Method for balancing service life and performance of fuel cell

Abstract

The invention relates to the technical field of battery energy storage system evaluation, and relates to the technical field of battery energy storage, in particular to a method for balancing the service life and performance of a fuel cell, which comprises the following steps of: 1, acquiring multi-source data of the fuel cell, and applying a series of preset pulse current load sequences to the fuel cell, measuring response parameters output by the fuel cell in real time; 2, Fourier transform processing is carried out on the response parameters, and frequency domain signals of the response parameters are obtained; the performance and service life balance model is constructed based on the fusion feature vector containing the dynamic performance, the service life characteristic and the operation state of the fuel cell, the load distribution weight can be dynamically adjusted, the performance and the service life of the single fuel cell are kept balanced, and therefore the overall service life of the fuel cell pack is prolonged, and the reliability of the fuel cell pack is improved. And the influence on the performance of the whole system due to overload or too fast loss of the single fuel cell is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of battery energy storage, and in particular to a method for balancing the life and performance of a fuel cell. Background Art

[0002] Fuel cells are highly efficient and clean electrochemical energy conversion devices and are considered one of the important options for future energy. Compared with traditional internal combustion engines, fuel cells have the advantages of zero emissions, higher energy conversion efficiency and low noise. Therefore, fuel cells have been widely used in transportation, fixed power supply and portable power supply. However, a major challenge facing fuel cells in practical applications is how to extend their operating life while maintaining high performance output.

[0003] After searching, Chinese patent number CN202310167828.5 discloses a method for balancing battery life and battery life, including: when the battery is running at a first voltage, regularly monitoring the current temperature of the battery; if the current temperature of the battery exceeds the preset warning temperature, recording high temperature data once; judging whether the number of consecutive recordings of the high temperature data is greater than the preset number; if the number of consecutive recordings of the high temperature data is greater than the preset number, controlling the battery to run at a second voltage as the battery full charge voltage; the second voltage is less than the first voltage; if the number of consecutive recordings of the high temperature data is not greater than the preset number, continuing to run at the first voltage as the battery full charge voltage. The above scheme can ensure the battery life and safety by lowering the battery voltage at high temperature, and can increase the battery usage time by raising the charging voltage at low temperature, and detect that the number of high temperature data recorded is greater than the preset number to prevent misjudgment. If the battery is judged to need voltage reduction processing only by recording high temperature data once, misjudgment is prone to occur. After the normal switching of low voltage as the full charge voltage, set a period of time to re-record the high temperature number, and if there is no high temperature exceeding the preset number of times in the middle, the battery full charge voltage is set to a higher standard voltage for operation.

[0004] However, in the existing battery life balancing methods, although the battery life is dynamically balanced by the battery temperature to alleviate the impact of the high temperature and high pressure environment on the battery life, the battery voltage is changed during the balancing process, which in turn affects the battery performance efficiency; but in the process of battery temperature and voltage changes, different parameter responses will also occur, which may cause the battery performance to change nonlinearly. The existing balancing methods lack an effective balancing mechanism to fuse multi-source information, resulting in insufficient robustness and accuracy of the evaluation results; in addition, in the battery pack, the performance and life balance of the single cells may be uneven. If the overall performance and life balancing strategy based on the battery pack fails to fully consider the differences between the single cells, it may cause the overall balance value of the battery pack to be unstable, further exacerbating the consistency problem between the single cells. Summary of the invention

[0005] In view of the above shortcomings of the prior art, the present invention provides a method for balancing the life and performance of a fuel cell, which can effectively solve the problem in the prior art of ignoring the balance of battery performance when adjusting the battery life based on temperature.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: The present invention provides a method for balancing the life and performance of a fuel cell. The technical solution adopted by the present invention is as follows: comprising the following steps: Step 1: Collect multi-source data of the fuel cell, apply a preset series of pulse current load sequences to the fuel cell, and measure the response parameters of the fuel cell output in real time; Step 2: Perform Fourier transform processing on the response parameters to obtain the frequency domain signal of the response parameters; Step 3: Extract frequency domain characteristics from frequency domain signals, fuse frequency domain characteristics and multi-source data at vector level to obtain fused feature vectors; Step 4: Construct a performance and life balance model based on the fusion feature vector to predict the balance index of fuel cell performance and life. ; Step 5: Balance index analysis based on single fuel cell Construct a fuel cell group control model and calculate the load distribution weight of the single fuel cell group , and then calculate the actual load distribution based on the fuel cell load distribution weight , using actual load distribution Its response parameters Update and obtain consistency control parameters; return the consistency control parameters to the performance and life balance model, and calculate the updated balance index score. Repeat the above steps to calculate the distribution weight difference of the single fuel cell ,in Assign weights to the updated load; if the assigned weights are different , calculate excess load distribution , The total output power of the battery pack and the excess load distribution distributed to other fuel cells.

[0007] The multi-source data includes: Electrochemical performance parameters: discharge capacity of fuel cells , open circuit voltage and resistor ; Environmental parameters: Temperature and humidity ; Gas supply parameters: oxygen flow rate and hydrogen flow rate ; Core component status: catalyst active area and PEM conductivity , where the conductivity of the proton exchange membrane is The calculation formula is: ; In the formula, is the optimal conductivity; is the humidity deviation, is the temperature deviation, is the humidity attenuation factor, is the temperature attenuation factor, is an exponential function.

