Real-time prediction method for performance degradation of vehicle-mounted hydrogen fuel cell

Abstract

The invention discloses a real-time prediction method for performance degradation of a vehicle-mounted hydrogen fuel cell, and relates to the technical field of vehicle fuel cells. The method comprises the following steps: establishing a working condition identification model based on a learning vector quantization neural network, and identifying four working condition types of variable load, start / stop, idle speed and high load; establishing an applicable improved real-time rain flow counting method; calculating a variable load cycle index, a start / stop cycle index, idling time and high load time based on an improved real-time rain flow counting method; and calculating the performance degradation rate of the fuel cell to obtain a life prediction result. According to the invention, the performance degradation rate of the hydrogen fuel cell is predicted in real time according to the automobile driving condition, and the engineering application difficulty of the performance degradation prediction method is reduced; through the application of the method, the fuel cell performance attenuation index of the fuel cell hybrid electric vehicle is established, real-time optimization of vehicle energy management is facilitated, hydrogen consumption is reduced, and the service life of the fuel cell is prolonged.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle fuel cells, and in particular, relates to a real-time prediction method for performance attenuation of a vehicle-mounted hydrogen fuel cell. Background Art

[0002] Proton exchange membrane fuel cells (PEMFCs) fueled by hydrogen are gradually maturing. They have the advantages of high energy conversion efficiency and truly zero emission and pollution-free, making them the most promising new energy system in the future, especially in the automotive field, which is gradually gaining more and more widespread applications. At present, the life of hydrogen fuel cells as a vehicle energy source is usually only 2500-3000 hours, but passenger cars and commercial vehicles require their lifespans to reach at least 5000 hours and 10000 hours respectively to meet the use requirements. Therefore, many scholars have been committed to the study of fuel cell life, involving factors affecting life, durability, and life (performance degradation) prediction methods. The factors affecting the life of automotive fuel cells are numerous and complex. Its dynamic response ability is one of the key factors, which is mainly affected by the structural parameters and operating parameters of the fuel cell. The study of fuel cell performance degradation prediction methods is of great significance to improving its durability and has become one of the most concerned areas.

[0003] In existing research, there are mainly model-driven prediction methods and data-driven prediction methods. The former requires the attenuation mechanism of the fuel cell, but due to the complexity of its attenuation process, it is usually difficult to describe it with an accurate mathematical model. The latter obtains a large amount of fuel cell data based on experimental measurements, converts the data into a corresponding model, and predicts its future performance attenuation. The methods involved are mostly based on modeling the factors that affect the performance attenuation of fuel cells under constant driving conditions, but in fact, on-board fuel cells have been working in dynamic variable load conditions. The vehicle's variable load, start / stop, idling and high load all have a great impact on its life. Therefore, these methods still have shortcomings in achieving dynamic real-time predictions.

[0004] As a data-driven prediction method, the rain flow counting method is mainly used in the engineering field, especially in the calculation of life loss. This method counts the nonlinear relationship between the stress-strain of the object, counts the number of full load cycles according to the established counting rules, and records the maximum and minimum values ​​of each cycle, calculates the amplitude and mean of the full cycle, and constructs a three-dimensional load spectrum of amplitude, mean, and number of cycles. The load spectrum is used to analyze the degree of damage to the object and predict the life of the object. The traditional rain flow counting method has the shortcomings of not being able to count all full cycles in the stress-strain response and requiring all experimental data for calculation, and lacks real-time performance. How to establish a real-time rain flow counting method and apply it to the prediction of fuel cell performance attenuation is a new method that is worth exploring.

[0005] In view of this, the present invention is proposed. Summary of the invention

[0006] The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a real-time prediction method for the performance attenuation of a vehicle-mounted hydrogen fuel cell, thereby solving the problems raised in the above-mentioned background technology.

