Decoupling method for advance control of fuel cell air system

Abstract

The present invention relates to a decoupling method for advance control of a fuel cell air system, which belongs to the field of fuel cell technology and comprises the following steps: S1: predicting the short-term power of a fuel cell engine by a timing prediction algorithm; S2: testing the required rotation speed of an air compressor and the required opening of a back-pressure valve under different pressures and flow rates; S3: obtaining the change data of the pressure and flow of an air system under the back-pressure valve opening and the air compressor rotation speed within a certain range through bench experiments, so as to provide data for decoupling control; S4: identifying the transfer function of the air system under a data-driven method, and then designing a corresponding decoupling device in combination with a diagonal matrix decoupling method; S5: obtaining the required air compressor rotation speed and back-pressure valve opening by a table lookup method for the predicted value of the short-term power; S6: analyzing the decoupling effect of the air system by the relative control errors of pressure and flow rate, and analyzing the fluctuation change of the actuator by the total change value of the actuator.

Description

Technical Field

[0001] The invention belongs to the technical field of fuel cells and relates to a decoupling method for advance control of an air system of a fuel cell. Background Art

[0002] The air system of a fuel cell can be considered its "lungs", where the key control parameters include intake flow and intake pressure. Effective control of intake flow helps prevent oxygen deficiency, while stable intake pressure can slow down pressure fluctuations in the stack, thereby extending the service life of the proton exchange membrane. Existing technologies cannot meet the fuel cell's high-precision control requirements for cathode intake flow and pressure, as well as its strict requirements for environmental adaptability, and it is difficult to control intake flow and pressure in advance on a time scale. Summary of the invention

[0003] In view of this, an object of the present invention is to provide a decoupling method for advance control of a fuel cell air system.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A decoupling method for advance control of a fuel cell air system comprises the following steps:

[0006] S1: Design a time series prediction algorithm to achieve short-term power prediction of fuel cell engines; use sliding window analysis to perform feature engineering on power demand time series data, combine phase space reconstruction theory to construct the input and output vector space of the recursive prediction model based on recursive least squares support vector machine, use autocorrelation analysis to determine the embedding dimension of phase space reconstruction, and propose a short-term power prediction algorithm;

[0007] S2: Test the required speed of the air compressor and the required opening of the back pressure valve under different pressures and flows on the test bench to provide data support for feedforward control;

[0008] S3: Through bench experiments, obtain the change data of air system pressure and flow rate under the back pressure valve opening and air compressor speed within a certain range to provide data for decoupling control;

[0009] S4: Using a data-driven approach, the transfer function of the air system is identified, and the corresponding solver is designed in combination with the diagonal matrix decoupling method;

[0010] S5: The predicted value of short-term power is used to obtain the required air compressor speed and back pressure valve opening by table lookup method;

[0011] S6: The decoupling effect of the air system can be analyzed through the relative control errors of pressure and flow, and the fluctuation of the actuator can be analyzed through the total change value of the actuator.

[0012] Further, step S1 specifically includes the following steps:

[0013] S11: Assume that the time series of the fuel cell vehicle power demand with N data points is [p1, p2, p3, …, p N ] T , select the embedding dimension m and time delay τ, and construct the input vector x t =[p t-(m-1)τ …, p t-2τ , p t-τ , p t ]∈R m ; The embedding dimension refers to the number of historical time series data points input, and the time delay refers to the time interval between the input historical time series data points; the corresponding predicted output variable is y t+h =[p t+h ]∈R 1 , where h is the prediction step length, called h-step ahead prediction. The recursive prediction model of the demand power time series of h-step ahead prediction is as follows:

[0014]

[0015] Where f(·) is a generalized smooth nonlinear mapping;

[0016] S12: Based on the phase space reconstruction method, the demand power time series prediction input and output sample pairs are established when the sliding window width L = 9, the embedding dimension m = 5, the time delay τ = 2 and the sliding period T = 4, as shown in the following formula:

[0017]

[0018] In the formula, are the input matrix and the expected output matrix after reconstructing the phase space, respectively. The required power time series is represented by [p1, p2, p3, …, p N ] T It means that the input sample matrix after reconstructing the phase space is used as the input of the prediction model F(·) to estimate the predicted data value at the future moment;

[0019] S13: The formula for calculating the autocorrelation coefficient is defined as:

[0020]

[0021] Wherein, the lag number k = 1, 2, ...; the autocorrelation coefficient is obtained by converting the time series data y = [y1, y2, y3, ..., y N ] T Two sample point sequences y constructed according to the lag number k t =[y 1+k,y 2+k ,y 3+k , …, y N ] T and t-k =[y1,y2,y3,…,y N-k ] T degree of relevance.