[0008] The series of pulse current load sequences are preset as follows: Define the pulse waveform type as rectangular wave, and define the pulse waveform parameters of the rectangular wave, including: maximum current density ; Minimum current density ; Duty cycle , the calculation formula is: ,in is the duration of the high current state, is the pulse cycle time; pulse frequency , the calculation formula is: ; Based on the maximum current density , minimum current density and active area Calculating Peak Current and valley current , the calculation formula is: and ; Based on the above parameters, calculate the average current density and average power density , the calculation formula is: ; ; is the average voltage actually output by the fuel cell during the pulse period, and the calculation formula is: ; represents the output voltage of the fuel cell at any time scale t, where t is the time scale; Combine the time scale t and the pulse waveform parameters to construct a pulse current load sequence ; The method of measuring the response parameter output by the fuel cell includes: Apply dynamic current to the fuel cell according to a preset series of pulse current load sequences and collect response parameters ; In the formula, is the dynamic output voltage of the fuel cell; is the dynamic output current of the fuel cell; is the dynamically changing internal resistance of the fuel cell; is the dynamic operating temperature of the fuel cell; is the dynamically changing humidity within the fuel in the fuel cell; is the dynamic supply of hydrogen to the anode; The dynamic supply of oxygen to the cathode; is the membrane dynamic conductivity; is the dynamic discharge capacitance of the fuel cell; is the catalyst active area of ​​the fuel cell.

[0009] The method of performing Fourier transform processing on the response parameters includes: From the response parameters Extract time series data, including the dynamic output voltage of the fuel cell and dynamic output current , defines the sampling frequency of time series data and number of sampling points ; Remove the DC offset in the time series data. The removal formula is: ; In the formula, is the original time series data; n is the discrete index of the time series data; Indicates the value of the time series data at the nth time scale t; Represents the DC component of time series data, and the calculation formula is: , is the total length of the time series data; represents the time series data after removing the DC offset; After removing the DC offset, the time series data Perform windowing processing, the processing formula is: ; In the formula, is the window function; is the signal processed by the window function, that is, the weighted time series data; Perform Fourier transform on the weighted time series data, and the transformation formula is: : In the formula, The frequency domain signal output by fast Fourier transform processing; is the discrete frequency, and the calculation formula is: , k is the frequency index; is the Fourier basis function; The frequency domain characteristics include the frequency component with the largest amplitude , harmonic amplitude , frequency domain energy distribution and energy ratio .

[0010] The method of vector-level fusion of the frequency domain characteristics and multi-source data includes: Calculate key battery characteristics based on frequency domain characteristics and multi-source data, including: Voltage response characteristics ; Discharge capacity retention rate ; In the formula, is the initial discharge capacity; Energy efficiency ; In the formula, is the time interval; is the calorific value of combustion of hydrogen; Catalyst activity ; In the formula, is the initial catalyst active area; Proton exchange membrane conductivity ; Performance decay rate ; In the formula, is the time derivative of the voltage drop; the above parameters are standardized and dimensionalized, and constructed as a fusion feature vector .

[0011] The method of constructing the performance and life balance model includes: Defining the balance between fuel cell performance and life , and its calculation formula is ; In the formula, is the short-term performance efficiency, which is calculated as: ; is the life loss rate; is the adjustment cost of the operating conditions, and the calculation formula is: ,in, , and The adjustment cost weights of the corresponding parameters are usually determined by providing experimental data; , and is the weight coefficient of the corresponding parameter, .

[0012] Among them, the life loss rate The way to obtain is: The fused feature vector Input into the pre-trained life loss model to predict the life loss rate ; The training methods of life loss model include: Define a recursive neural network as the basic structure of the life loss model, which includes an input layer, a hidden layer, and an output layer; Collect the frequency domain characteristics and multi-source data of multiple fuel cells in the past fixed time, and construct the corresponding fusion feature vector, which is recorded as the historical fusion feature vector ,in Represents a vector time step; fuses the history with the feature vector Expand by time step to form sequence data as the training input of the life loss model, and label the sequence data according to the vector time step; the label is the life state index of the fuel cell; the labeling range of the life state index is 0 to 100%; Initialize the network parameters of the life loss model; the network parameters include the weight matrix from the input layer to the hidden layer , the recurrent weight matrix of the hidden layer , the bias vector of the hidden layer , the weight matrix from hidden layer to output layer and the bias vector of the output layer ; Define the loss function for the lifetime loss model ;in, is the length of the sequence data; is the loss weight function weighted according to the life state; It is A vector of life state indices for time steps; is the quantile loss function; It is The life state index is annotated with a vector of time steps; It is The predicted values ​​of the life loss model for vector time steps; is the quantile level; ;in, is the indicative function; The threshold interval of the preset life state, the upper limit of the threshold interval is ; The lower limit of the threshold interval is ;like Less than ,but for ;like If it is within the threshold range, is 1; if Greater than ,but for ;in, and is the preset weight coefficient, and greater than 1, Less than 1; The sequence data is passed to the input layer, and then passes through the hidden layer and the output layer in turn to output the predicted value; the value of the corresponding loss function is calculated, and the error gradient is back-propagated from the output layer to the input layer; according to the error gradient, the network parameters are updated using the optimization algorithm; the sequence data is repeatedly trained until the life loss model converges or reaches the preset number of iterations, that is, the training of the life loss model is completed.

