply pressed, and the energy recovery system is frequently involved), and frequent start-stop (the power demand fluctuates in a sawtooth shape);

[0158] During the process of obtaining the energy demand pattern and energy consumption pattern, extract the temporal characteristics representing the driving behavior and construct a feature vector: throttle operation feature and brake operation feature;

[0159] Throttle operation feature: Calculate the average throttle opening

[0160]

[0161] where T is the time window length; α(t) is the throttle opening value at time point t;

[0162] Calculate the throttle change rate Δα(t):

[0163]

[0164] Among them, α(t - 1) is the throttle opening value at time point t - 1; Δt is the time interval, representing the time difference between two consecutive time points; Hard acceleration count: Count the number of times the throttle opening change rate exceeds the threshold (such as Δα > 50% / s);

[0165] Braking operation characteristics: Braking frequency: The number of braking times per unit time; Calculate the average braking depth

[0166] Among them, β(t) is the braking depth value at time point t;

[0167] Energy demand pattern analysis: For a specific driving behavior, calculate the average value P of the vehicle's demand power avg :

[0168]

[0169] Among them, T P is the time period of a specific driving behavior; P(t) is the change of the vehicle's demand power over time; dt is the infinitesimal change of time, that is, the differential of time;

[0170] Energy consumption pattern analysis: For a specific driving behavior, calculate the total energy consumption E total :

[0171]

[0172] Among them, T E is the duration of a specific driving behavior.

[0173] In this embodiment, in the step of S32, under different environmental state data, the influence on the vehicle's energy data and the driver's driving data is analyzed by using the Pearson correlation coefficient, and the correlation analysis steps are the same as those in S21;

[0174] Feature enhancement of fuel cell operation data includes: Interaction term feature construction, environmental dynamic weighted features, and polynomial feature expansion;

[0175] Interaction term feature construction:

[0176] Z = X env × X fc ;

[0177] Among them, Z is the enhanced interaction feature, reflecting the adjustment effect of the environment on fuel cell parameters; X envis an environmental parameter (such as temperature, humidity); X fc is a fuel cell operating parameter (such as output power, hydrogen consumption rate);

[0178] Environmental dynamic weighted feature:

[0179] X enhanced = w env ·Y fc ;

[0180] Among them, X enhanced is the enhanced feature, reflecting the dynamic influence of the environment on the parameter; w env is the environmental weight coefficient; Y fc is the original fuel cell parameter (such as voltage, efficiency);

[0181] Environmental weight coefficient w env Calculation: If the correlation coefficient between the environmental temperature T C and the fuel cell efficiency η fc is r T,η , the temperature weight

[0182] Among them, δ is the scaling factor; T nominal is the standard environmental temperature;

[0183] Polynomial feature expansion:

[0184]

[0185] Among them, β i is the regression coefficient, fitted by the least squares method;

[0186] Normalize the constructed new features to ensure that the data scales input into the model are consistent:

[0187]

[0188] Among them, μZ is the mean of feature Z; σZ is the standard deviation of feature Z.

[0189] In this embodiment, in the step of S33, the training step of the Transformer model is the same as the training step in S24.

[0190] In some specific implementation schemes, in the step of S4, the operation data includes: vehicle energy data, driver driving data, fuel cell operation data, and environmental status data; the preprocessing step is the same as the preprocessing step in S1.

[0191] In some specific implementation schemes, in the step of S5, it includes the following sub-steps:

[0192] S51. According to the prediction results in step S4, an optimization algorithm is used to generate an optimized control strategy for the fuel cell system, and the adjusted parameters and ranges are determined.

[0193] S52. Convert the optimized control strategy into corresponding control signals, and send the control signals to the execution components of the fuel cell system to adjust the operating parameters of the fuel cell in real time and achieve optimized control.

[0194] It should be noted that generating an optimized control strategy in real time according to the prediction results can dynamically adjust the operating parameters of the fuel cell system, perform more precise control and optimization on the fuel cell system, adapt to load changes and environmental condition changes, and improve the dynamic adaptability of the system.