[0007] In order to solve the above technical problems, the basic concept of the technical solution adopted by the present invention is:

[0008] A real-time prediction method for vehicle-mounted hydrogen fuel cell performance degradation, comprising:

[0009] S1: According to the characteristic parameters of driving conditions, a working condition recognition model based on learning vector quantization neural network is established to identify four working condition categories: variable load, start / stop, idling and high load;

[0010] S1.1: Determine the operating condition category. According to the main driving conditions that affect the performance degradation of the fuel cell, select four operating condition categories: variable load, start / stop, idling and high load;

[0011] S1.2: Determine the driving condition characteristic parameters and select 11 driving condition characteristic parameters;

[0012] S1.3: Construct a neural network based on learning vector quantization for working condition identification, set up a three-layer network of input layer, competition layer and linear output layer to achieve the purpose of working condition identification; the input layer receives the working condition category label and 11 characteristic parameter values, and the competition layer establishes the relationship between the characteristic parameters of each short-trip segment used for training and the working condition category to which the segment belongs through learning; the linear output layer converts the category learned by the competition layer into a defined classification target, that is, outputs the category of the driving condition, so as to achieve the purpose of working condition identification;

[0013] S1.4: Determine the learning algorithm, and select the LVQ2 algorithm that activates two competitive layer neurons at the same time as the learning algorithm for the working condition recognition model based on the learning vector quantization neural network;

[0014] S1.5: Use the operating condition characteristic parameters to perform network training to obtain an operating condition recognition model based on a learning vector quantization neural network.

[0015] S2: Establish an improved real-time rainflow counting method applicable to four types of working conditions;

[0016] S2.1: Use the "three-point method" to determine the peak or valley value, and extract three data Y(n-1), Y(n), and Y(n+1) of a certain process in real time. If (Y(n)-Y(n-1))×(Y(n)-Y(N+1))>0, then Y(n) is determined to be a peak or valley value;

[0017] S2.2: Three areas are set to store peak and valley values ​​to meet the requirements of real-time calculation; the three areas are the middle area, the maximum area and the minimum area to store data, the middle area stores the peak value or valley value obtained in step S2.1, when the middle area stores at least 2 data, they are compared in order, the larger data is stored in the maximum area, and the smaller data is stored in the minimum area;

[0018] S2.3: The peak-to-valley comparison method is combined with the multi-range counting method to extract the fluctuation amplitude, mean value and number of cycles; the "four-point method" peak-to-valley comparison method is used, and 4 peak-to-valley values ​​are used for each calculation. The range is the absolute value of the difference between two adjacent peak-to-valley values, which is essentially the fluctuation amplitude, and multiple ranges are used for counting;

[0019] S2.3.1: Extract four adjacent peak-to-valley values ​​from the data region in step S2.2, calculate the pairwise difference, and take their absolute values ​​to obtain the three ranges of the adjacent sequence;

[0020] S2.3.2: Determine whether the range of the intermediate sequence is the minimum value. If so, extract the range, count a full cycle number, and calculate the mean value corresponding to the range;

[0021] S2.3.3: The remaining two ranges in step S2.3.2 are combined into a new range, and a new round of calculation is performed with the new ranges of the next two sequences obtained by the new sampling values;

[0022] S2.3.4: In step S2.3.2, when the range of the intermediate sequence does not meet the minimum condition, the first range is temporarily stored, and the second and third ranges are compared with the new fourth range. If the minimum condition is met, the above process is followed, and the second and third ranges are combined into a new range, which is compared with the temporarily stored first range and the fifth range taken out;

[0023] S2.3.5: Repeat the above steps and record the fluctuation amplitude, mean value and number of cycles for subsequent performance degradation prediction calculations.

[0024] S3: Calculate the number of load-changing cycles, the number of start / stop cycles, the idling time and the high-load time according to the established improved real-time rainflow counting method; use the improved real-time rainflow counting method to determine the corresponding relationship between the real-time driving speed of each working condition and the fluctuation amplitude, mean value and number of cycles of the rainflow counting method; use the improved real-time rainflow counting method to record the number of cycles of the four types of working conditions and calculate the working condition cycle time;

[0025] S4: Calculate the performance decay rate of the fuel cell according to the number of load-changing cycles, the number of start / stop cycles, the idle time and the high-load time to obtain a battery life prediction result;

[0026] The formula for calculating the performance decay rate of a fuel cell is as follows:

[0027]

[0028] Wherein, ΔV is the maximum voltage decay value allowed for the proton exchange membrane fuel cell; K μ is the correction factor for different proton exchange membrane fuel cell systems; μ 1 , μ 2 , μ 3 , μ 4 are the performance attenuation rates under variable load, start / stop, idling and high load conditions respectively; n 1 、n 2 ,t 1 ,t 2 They are the number of cycles of variable load condition, the number of cycles of start / stop condition, the idling condition time and the high load condition time;

[0029] In step S3, n is recorded 1 、n 2 ,t 1 ,t 2 , calculate the performance decay rate of the fuel cell and obtain its life prediction result.