[0013] Among them, the input layer is used to receive the fused feature vector as input; the hidden layer consists of It is composed of neural units; The calculation formula for each neural unit is: ;in, is the current vector time step The hidden state vector of is the hidden state vector of the previous vector time step; is the activation function; is the weight matrix from the input layer to the hidden layer; is the recurrent weight matrix of the hidden layer; is the bias vector of the hidden layer; The output layer calculates the predicted value based on the output of the hidden layer ;in, is the activation function of the output layer; is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

[0014] The fuel cell group control model is constructed in the following manner: Defining load distribution weights The calculation formula is: ; Indicates The load weight ratio that should be allocated to each single fuel cell; It is The balance index of each fuel cell monomer; It represents the sum of the balance indexes of all the individual fuel cells in the fuel cell stack. is the total number of single fuel cells, is the index of the fuel cell; Defining the actual load distribution The calculation formula is: ; In the formula is the total output probability of the battery pack, and the calculation formula is: ,in and are the total output voltage and total output current of the fuel cell group respectively; The update formula that defines the consistency control parameters includes: ; In the formula, For fuel cells in actual load distribution The actual current under For the The operating voltage of each fuel cell; For the The internal resistance of a fuel cell; is the voltage drop in the ohmic region; is the power loss caused by the internal resistance of the fuel cell; For the The actual energy utilization efficiency of a fuel cell; is the input energy of the single fuel cell, and its calculation formula is: ,in is the Gibbs free energy of the reaction; Calculate the distribution weight difference of a single fuel cell , the calculation formula is: .

[0015] Among them, the updated load distribution weight The constraints are: ; Excess load distribution The constraints are: .

[0016] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. In the present invention, the operating status of the battery can be fully reflected by collecting multi-source data, providing a high-quality basis for subsequent data processing and ensuring the accuracy of the analysis results.

[0017] 2. In the present invention, by applying a pulse current load sequence to the fuel cell, the dynamic response characteristics of the fuel cell under different load conditions can be captured, revealing its performance potential and limitations; and the response parameters of the fuel cell under different loads are collected. By performing Fourier transform processing on the response parameters, the nonlinear behavior and dynamic response capability of the battery can be revealed, and the response capability of the fuel cell to loads of different frequencies can be quantified, providing a basis for performance evaluation.

[0018] 3. In the present invention, by fusing frequency domain characteristics and multi-source data at vector level, the obtained fused feature vector contains the dynamic performance, life characteristics and operating status of the fuel cell, which can comprehensively reflect the health status of the battery, and the unified feature vector form is convenient for input into the machine learning model, thereby improving the prediction ability of the model and ensuring that the model can reflect the latest status of the battery; then, a performance and life balance model is constructed based on the fused feature vector, which can dynamically adjust the load distribution weight to balance the performance and life of the single fuel cell, thereby extending the overall life of the fuel cell group and avoiding the performance of the entire system being affected by overload or excessive loss of a single fuel cell.

[0019] 4. In the present invention, the load distribution differences of individual fuel cells are calculated, the load distribution is dynamically adjusted, low-scoring cells are protected, and high-scoring cells are preferentially utilized, which can improve the operating consistency and overall health of the battery pack, optimize the load distribution and operating conditions, and enhance the overall efficiency and reliability of the battery pack. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The following is a simplified diagram showing the method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] Example: Refer to Figure 1 , this case proposes a method for balancing fuel cell life and performance, including the following steps: Step 1: Collect multi-source data of the fuel cell, apply a preset series of pulse current load sequences to the fuel cell, and measure the response parameters of the fuel cell output in real time; Step 2: Perform Fourier transform processing on the response parameters to obtain the frequency domain signal of the response parameters; Step 3: Extract frequency domain characteristics from frequency domain signals, fuse frequency domain characteristics and multi-source data at vector level to obtain fused feature vectors; Step 4: Construct a performance and life balance model based on the fusion feature vector to predict the balance index of fuel cell performance and life. ; Step 5: Balance index analysis based on single fuel cell Construct a fuel cell group control model and calculate the load distribution weight of the single fuel cell group , and then calculate the actual load distribution based on the fuel cell load distribution weight , using actual load distribution Its response parameters Update and obtain consistency control parameters; return the consistency control parameters to the performance and life balance model, and calculate the updated balance index score. Repeat the above steps to calculate the distribution weight difference of the single fuel cell ,in Assign weights to the updated load; if the assigned weights are different , calculate excess load distribution , The total output power of the battery pack and the excess load distribution distributed to other fuel cells.