[0195] In this embodiment, in step S51, the optimization algorithm is a rule-based optimization algorithm. For example, if the decay rate of the stack performance exceeds a certain threshold threshold_rate, the heat dissipation power of the cooling system is adjusted; if the stack temperature exceeds a certain threshold threshold_temp, the heat dissipation power of the cooling system is adjusted; if the gas flow is unstable, the rotation speed of the gas supply pump or the opening degree of the control valve is adjusted.

[0196] In this embodiment, in step S52, the control signal can be a value or an instruction for adjusting the parameters of the execution component; the sending of the control signal to the execution component of the fuel cell system can be implemented through a control system (such as PLC or DCS).

[0197] In some specific implementation schemes, in step S6, the following sub-steps are included:

[0198] S61. Continuously collect the operating data of the fuel cell system. Every certain period of time or when the amount of newly collected data reaches a certain scale, the new data is merged with the historical data.

[0199] S62. Re-divide the training set and the validation set, and update the training of the Transformer model to adapt to the possible component aging and operating environment changes of the system, and ensure that the prediction accuracy and optimization effect of the model remain at a high level.

[0200] It should be noted that by continuously collecting new operating data and regularly updating the model, the prediction accuracy and optimization effect of the model are ensured to always remain at a high level, the service life of the model is extended, and the robustness and generalization ability of the model are enhanced.

[0201] In this embodiment, in step S61, in the model update and maintenance stage, every certain period of time (such as 1 month) or when the amount of newly collected data reaches a certain scale (such as 100,000 pieces), the new data is merged with the historical data.

[0202] In this embodiment, in step S62, the step of updating and training the Transformer model is the same as step S24.

[0203] Based on the ideal embodiments of the present invention as the inspiration, through the above description, for those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be embraced within the present invention. Any reference signs in the claims should not be construed as limiting the claimed rights.

[0204] In addition, it should be understood that although this specification is described according to embodiments, not every embodiment only contains an independent technical solution. This narrative way of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for optimizing the lifespan of a fuel cell system based on Transformer, characterized in that, It includes the following steps: S1. Collect the vehicle's whole - vehicle energy data, driver driving data, fuel cell operation data, and environmental status data under normal driving behaviors, and pre - process the collected multi - dimensional data; S2. Based on the pre - processed multi - dimensional data, construct a Transformer model for predicting the degradation status of the fuel cell system and the remaining life of the fuel cell stack; S3. Collect the vehicle's whole - vehicle energy data and driver driving data under specific driving behaviors, analyze the vehicle's energy demand pattern and energy consumption pattern, and train the Transformer model; S4. Real - time collect the current operation data of the fuel cell system, pre - process it, input it into the Transformer model, and predict the current degradation status of the fuel cell system and the remaining life of the fuel cell stack according to the driving behavior; S5. According to the prediction results, generate an optimized control strategy, which acts on the execution components in the fuel cell system to perform real - time optimization and adjustment of the operation parameters.

2. The method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 1, wherein: In the step of S1, it includes the following sub - steps: S11. Collect the vehicle's whole - vehicle energy data, driver driving data, fuel cell operation data, and environmental status data under normal driving behaviors; S12. Pre - process the collected multi - dimensional data; It includes removing duplicate, incomplete or abnormal data items, and performing normalization processing to ensure that the data is on the same scale.

3. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 1, characterized in that: In the step of S2, it includes the following sub - steps: S21. Perform a relevance analysis on the pre - processed multi - dimensional data that affects the degradation of the fuel cell system; Calculate the Pearson correlation coefficient r for each influencing factor x and the degradation status y; where x i is the multi-dimensional data value of the i-th sample; y i is the attenuation condition of the i-th sample corresponding to x i ; is the mean of the multi-dimensional data, that is is the mean of the attenuation conditions, that is n is the number of samples, that is, the total number of observations; S22. Assign weights to the multi - dimensional data that affects the degradation of the fuel cell system; Calculate the weight w of each influencing factor i : where d j ′ = d j ·|r j | is the adjusted coefficient of variation; S23. Construct a Transformer model for outputting the prediction of the degradation status of the fuel cell system and the remaining life of the fuel cell stack; S24. Divide a large amount of fuel cell system operation data into a training set and a validation set, use the training set to train the Transformer model, and use the validation set to evaluate the performance of the Transformer model to ensure that the model has good generalization ability.

4. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 3, characterized in that: In the step of S23, the construction of the Transformer model specifically includes the following steps: S231. The input layer converts the multi - dimensional data into a form that the model can process; S232. The encoder is composed of multiple encoder layers, and each encoder layer contains a multi - head self - attention mechanism and a feed - forward neural network; S233. The decoder is composed of multiple decoder layers, and each decoder layer contains a masked multi - head self - attention mechanism, a multi - head self - attention mechanism, and a feed - forward neural network; S234. The output layer converts the output of the decoder into a prediction result, including a regression task: predicting the remaining life of the fuel cell stack, and a classification task: predicting the degradation status of the fuel cell system.

5. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 1, characterized in that: In the step of S3, it includes the following sub - steps: S31. Collect the vehicle's whole - vehicle energy data and driver driving data under different specific driving behaviors to obtain the vehicle's energy demand pattern and energy consumption pattern; S32. Analyze the influence of different environmental status data on the vehicle's whole - vehicle energy data and driver driving data, and according to the analysis results, enhance the characteristics of the fuel cell operation data. S33. Use the enhanced fuel cell operation data features to train the Transformer model to obtain a Transformer model for predicting the degradation status of the fuel cell system and the remaining life of the stack under specific driving behaviors.

6. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 5, characterized in that: In the step of S31, the specific driving behaviors include: rapid acceleration, rapid deceleration, and frequent start-stop. The energy demand pattern analysis: For the specific driving behavior, calculate the average value P of the vehicle's total demand power avg : Among them, T P is the time period of a specific driving behavior; P(t) is the change of the vehicle's demand power over time; dt is the infinitesimal change in time, that is, the differential of time; Analysis of the energy consumption pattern: For the specific driving behavior, calculate the total energy consumption E total : where T E is the duration of a specific driving behavior.

7. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 5, characterized in that: In the step of S32, the enhancement of the fuel cell operation data features includes: interaction term feature construction, environmental dynamic weighted features, and polynomial feature expansion. Interaction term feature construction: Z = X env × X fc ; Among them, Z is the enhanced interaction feature; X env is the environmental parameter; X fc is the fuel cell operation parameter; Environmental dynamic weighted features: X enhanced = w env · Y fc ; Among them, X enhanced is the enhanced feature; w env is the environmental weight coefficient; Y fc is the original fuel cell parameter; Polynomial feature expansion: Among them, β i is the regression coefficient, which is fitted by the least squares method.

8. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 1, characterized in that: In the step of S5, it includes the following sub-steps: S51. According to the prediction results in step S4, use an optimization algorithm to generate an optimized control strategy for the fuel cell system, and determine the adjusted parameters and ranges. S52. Convert the optimized control strategy into corresponding control signals, and send the control signals to the execution components of the fuel cell system to adjust the operation parameters of the fuel cell in real time to achieve optimized control.

9. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 1, characterized in that: It also includes step S6. Continuously collect the operation data of the fuel cell system, and update and train the Transformer model to keep the model with high prediction accuracy and optimization effect.

10. A method for optimizing the lifespan of a fuel cell system based on Transformer according to claim 9, characterized in that: In the step of S6, it includes the following sub-steps: S61. Continuously collect the operation data of the fuel cell system. At regular intervals or when the amount of newly collected data reaches a certain scale, merge the new data with the historical data. S62. Re-partition the training set and the validation set, and update and train the Transformer model to adapt to the possible component aging and operating environment changes of the system, and ensure that the prediction accuracy and optimization effect of the model remain at a high level.

Images