[0022] Furthermore, an iterative learning framework is constructed to replace the one-time prediction of a long time span with several iterative predictions with shorter time steps. Based on the theory of reconstructed embedded phase space, the training data sets of the learning elements used to predict power and power change rate in the xth iteration are constructed respectively. and As shown in the following formula:

[0023] [P(t+β);P(t+2β);…;P(t+xβ);P′(t+β);P′(t+2β);…;P′(t+xβ)]=f(P(t),P′(t),P″(t))

[0024] Where f(·) represents the abstract nonlinear mapping of the trained ILF iterative learning framework.

[0025] Further, step S3 specifically includes the following steps:

[0026] S31: Build and debug the test bench;

[0027] S32: Setting the amplitude range of the air compressor speed M sequence and the back pressure valve opening;

[0028] S33: Considering the identification algorithm, the sampling period of the control system and the sensitivity of the sensor at the same time, the sampling period of the M sequence is selected and the sampling time is set.

[0029] Further, the construction and debugging of the test bench in step S31 specifically includes:

[0030] (1) Determine the topology of the test bench, determine the component selection and pipeline design;

[0031] (2) Make mechanical and electrical connections according to the topology and test the mechanical and electrical connections;

[0032] (3) Calibrate each sensor;

[0033] (4) Test the status of the air compressor and back pressure valve;

[0034] (5) Calibrate the opening of each ball valve according to the pressure drop of the fuel cell stack, intercooler, and humidifier;

[0035] (6) Bench test the system under the full range of operating conditions.

[0036] Further, in step S33, the sampling time is set to:

[0037]

[0038] T0 is the sampling period, T 95 It is the response time for the controlled object to reach 95% of the target value. The denominator is set to any constant in the range of 5 to 15 according to the actual situation.

[0039] Further, step S4 specifically includes the following steps:

[0040] S41: within a certain range of compressor speed, select multiple equilibrium operating points for system identification;

[0041] S42: The input-output relationship of the system is described as a first-order inertia link. The model structure of the air system is expressed by the following formula:

[0042]

[0043] in is the increment of air flow, Δp is the increment of pressure, Δn is the increment of compressor speed, and Δγ is the increment of back pressure valve opening.

[0044] S43: Identify the dual-input and dual-output transfer function of the air system through a data-driven method;

[0045] S44: By adding matrices to the control system, a diagonal matrix decoupling controller is constructed, including: Multiply with the object feature matrix The resulting generalized object matrix becomes a diagonal matrix, thus achieving system decoupling:

[0046]

[0047] When the mathematical model matrix of the controlled object is non-singular, The decoupling compensation matrix is ​​written as:

[0048]

[0049] Further, step S6 specifically includes:

[0050] The total variation of the actuators including the air compressor and the back pressure valve is analyzed simultaneously using the mean square of the absolute error of flow and pressure as shown below:

[0051]

[0052] Where Pressure(t) and Flow(t) represent the pressure and flow at a certain moment, and Pressure_MAE, Flow_MAE, and Pressure_MAE represent the average values ​​of pressure and flow. The better the decoupling effect of the decoupling strategy, the smaller the Flow_MAE and Pressure_MAE, the smaller the Tv, and the better the stability and anti-interference performance.

[0053] The beneficial effects of the present invention are: the fuel cell air system advance decoupling method proposed in the present invention and the design of the timing prediction algorithm can realize the short-term power prediction of the fuel cell engine, and the advance control of the pressure flow can be realized by obtaining the air system flow pressure, air compressor speed and back pressure valve opening data. The diagonal matrix decoupling method can alleviate the coupling of the fuel cell air system. The above method can further improve the stability of the fuel cell air system and the system service life.

[0054] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0056] Figure 1 Check the representation for feedforward control;

[0057] Figure 2 A short-term prediction method for power;

[0058] Figure 3 Schematic diagram of decoupling design. DETAILED DESCRIPTION

[0059] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0060] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0061] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0062] The present invention provides a decoupling method for advance control of a fuel cell air system, comprising the following steps:

[0063] S1: Design a time series prediction algorithm to achieve short-term power prediction of fuel cell engines. First, use sliding window analysis to perform feature engineering on the power demand time series data, and combine the phase space reconstruction theory to construct the input and output vector space of the recursive prediction model based on the machine learning algorithm (recursive least squares support vector machine). Finally, use autocorrelation analysis to determine the embedding dimension of phase space reconstruction and propose a short-term power prediction algorithm; including the following steps:

[0064] S11: Assume that the time series of the fuel cell vehicle power demand with N data points is [p1, p2, p3, …, p N ] T , choosing the embedding dimension m and the time delay τ, we can construct the input vector x t =[p t-(m-1)τ ,…,p t-2τ , p t-τ , p t ]∈R m The embedding dimension refers to the number of historical time series data points input, and the time delay refers to the time interval between the input historical time series data points. The corresponding predicted output variable is y t+h =[p t+h ]∈R 1 , where h is the prediction step length, which is usually called h-step ahead prediction. The figure shows the recursive prediction model of the demand power time series with h-step ahead prediction.