[0023] Specifically, in this case, the multi-source data of fuel cells include: Electrochemical performance parameters: discharge capacity of fuel cells , open circuit voltage and resistor ; Discharge capacity A direct measure of the remaining useful life of a fuel cell; open circuit voltage Characterize the internal electrochemical energy of the fuel cell under no-current conditions, which is used as a reference voltage to evaluate whether there is significant degradation; Environmental parameters: Temperature and humidity ;temperature It affects the reaction rate of the catalyst surface, thus affecting the output performance of the fuel cell. It also has the function of correcting the conductivity of the proton exchange membrane and is a key regulating factor of the conductivity of the proton exchange membrane. Deviation from the optimal range may lead to dehydration or overhydration of the PEM, which in turn affects the conductivity of the PEM; humidity For hydration status assessment and gas management optimization, humidity Humidity is critical to the ionic conductivity of the proton exchange membrane. Too low will cause membrane dehydration, humidity Too high a temperature will cause difficulty in gas transmission and flooding, and is also an important environmental condition for gas diffusion and reaction rate in fuel cells; Gas supply parameters: oxygen flow rate and hydrogen flow rate ; Used for gas utilization analysis and mass permeation impedance analysis. The dynamic change of gas flow rate reflects the utilization efficiency of reactants. Whether the hydrogen and oxygen supply is sufficient directly affects the working performance of the fuel cell. Insufficient gas supply will lead to increased concentration polarization in the oxygen or hydrogen region, thereby increasing internal resistance and affecting performance. Core component status: catalyst active area and PEM conductivity , active area The range that affects the total current is used to convert the current density into the actual applied current value, usually obtained by measuring the surface area of ​​the actual anode or cathode of the fuel cell; the conductivity of the proton exchange membrane It indicates the dynamic change of the conductivity of the proton exchange membrane per unit time and per unit thickness. It is the core indicator of the proton transport capacity in fuel cells and is affected by temperature. and humidity The significant influence of is calculated as follows: ;in, The optimal conductivity of the proton exchange membrane under ideal humidity and temperature conditions; is the humidity deviation, which is the current humidity With ideal humidity The deviation is calculated as: ; is the temperature deviation, which is the current temperature With ideal temperature The deviation is calculated as: ; is the humidity attenuation factor, which describes the coefficient of the effect of humidity deviation on the conductivity of the proton exchange membrane; is the temperature attenuation factor, which describes the coefficient of the effect of temperature deviation on the conductivity of the proton exchange membrane; is an exponential function.

[0024] By collecting multi-source data, the operating status of the battery can be fully reflected, providing a high-quality foundation for subsequent data processing and ensuring the accuracy of the analysis results.

[0025] More specifically, a series of pulse current load sequences are preset as follows: Define the pulse waveform type as rectangular wave, and define the pulse waveform parameters of the rectangular wave, including: Maximum current density , is the current density of the peak part of the pulse waveform, representing the current intensity at the maximum dynamic load. It is usually set according to the ultimate capacity of the fuel cell to avoid exceeding the rated operating range of the battery. It is generally set to 70%-80% of the rated operating density of the fuel cell to reflect the typical dynamic load capacity of the fuel cell, to avoid setting it too high and causing problems such as overheating, proton exchange membrane drying or insufficient fluid supply; Minimum current density , is the current density of the valley part of the pulse waveform, representing the current intensity at the minimum dynamic load. It is generally set to 20%-30% of the rated working density of the battery to ensure that the electrochemical reaction is still within the normal range. Avoid setting it to 0 (open circuit state), which will lead to unstable experimental results or inability to test dynamic response; Duty Cycle , which represents the proportion of high current load duration in the pulse waveform, is an important parameter for simulating actual dynamic load conditions. The calculation formula is: ,in is the duration of the high current state, is the pulse cycle time (including the sum of high and low current state time); Pulse frequency , that is, the switching rate of the pulse waveform, is the frequency of repeated changes of the waveform, and the calculation formula is: ; Based on the maximum current density , minimum current density and active area Calculating Peak Current and valley current , the calculation formula is: and ; Based on the above parameters, the dynamic relationship between the pulse waveform parameters is calculated, including the average current density and average power density , the calculation formula is: ; Average current density It represents the average current density of the fuel cell in one pulse cycle and is used to evaluate the overall capacity of the fuel cell under continuous load; ; Among them, the average power density Represents the unit active area of ​​the fuel cell in one pulse cycle The average probability output is the core indicator for measuring the overall output capacity and energy utilization efficiency of fuel cells; is the average voltage actually output by the fuel cell during the pulse period, and the calculation formula is: ; represents the output voltage of the fuel cell at any time scale t, where t is the time scale; Combine the time scale t and the pulse waveform parameters to construct a pulse current load sequence ; Apply dynamic current to the fuel cell according to a preset series of pulse current load sequences and collect response parameters ; in, It is the dynamic output voltage of the fuel cell, reflecting the instantaneous energy performance and dynamic response efficiency of the fuel cell; It is the dynamic output current of the fuel cell, an important indicator of gas consumption, and is closely related to the dynamic load; The dynamically changing internal resistance of the fuel cell reflects the health status of the internal materials of the fuel cell and the dynamic operation stability; The dynamic operating temperature of the fuel cell affects the catalyst performance, proton exchange membrane conductivity and reaction rate; The dynamically changing humidity in the fuel in the fuel cell affects the hydration state of the proton exchange membrane; The dynamic supply of hydrogen to the anode is related to the chemical reaction amount of the fuel cell; The dynamic supply of oxygen to the cathode; is the membrane dynamic conductivity; is the dynamic discharge capacitance of the fuel cell; is the catalyst active area of ​​the fuel cell.

[0026] By applying a pulse current load sequence, the dynamic response characteristics of the fuel cell under different load conditions can be captured, revealing its performance potential and limitations.