[0030] After adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art. Of course, any product implementing the present invention does not necessarily need to achieve all the advantages described below at the same time:

[0031] 1. The present invention provides a new real-time rain flow counting method, which can use less historical data, adopt real-time data collection, cyclic calculation, and obtain the fluctuation amplitude, mean value and number of cycles of relevant physical quantities in real time, laying a foundation for various applications. This method avoids the shortcomings of traditional methods such as using all historical data for calculation and poor real-time performance;

[0032] 2. The applied learning vector quantization neural network operating condition prediction model and improved real-time rain flow counting method can realize the real-time prediction of hydrogen fuel cell performance attenuation rate according to the vehicle driving conditions, which greatly reduces the engineering application difficulty of the fuel cell performance attenuation prediction method;

[0033] 3. Through the application of the present invention, a fuel cell performance attenuation performance index for fuel cell hybrid vehicles can be established, which is beneficial to the real-time optimization of energy management of such vehicles, achieving optimal energy distribution of various power sources such as fuel cells, reducing hydrogen consumption, and improving the fuel economy of the entire vehicle; at the same time, real-time optimization of energy is beneficial to extending the service life of the fuel cell.

[0034] The specific implementation modes of the present invention are further described in detail below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings described below are only some embodiments. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0036] In the figure:

[0037] Figure 1 This is the overall framework diagram of the real-time prediction method for hydrogen fuel cell performance degradation;

[0038] Figure 2 This is the overall framework diagram of the neural network based on learning vector quantization;

[0039] Figure 3 To improve the real-time rainflow counting method flow chart;

[0040] Figure 4 A flow chart of a method for real-time prediction of hydrogen fuel cell performance degradation;

[0041] Figure 5 This is a topological diagram of a hydrogen fuel cell hybrid vehicle.

[0042] It should be noted that these drawings and textual descriptions are not intended to limit the conceptual scope of the present invention in any way, but are intended to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0043] The present invention will now be described in further detail with reference to the accompanying drawings.

[0044] See also Figure 1-5 As shown, in this embodiment, a method for real-time prediction of vehicle-mounted hydrogen fuel cell performance degradation is provided, including:

[0045] S1: According to the characteristic parameters of driving conditions, a working condition recognition model based on learning vector quantization neural network is established to identify four working condition categories: variable load, start / stop, idling and high load;

[0046] S1.1: Determine the operating condition category. According to the main driving conditions that affect the performance degradation of the fuel cell, select four operating condition categories: variable load, start / stop, idling and high load. Among them, according to the established short-trip rules, the vehicle driving data is subjected to noise reduction, filling, interpolation and other operations, and the overspeed, abnormal acceleration and deceleration, idling and other conditions are processed to obtain effective driving data for short-trip segment division and segment feature parameter calculation. The short-trip segments are classified using a combination of principal component analysis and K-means cluster analysis, and finally four operating condition categories are obtained;

[0047] S1.2: Determine the driving condition characteristic parameters and select 11 condition characteristic parameters as the necessary prerequisite for establishing a high-performance condition identification model;

[0048] In this embodiment, 11 operating condition characteristic parameters are selected from the 22 operating condition characteristic parameters in the whole set, as shown in Table 1.

[0049] Table 1 Parameters and meanings of optimal subsets of driving condition characteristic parameters

[0050]

[0051] S1.3: Construct a neural network based on learning vector quantization for working condition identification, set up a three-layer network of input layer, competition layer and linear output layer to achieve the purpose of working condition identification; the input layer receives the working condition category label and 11 characteristic parameter values, and the competition layer establishes the relationship between the characteristic parameters of each short-trip segment used for training and the working condition category to which the segment belongs through learning; the linear output layer converts the category learned by the competition layer into a defined classification target, that is, outputs the category of the driving condition, so as to achieve the purpose of working condition identification;

[0052] The neural network based on learning vector quantization selected in this embodiment is as follows Figure 2 As shown, P is the R-dimensional input vector; S 1 is the number of neurons in the competition layer; S 2 is the number of neurons in the linear output layer; IW1.1 is the number of neurons between the input layer and the competition layer 1 ×R-dimensional weight matrix; n 1 For the competition layer S 1 ×1-dimensional input; a 1 For the competition layer S 1 ×1-dimensional output; LW2.1 is the S1×S2-dimensional weight matrix between the competition layer and the linear output layer; n 2 is the linear output layer S 1 ×1-dimensional input; a 2 is the linear output layer S 1 ×1-dimensional output.