[0027] Response parameters Ways to perform Fourier transform processing include: From the response parameters Extract key time series data, including the dynamic output voltage of the fuel cell and dynamic output current , defines the sampling frequency of time series data and number of sampling points ; Remove the DC offset in the time series data so that the dynamic change part can be analyzed. The removal formula is: ; In the formula, is the original time series data; n is the discrete index of the time series data, indicating a discrete time scale; Indicates the value of the time series data at the nth time scale t; Represents the DC component of the time series data, that is, the average value, which is used to extract the static, low-frequency bias part of the time series data. The calculation formula is: , is the total length of the time series data; It represents the time series data after the DC offset is removed, that is, the remaining time series data only represents the dynamic characteristics that change with time, and in the time domain, it fluctuates more around zero than the original data; After removing the DC offset, the time series data Perform windowing to reduce spectrum leakage. The processing formula is: ; In the formula, is the window function, a weight function used to force the boundaries of time series data to gradually decrease to zero; is the signal processed by the window function, that is, the weighted time series data; Perform Fourier transform on the weighted time series data, and the transformation formula is: : In the formula, It is the frequency domain signal output by fast Fourier transform processing, which is a complex number representing the response parameter at discrete frequency. Amplitude and phase on discrete frequencies The calculation formula is: , k is the frequency index; is the Fourier basis function.

[0028] From the frequency domain signal The way to extract frequency domain characteristics is: Calculate the amplitude of a frequency domain signal step by step , the calculation formula is: ; In the formula, is the square of the real part, indicating the intensity based on the cosine wave; is the square of the imaginary part, indicating the intensity based on the sine wave; Find the frequency domain signal The frequency component with the largest amplitude , which usually corresponds to the main dynamic characteristics of time series data, such as the pulse frequency in a pulse current load sequence , the calculation formula is: ; Extract the amplitude of the harmonic component, which is the pulse frequency The integer multiple frequency components of the harmonic components are normalized to the pulse frequency. The ratio of the amplitude, i.e. the harmonic amplitude , to quantify the nonlinear characteristics; the calculation formula is: ;in, represents the amplitude of the kth harmonic component; Calculate frequency components Energy distribution in the mid-frequency domain , the calculation formula is: , where the integral range [a, b] represents the frequency interval of energy calculation; frequency domain energy distribution Including low frequency energy (energy in the frequency range [0, 1Hz]) and high frequency energy (Energy in the frequency range [1, 5Hz]), the calculation formulas are: , ;in, is the discrete frequency interval, and the calculation formula is ; Based on low frequency energy and high frequency energy Compute energy ratios for evaluating dynamical complexity of time series data , the calculation formula is: ; From the response parameters Extract other parameters, including the dynamic internal resistance of the fuel cell , Dynamic operating temperature , Dynamically changing humidity , Dynamic supply of hydrogen , Dynamic oxygen supply and membrane dynamic conductivity , used to explain the frequency domain characteristics, the expressions explaining the relationship include: , , and ; Finally, the frequency domain characteristics are output, including the frequency component with the largest amplitude. , harmonic amplitude , frequency domain energy distribution and energy ratio ; By passing the response parameter Convert to frequency domain signal , the dynamic characteristics of the fuel cell can be extracted, and then the frequency domain signal Extracting frequency domain characteristics can reveal the nonlinear behavior and dynamic response capability of the battery, quantify the response capability of the fuel cell to loads of different frequencies, and provide a basis for performance evaluation.

[0029] The methods of vector-level fusion of frequency domain characteristics and multi-source data include: Calculate key battery characteristics based on frequency domain characteristics and multi-source data, including: Voltage response characteristics , reflects the pulse frequency of the fuel cell in the pulse current load sequence The response capability to load changes is calculated as follows: ;when When , it indicates that the fuel cell is running stably; Discharge capacity retention rate , reflects the remaining available capacity of the fuel cell, and the calculation formula is: ; In the formula, is the initial discharge capacity, when When When , it indicates that the battery discharge capacity has declined due to aging or insufficient fuel; Energy efficiency , directly measures the utilization of input energy (chemical fuel) by the fuel cell and is an important reference for optimizing operating performance. The calculation formula is: ; In the formula, is the time interval; is the combustion calorific value of hydrogen; When When , it indicates that there is efficiency loss, which is often caused by problems such as excessive internal resistance or side reactions; Catalyst activity , which reflects the quantitative value of the catalyst health status, is calculated as follows: ; In the formula, is the initial catalyst active area, catalyst activity characteristics The value of indicates the health status of the current activity. When When , it means that the activity of the catalyst decreases due to aging or contamination; Proton exchange membrane conductivity , which affects the conductivity of the proton exchange membrane , the calculation formula is: ; Proton exchange membrane conductivity The value represents the conductivity of the proton exchange membrane performance status, when When , it indicates that the conductivity of the proton exchange membrane is the best. When , it means that the conductivity of the proton exchange membrane has declined due to drying, degradation or high temperature; Performance decay rate , reflects the performance decline rate of the fuel cell, and the calculation formula is: ; In the formula, is the time derivative of the voltage drop, indicating the rate at which the fuel cell performance decays; when When When , it indicates that the performance is stable and unchanged; The above parameters are standardized and dimensionalized and constructed as a fusion feature vector , which can comprehensively reflect the status and performance of the fuel cell.

[0030] Fusion feature vector It includes the dynamic performance, life characteristics and operating status of the fuel cell, which can comprehensively reflect the health status of the battery. The unified feature vector form is easy to input into the machine learning model, improving the model's predictive ability and ensuring that the model can reflect the latest status of the battery.

[0031] Ways to build a performance and life balance model include: Defining the balance between fuel cell performance and life , which is used to comprehensively evaluate the current operating status of the fuel cell. The calculation formula is: ; In the formula, The higher the value, the better the balance between battery performance and life. A value of is too low, indicating that operating conditions need to be optimized or maintenance needs to be performed; It is the short-term performance efficiency, which reflects the energy utilization efficiency of the fuel cell under the current operating conditions. Its calculation formula is: ; is the life loss rate, which reflects the degradation rate of the fuel cell; The potential impact of performance optimization on life is quantified by the adjustment cost of operating conditions, calculated as: ,in, , and The adjustment cost weights of the corresponding parameters are usually determined by providing experimental data; , and is the weight coefficient of the corresponding parameter, which is used to adjust the importance of performance, life and adjustment cost in the comprehensive index. ; Life loss rate The way to obtain is: The fused feature vector Input into the pre-trained life loss model to predict the life loss rate ; The training methods of life loss model include: Define a recursive neural network as the basic structure of the life loss model, which includes an input layer, a hidden layer, and an output layer; The input layer is used to receive the fused feature vector As input, represents the vector time step; the hidden layer is composed of It is composed of neural units; The calculation formula for each neural unit is: ;in, is the current vector time step The hidden state vector of is the hidden state vector of the previous vector time step; is the activation function (such as tanh or ReLU); is the weight matrix from the input layer to the hidden layer; is the recurrent weight matrix of the hidden layer; is the bias vector of the hidden layer; The output layer calculates the predicted value based on the output of the hidden layer ;in, is the activation function of the output layer (such as a linear function or softmax); is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

[0032] Collect the frequency domain characteristics and multi-source data of multiple fuel cells in the past fixed time, and construct the corresponding fusion feature vector, which is recorded as the historical fusion feature vector ; Fusion of history feature vector Expand by time step to form sequence data as the training input of the life loss model, and label the sequence data according to the vector time step; the label is the life state index of the fuel cell; the labeling range of the life state index is 0 to 100%; Initialize the network parameters of the life loss model; the network parameters include the weight matrix from the input layer to the hidden layer , the recurrent weight matrix of the hidden layer , the bias vector of the hidden layer , the weight matrix from hidden layer to output layer and the bias vector of the output layer ; Define the loss function for the lifetime loss model ;in, is the length of the sequence data; is the loss weight function weighted according to the life state; It is A vector of life state indices for time steps; is the quantile loss function; It is The annotated life state index for each vector time step; It is The predicted values ​​of the life loss model for vector time steps; is the quantile level; ;in, is an indicator function; an indicator function is a function used to indicate whether a certain condition is true or not; The threshold interval of the preset life state, the upper limit of the threshold interval is ; The lower limit of the threshold interval is ;like Less than ,but for ;like If it is within the threshold range, is 1; if Greater than ,but for ;in, and is the preset weight coefficient, and Greater than 1, the purpose is to increase the loss weight when the life state is poor, so that the model pays more attention to this state; Less than 1; the purpose is to appropriately reduce the loss weight when the life state is good to avoid over-tracking this state; specific and Value, based on the analysis of historical data, when the life status is poor or good, the deviation distribution of the model prediction is obtained, and according to the actual application scenario, the tolerance for poor or good life status is determined to obtain; The setting of the life status threshold range and weight coefficient effectively guides the model to pay more attention to key areas and improves the accuracy of the battery life status assessment.

[0033] The sequence data is passed to the input layer, then passes through the hidden layer and the output layer in turn, and the predicted value, i.e., the life loss rate, is output. ; and calculate the value of the corresponding loss function, and back-propagate the error gradient along the output layer to the input layer; according to the error gradient, use the optimization algorithm (such as gradient descent, Adam, etc.) to update the network parameters to reduce the value of the loss function; repeat the training of the sequence data until the life loss model converges or reaches the preset number of iterations, that is, the training of the life loss model is completed; the convergence of the life loss model can be understood as the value of the loss function no longer changes.

[0034] The balance index between fuel cell performance and life is divided into Compared with the preset balance threshold interval, the upper limit of the balance threshold interval is , the lower limit of the equilibrium threshold interval is ;like , indicating that the performance and life of the fuel cell have reached the best balance and no additional optimization is required; if , indicating that the performance and life of the fuel cell are not completely balanced and need to be regulated; if , indicating that the operating state of the fuel cell has deteriorated significantly and the operating conditions need to be optimized as a priority.

[0035] The machine learning model predicts the balance index of each fuel cell performance and life , which can dynamically reflect the latest status of the battery. The performance parameters provide the battery's immediate operating performance, the life parameters reveal the long-term health trend, and the balance indicators combine the two to achieve a comprehensive operating status assessment.