[0053] S1.4: Determine the learning algorithm, and select the LVQ2 algorithm that simultaneously activates two competitive layer neurons as the learning algorithm for the working condition recognition model based on the learning vector quantization neural network; Since the LVQNN learning algorithm combines the competitive learning algorithm and the supervised learning algorithm, in theory, the more neurons are activated at the same time for weight adjustment, the higher the training efficiency of the neural network can be. Therefore, the LVQ2 algorithm that simultaneously activates two competitive layer neurons is selected as the learning algorithm for the working condition recognition model based on the learning vector quantization neural network;

[0054] S1.5: Use the operating condition characteristic parameters to perform network training to obtain an operating condition recognition model based on a learning vector quantization neural network; randomly extract 20% of the speed fragments from each type of operating condition data as the training data set, and the remaining 80% of the speed fragments as the test data set. Through training and testing, a high-precision operating condition recognition model based on a learning vector quantization neural network is finally obtained.

[0055] S2: Establish an improved real-time rainflow counting method applicable to four types of working conditions;

[0056] S2.1: Use the "three-point method" to determine the peak or valley value, extract three data Y(n-1), Y(n), and Y(n+1) of a certain process in real time, and if (Y(n)-Y(n-1))×(Y(n)-Y(N+1))>0, then determine that Y(n) is the peak or valley value; for driving conditions, mainly collect three vehicle driving speed data in real time, and use the "three-point method" to determine the peak or valley value;

[0057] S2.2: Three areas are set to store peak and valley values ​​to meet the requirements of real-time calculation; the three areas are the middle area, the maximum area and the minimum area to store data, the middle area stores the peak value or valley value obtained in step S2.1, when the middle area stores at least 2 data, they are compared in order, the larger data is stored in the maximum area, and the smaller data is stored in the minimum area;

[0058] S2.3: The peak-to-valley comparison method is combined with the multi-range counting method to extract the fluctuation amplitude, mean value and number of cycles; the "four-point method" peak-to-valley comparison method is used, and 4 peak-to-valley values ​​are used for each calculation. The range is the absolute value of the difference between two adjacent peak-to-valley values, which is essentially the fluctuation amplitude, and multiple ranges are used for counting;

[0059] S2.3.1: Extract four adjacent peak-to-valley values ​​from the data region in step S2.2, calculate the pairwise difference, and take their absolute values ​​to obtain the three ranges of the adjacent sequence;

[0060] S2.3.2: Determine whether the range of the intermediate sequence is the minimum value. If so, extract the range, count a full cycle number, and calculate the mean value corresponding to the range;

[0061] S2.3.3: The remaining two ranges in step S2.3.2 are combined into a new range, and a new round of calculation is performed with the new ranges of the next two sequences obtained by the new sampling values;

[0062] S2.3.4: In step S2.3.2, when the range of the intermediate sequence does not meet the minimum condition, the first range is temporarily stored, and the second and third ranges are compared with the new fourth range. If the minimum condition is met, the above process is followed, and the second and third ranges are combined into a new range, which is compared with the temporarily stored first range and the fifth range taken out;

[0063] S2.3.5: Repeat the above steps and record the fluctuation amplitude, mean value and number of cycles for subsequent performance degradation prediction calculations.

[0064] S3: The steps of calculating the number of load-changing cycles, the number of start / stop cycles, the idle time and the high-load time according to the established improved real-time rain flow counting method are as follows;

[0065] Determine the corresponding relationship between the real-time driving speed of each working condition and the fluctuation amplitude, mean value and cycle number of the rain flow counting method; apply the improved real-time rain flow counting method of step S2 to record the cycle number of the four types of working conditions. After setting the sampling period, such as 1s in this embodiment, the cycle time of various working conditions can be accumulated, and finally the cycle number and cycle time of the four working conditions are obtained;

[0066] S4: Calculate the performance decay rate of the fuel cell according to the number of load-changing cycles, the number of start / stop cycles, the idle time and the high-load time to obtain a battery life prediction result;

[0067] The formula for calculating the performance decay rate of a fuel cell is as follows:

[0068]

[0069] Wherein, ΔV is the maximum voltage decay value allowed for the proton exchange membrane fuel cell; K μ is the correction factor for different proton exchange membrane fuel cell systems; μ 1 , μ 2 , μ 3 , μ 4 are the performance attenuation rates under variable load, start / stop, idling and high load conditions respectively; n 1 、n 2 ,t 1 ,t 2 They are the number of cycles of variable load condition, the number of cycles of start / stop condition, the idling condition time and the high load condition time;

[0070] In step S3, n is recorded 1 、n 2 ,t 1 ,t 2 , calculate the performance decay rate of the fuel cell and obtain its life prediction result.