[0036] The fuel cell stack control model is constructed as follows: Based on the balance index of each fuel cell performance and life Calculate its load distribution weight , the calculation formula is: ; Indicates The load weight ratio that each single fuel cell should be allocated is The value is between 0 and 1, which is used to adjust the load distribution weight. , allocating more load to the single fuel cells with better performance and life balance, reducing the pressure on the single cells with poor performance and life balance, thereby optimizing the consistency and overall life of the battery pack; It is The balance index of each fuel cell monomer; It represents the sum of the balance indexes of all the individual fuel cells in the fuel cell stack. is the total number of single fuel cells, is the index of the fuel cell; Load distribution weight based on each single fuel cell Calculate its actual load distribution , the calculation formula is: ; In the formula is the total output probability of the battery pack, and the calculation formula is: ,in and are the total output voltage and total output current of the fuel cell group respectively; Based on the actual load distribution of each single fuel cell Its response parameters Update to obtain consistency control parameters, including: ; In the formula, For fuel cells in actual load distribution The actual current under For the The operating voltage of each fuel cell; For the The internal resistance of a fuel cell; is the voltage drop in the ohmic region, obtained by experimental measurement; is the power loss caused by the internal resistance of the fuel cell; For the The actual energy utilization efficiency of a fuel cell reflects how much of the input energy is converted into effective output power; is the input energy of the single fuel cell, and its calculation formula is: ,in is the Gibbs free energy of the reaction; Return the consistency control parameters to the performance and life balance model, and calculate the updated balance index score. , based on the updated balance index Calculate updated load distribution weights ; and the updated load distribution weight The following conditions are met: ; Calculate the distribution weight difference of a single fuel cell , the calculation formula is: ; If the weight difference , calculate excess load distribution , and distribute the excess load Distributed to other single fuel cells to ensure the total power demand of the fuel cell group , so that each fuel cell in the fuel cell stack maintains the highest efficiency and longest life, while the fuel cell stack maintains the highest efficiency and longest life.

[0037] By calculating the load distribution differences of individual fuel cells, dynamically adjusting the load distribution, protecting low-scoring batteries, and giving priority to high-scoring batteries, the operating consistency and overall health of the battery pack can be improved, and the load distribution and operating conditions can be optimized to enhance the overall efficiency and reliability of the battery pack.

[0038] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for balancing fuel cell life and performance, characterized in that: The following steps are involved: Step 1: Collect multi-source data of the fuel cell, apply a preset series of pulse current load sequences to the fuel cell, and measure the response parameters of the fuel cell output in real time; Step 2: Perform Fourier transform processing on the response parameters to obtain the frequency domain signal of the response parameters; Step 3: Extract frequency domain characteristics from frequency domain signals, fuse frequency domain characteristics and multi-source data at vector level to obtain fused feature vectors; Step 4: Construct a performance and life balance model based on the fusion feature vector to predict the balance index of fuel cell performance and life. ; Step 5: Balance index analysis based on single fuel cell Construct a fuel cell group control model and calculate the load distribution weight of the single fuel cell group , and then calculate the actual load distribution based on the fuel cell load distribution weight , using actual load distribution Its response parameters Update and obtain consistency control parameters; The consistency control parameters are returned to the performance and life balance model to calculate the updated balance index score. Repeat the above steps to calculate the distribution weight difference of the single fuel cell ,in Assign weights to updated loads; If the weight difference , calculate excess load distribution , The total output power of the battery pack and the excess load distribution distributed to other fuel cells.

2. A method for balancing fuel cell life and performance as claimed in claim 1, characterized in that: The multi-source data includes: Electrochemical performance parameters: discharge capacity of fuel cells , open circuit voltage and resistor ; Environmental parameters: Temperature and humidity ; Gas supply parameters: oxygen flow rate and hydrogen flow rate ; Core component status: catalyst active area and PEM conductivity , where the conductivity of the proton exchange membrane is The calculation formula is: ; In the formula, is the optimal conductivity; is the humidity deviation, is the temperature deviation, is the humidity attenuation factor, is the temperature attenuation factor, is an exponential function.

3. A method for balancing fuel cell life and performance as claimed in claim 2, characterized in that: The preset mode of the series of pulse current load sequences is: Define the pulse waveform type as rectangular wave, and define the pulse waveform parameters of the rectangular wave, including: maximum current density ; Minimum current density ; Duty cycle , the calculation formula is: ,in is the duration of the high current state, is the pulse cycle time; pulse frequency , the calculation formula is: ; Based on the maximum current density , minimum current density and active area Calculating Peak Current and valley current , the calculation formula is: and ; Based on the above parameters, calculate the average current density and average power density , the calculation formula is: ； ; is the average voltage actually output by the fuel cell during the pulse period, and the calculation formula is: ; represents the output voltage of the fuel cell at any time scale t, where t is the time scale; Combine the time scale t and the pulse waveform parameters to construct a pulse current load sequence ; The method of measuring the response parameter output by the fuel cell includes: Apply dynamic current to the fuel cell according to a preset series of pulse current load sequences and collect response parameters ; In the formula, is the dynamic output voltage of the fuel cell; is the dynamic output current of the fuel cell; is the dynamically changing internal resistance of the fuel cell; is the dynamic operating temperature of the fuel cell; is the dynamically changing humidity within the fuel in the fuel cell; is the dynamic supply of hydrogen to the anode; The dynamic supply of oxygen to the cathode; is the membrane dynamic conductivity; is the dynamic discharge capacitance of the fuel cell; is the catalyst active area of ​​the fuel cell.

4. A method for balancing fuel cell life and performance as claimed in claim 3, characterized in that: The method of performing Fourier transform processing on the response parameters includes: From the response parameters Extract time series data, including the dynamic output voltage of the fuel cell and dynamic output current , defines the sampling frequency of time series data and number of sampling points ; Remove the DC offset in the time series data. The removal formula is: ; In the formula, is the original time series data; n is the discrete index of the time series data; Indicates the value of the time series data at the nth time scale t; Represents the DC component of time series data, and the calculation formula is: , is the total length of the time series data; represents the time series data after removing the DC offset; After removing the DC offset, the time series data Perform windowing processing, the processing formula is: ; In the formula, is the window function; is the signal processed by the window function, that is, the weighted time series data; Perform Fourier transform on the weighted time series data, and the transformation formula is: ： In the formula, The frequency domain signal output by fast Fourier transform processing; is the discrete frequency, and the calculation formula is: , k is the frequency index; is the Fourier basis function; The frequency domain characteristics include the frequency component with the largest amplitude , harmonic amplitude , frequency domain energy distribution and energy ratio .