[0071] In this embodiment, the fuel cell voltage attenuation values ​​under various operating conditions obtained from experiments provided in existing literature are used for calculation, as shown in Table 2.

[0072] Table 2 Fuel cell voltage attenuation values ​​under various working conditions

[0073] Driving conditions symbol Attenuation Variable load <![CDATA[μ 1 ]]> 0.00126uV / time Start and stop <![CDATA[μ 2 ]]> 0.00196uV / time Idle <![CDATA[μ 3 ]]> 0.0000593uV / h High load <![CDATA[μ 4 ]]> 0.00147uV / h

[0074] The embodiment of the present invention is based on the power system of a hydrogen fuel cell hybrid vehicle equipped with three energy sources, and the actual reference object is a four-wheel vehicle, whose power system topology is as follows: Figure 5 As shown in Table 3, the parameters of the three-energy fuel cell hybrid vehicle and its main components are shown in Table 3.

[0075] Table 3 Fuel cell vehicles and their main basic parameters

[0076]

[0077] like Figure 5 As shown, the power system assembly of a hydrogen fuel cell hybrid vehicle can be divided into two parts: an energy source subsystem and a traction subsystem, wherein the control strategy of the energy source subsystem includes an energy management strategy and a control strategy of a direct current / direct current (DC / DC) converter, and the control method of the traction subsystem implements control on an inverter (DC / AC) to achieve the goal of controlling the traction subsystem, including regenerative braking control of a fuel cell vehicle, and the energy management strategy serves as an upper-level control to allocate output power and / or absorb braking feedback power to three energy sources, and accordingly implement corresponding control on the DC / DC converter. The present invention is mainly aimed at real-time prediction of the performance attenuation rate of a hydrogen fuel cell during vehicle driving.

[0078] The following is an analysis of the present invention using dual fuzzy controllers with different parameters to implement energy management for hydrogen fuel cell hybrid vehicles under the World Harmonized Light Vehicle Test Cycle (WLTC), and real-time calculation of the hydrogen fuel cell performance decay rate during simulation operation. Six fuzzy controllers with different parameters are given, marked as ①-⑥, and the real-time calculation results are shown in Table 4.

[0079] Table 4 Real-time calculation results of hydrogen fuel cell performance attenuation rate

[0080]

[0081] The feasibility of the present invention is verified through comparative analysis of the above results, and it is not difficult to find that:

[0082] (1) The embodiment of the present invention can calculate the performance attenuation rate of the hydrogen fuel cell in real time during the real-time simulation driving of the car;

[0083] (2) Implementing vehicle energy management under fuzzy controllers with different parameters has a significant impact on the performance degradation of hydrogen fuel cells, and the decay rate value changes significantly. It can be inferred that reasonable and optimized power allocation among vehicle energy sources can not only reduce vehicle energy consumption, but also reduce the performance degradation of hydrogen fuel cells, which is beneficial to extending the service life of hydrogen fuel cells.

[0084] The present invention is not limited to the above-mentioned embodiments. Anyone should be aware that any structural changes made under the enlightenment of the present invention, and any technical solutions that are the same or similar to the present invention, fall within the protection scope of the present invention. The technology, shape, and structural parts not described in detail in the present invention are all well-known technologies.

Claims

1. A real-time prediction method for vehicle-mounted hydrogen fuel cell performance attenuation, characterized in that: include: S1: According to the characteristic parameters of driving conditions, a working condition recognition model based on learning vector quantization neural network is established to identify four working condition categories: variable load, start / stop, idling and high load; S2: Establish an improved real-time rainflow counting method applicable to four types of working conditions; S3: Calculate the number of load-changing cycles, the number of start / stop cycles, the idle time and the high-load time according to the established improved real-time rainflow counting method; S4: Calculate the performance decay rate of the fuel cell according to the number of variable load cycles, the number of start / stop cycles, the idling time, and the high load time to obtain a battery life prediction result.

2. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 1 is characterized in that: Step S1 specifically includes the following steps: S1.1: Determine the operating condition category. According to the main driving conditions that affect the performance degradation of the fuel cell, select four operating condition categories: variable load, start / stop, idling and high load; S1.2: Determine the driving condition characteristic parameters and select 11 driving condition characteristic parameters; S1.3: Construct a neural network based on learning vector quantization for working condition identification, set up a three-layer network of input layer, competition layer and linear output layer to achieve the purpose of working condition identification; S1.4: Determine the learning algorithm, and select the LVQ2 algorithm that activates two competitive layer neurons at the same time as the learning algorithm for the working condition recognition model based on the learning vector quantization neural network; S1.5: Use the operating condition characteristic parameters to perform network training to obtain an operating condition recognition model based on a learning vector quantization neural network.

3. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 2 is characterized in that: In step S1.3, the input layer receives the operating condition category label and 11 feature parameter values. The competitive layer establishes the relationship between the feature parameters of each short-trip segment used for training and the operating condition category to which the segment belongs through learning. The linear output layer converts the category learned by the competitive layer into a defined classification target, that is, outputs the category of the driving condition to achieve the purpose of operating condition identification.

4. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 1 is characterized in that: Step S2 specifically includes the following steps: S2.1: Use the "three-point method" to determine the peak or valley value, and extract three data Y(n-1), Y(n), and Y(n+1) of a certain process in real time. If (Y(n)-Y(n-1))×(Y(n)-Y(N+1))>0, then Y(n) is determined to be a peak or valley value; S2.2: Set three areas to store peak and valley values ​​to meet the requirements of real-time calculation; S2.3: The peak-to-valley comparison method is combined with the multi-variable counting method to extract the fluctuation amplitude, mean value and number of cycles.

5. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 4 is characterized in that: In step S2.2, three areas are used to store data, namely the middle area, the maximum area and the minimum area. The middle area stores the peak value or valley value obtained in step S2.

1. When the middle area stores at least two data, they are compared in sequence. The larger data is stored in the maximum area, and the smaller data is stored in the minimum area.

6. A method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 4, characterized in that: In step S2.3, the "four-point method" peak-to-valley value comparison method is adopted, and 4 peak-to-valley values ​​are used for each calculation. The range is the absolute value of the difference between two adjacent peak-to-valley values, which is essentially the fluctuation amplitude, and multiple ranges are used for counting.

7. A method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 6, characterized in that: Step S2.3 specifically includes the following steps: S2.3.1: Extract four adjacent peak-to-valley values ​​from the data region in step S2.2, calculate the pairwise difference, and take their absolute values ​​to obtain the three ranges of the adjacent sequence; S2.3.2: Determine whether the range of the intermediate sequence is the minimum value. If so, extract the range, count a full cycle number, and calculate the mean value corresponding to the range; S2.3.3: The remaining two ranges in step S2.3.2 are combined into a new range, and a new round of calculation is performed with the new ranges of the next two sequences obtained by the new sampling values; S2.3.4: In step S2.3.2, when the range of the intermediate sequence does not meet the minimum condition, the first range is temporarily stored, and the second and third ranges are compared with the new fourth range. If the minimum condition is met, the above process is followed, and the second and third ranges are combined into a new range, which is compared with the temporarily stored first range and the fifth range taken out; S2.3.5: Repeat the above steps and record the fluctuation amplitude, mean value and number of cycles for subsequent performance degradation prediction calculations.

8. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 1 is characterized in that: In step S3, the improved real-time rain flow counting method is used to determine the corresponding relationship between the real-time driving speed of each working condition and the fluctuation amplitude, mean value and number of cycles of the rain flow counting method.

9. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 1, characterized in that: In step S3, the improved real-time rain flow counting method is used to record the cycle times of the four types of working conditions and calculate the working condition cycle time.

10. The method for real-time prediction of vehicle-mounted hydrogen fuel cell performance attenuation according to claim 1, characterized in that: In step S4, the formula for calculating the performance decay rate of the fuel cell is as follows: Wherein, ΔV is the maximum voltage decay value allowed for the proton exchange membrane fuel cell; K μ is the correction factor for different proton exchange membrane fuel cell systems; μ1, μ2, μ3, and μ4 are the performance attenuation rates under variable load, start / stop, idle, and high load conditions, respectively; n1, n2, t1, and t2 are the number of cycles under variable load conditions, the number of cycles under start / stop conditions, the idle condition time, and the high load condition time, respectively; By recording n1, n2, t1, and t2 in step S3, the performance decay rate of the fuel cell is calculated to obtain a life prediction result.

Images