5. A method for balancing fuel cell life and performance as claimed in claim 4, characterized in that: The method of vector-level fusion of the frequency domain characteristics and multi-source data includes: Calculate key battery characteristics based on frequency domain characteristics and multi-source data, including: Voltage response characteristics ; Discharge capacity retention rate ; In the formula, is the initial discharge capacity; Energy efficiency ; In the formula, is the time interval; is the calorific value of combustion of hydrogen; Catalyst activity ; In the formula, is the initial catalyst active area; Proton exchange membrane conductivity ; Performance decay rate ; In the formula, is the time derivative of the voltage drop; the above parameters are standardized and dimensionalized, and constructed as a fusion feature vector .

6. A method for balancing fuel cell life and performance as claimed in claim 5, characterized in that: The method of constructing the performance and life balance model includes: Defining the balance between fuel cell performance and life , and its calculation formula is ; In the formula, is the short-term performance efficiency, which is calculated as: ; is the life loss rate; is the adjustment cost of the operating conditions, and the calculation formula is: ,in, , and The adjustment cost weights of the corresponding parameters are usually determined by providing experimental data; , and is the weight coefficient of the corresponding parameter, .

7. A method for balancing fuel cell life and performance as claimed in claim 6, characterized in that: The life loss rate The way to obtain is: The fused feature vector Input into the pre-trained life loss model to predict the life loss rate ; The training methods of the life loss model include: Define a recursive neural network as the basic structure of the life loss model, which includes an input layer, a hidden layer, and an output layer; Collect the frequency domain characteristics and multi-source data of multiple fuel cells in the past fixed time, and construct the corresponding fusion feature vector, which is recorded as the historical fusion feature vector ,in Represents a vector time step; fuses the history with the feature vector Expand by time step to form sequence data as the training input of the life loss model, and label the sequence data according to the vector time step; the label is the life state index of the fuel cell; the labeling range of the life state index is 0 to 100%; Initialize the network parameters of the life loss model; the network parameters include the weight matrix from the input layer to the hidden layer , the recurrent weight matrix of the hidden layer , the bias vector of the hidden layer , the weight matrix from hidden layer to output layer and the bias vector of the output layer ; Define the loss function for the lifetime loss model ;in, is the length of the sequence data; is the loss weight function weighted according to the life state; It is A vector of life state indices for time steps; is the quantile loss function; It is The life state index is annotated with a vector of time steps; It is The predicted values ​​of the life loss model for vector time steps; is the quantile level; ;in, is the indicative function; The threshold interval of the preset life state, the upper limit of the threshold interval is ; The lower limit of the threshold interval is ;like Less than ,but for ;like If it is within the threshold range, is 1; if Greater than ,but for ;in, and is the preset weight coefficient, and greater than 1, Less than 1; The sequence data is passed to the input layer, and then passes through the hidden layer and the output layer in turn to output the predicted value; the value of the corresponding loss function is calculated, and the error gradient is back-propagated from the output layer to the input layer; according to the error gradient, the network parameters are updated using the optimization algorithm; the sequence data is repeatedly trained until the life loss model converges or reaches the preset number of iterations, that is, the training of the life loss model is completed.

8. A method for balancing fuel cell life and performance as claimed in claim 7, characterized in that: The input layer is used to receive the fused feature vector as input; the hidden layer consists of It is composed of neural units; The calculation formula for each neural unit is: ;in, is the current vector time step The hidden state vector of is the hidden state vector of the previous vector time step; is the activation function; is the weight matrix from the input layer to the hidden layer; is the recurrent weight matrix of the hidden layer; is the bias vector of the hidden layer; The output layer calculates the predicted value based on the output of the hidden layer ;in, is the activation function of the output layer; is the weight matrix from the hidden layer to the output layer; is the bias vector of the output layer.

9. A method for balancing fuel cell life and performance as claimed in claim 8, characterized in that: The method of constructing the fuel cell group control model is: Defining load distribution weights The calculation formula is: ; Indicates The load weight ratio that should be allocated to each single fuel cell; It is The balance index of each fuel cell monomer; It represents the sum of the balance indexes of all the individual fuel cells in the fuel cell stack. is the total number of single fuel cells, is the index of the fuel cell; Defining the actual load distribution The calculation formula is: ; In the formula is the total output probability of the battery pack, and the calculation formula is: ,in and are the total output voltage and total output current of the fuel cell group respectively; The update formula that defines the consistency control parameters includes: ； In the formula, For fuel cells in actual load distribution The actual current under For the The operating voltage of each fuel cell; For the The internal resistance of a fuel cell; is the voltage drop in the ohmic region; is the power loss caused by the internal resistance of the fuel cell; For the The actual energy utilization efficiency of a fuel cell; is the input energy of the single fuel cell, and its calculation formula is: ,in is the Gibbs free energy of the reaction; Calculate the distribution weight difference of a single fuel cell , the calculation formula is: .

10. A method for balancing fuel cell life and performance as claimed in claim 9, characterized in that: The updated load distribution weight The constraints are: ; Excess load distribution The constraints are: .

Images