[0065]

[0066] Where f(·) is a generalized smooth nonlinear mapping.

[0067] S12: Based on the phase space reconstruction method, the demand power time series prediction input-output sample pair can be established when the sliding window width L=9, the embedding dimension m=5, the time delay τ=2 and the sliding period T=4, as shown in the following formula.

[0068]

[0069] In the formula, are the input matrix and expected output matrix after reconstructing the phase space. The input sample matrix after reconstructing the phase space is used as the input of the prediction model F(·) to estimate the predicted data value at the future moment.

[0070] S13: The autocorrelation coefficient can be understood as the time series data y=[y1, y2, y3, ..., y N ] T Two sample point sequences y constructed according to the lag number k t =[y 1+k ,y 2+k ,y 3+k , …, y N ] T and t-k =[y1,y2,y3,…,y N-k ]] T The autocorrelation coefficient calculation formula can be defined as:

[0071]

[0072] In the formula, the lag number k = 1, 2, ….

[0073] According to the results of autocorrelation analysis, when it is necessary to predict the future power demand p h sampling points backward relative to the current time t t+h and power change rate p′ t+h When , the phase space reconstruction embedding dimension m of the demand power and power change rate time series data should not be greater than 30-h and 50-h, respectively. The smaller the time range Δt of the one-step prediction, the smaller the error term μ pThe smaller (t, Δt), the lighter the time phase mismatch caused. Based on this, the present invention proposes an iterative learning framework (replacing the one-time prediction of a larger time span in the previous text by several iterative predictions with shorter time steps, for example, splitting the one-step prediction from point a to point f into multiple iterative predictions, that is, first predicting the data at point b, and then merging the predicted data at point b and the original input data at point a as new input to predict the next data at point c, and so on. Based on the theory of reconstructed embedded phase space, the training data sets of the learning elements used to predict power and power change rate in the xth iteration can be constructed respectively. and The following formula is shown.

[0074] [P(t+β);P(t+2β);…;P(t+xβ);P′(t+β);P′(t+2β);…;P′(t+xβ)]=f(P(t),P′(t),P″(t))

[0075] Where f(·) represents the abstract nonlinear mapping of the trained ILF iterative learning framework.

[0076] S2: Test the required speed of the air compressor and the required opening of the back pressure valve under different pressures and flows on the test bench to provide data support for feedforward control;

[0077] S3: Through bench experiments, obtain the change data of air system pressure and flow rate under the back pressure valve opening and air compressor speed within a certain range to provide data for decoupling control; step S3 specifically includes:

[0078] S31: The specific process of test bench construction and debugging is as follows: (1) Determine the topology of the test bench, determine the component selection and pipeline design; (2) Make mechanical and electrical connections according to the topology, and test the mechanical and electrical connections; (3) Calibrate each sensor; (4) Test the status of the air compressor and back pressure valve; (5) Calibrate the opening of each ball valve according to the pressure drop of the fuel cell stack, intercooler, and humidifier; (6) Test the system bench under full range operating conditions.

[0079] S32: To ensure that the identification input has a certain signal-to-noise ratio, its amplitude should not be too small. Combined with the actual situation of this article, the amplitude range of the air compressor speed M sequence is ±1000RPM, and the back pressure valve opening is ±10%.

[0080] S33: When selecting the sampling period of the M sequence, the identification algorithm, the sampling period of the control system and the sensitivity of the sensor should be considered at the same time. And the sampling time should be consistent with the actual application of the control algorithm as much as possible. The following empirical formula is generally used in engineering:

[0081]

[0082] T0 is the sampling period, T 95 is the response time for the control object to reach 95% of the target value. The sampling time of the existing air supply system controller is 0.01s, which satisfies the above formula. Therefore, in order to maintain consistency, the sampling time of the system identification test is determined to be 0.01s.

[0083] S4: Under the data-driven method, the transfer function of the air system is identified, and then the corresponding solver is designed in combination with the diagonal matrix decoupling method; step S4 specifically includes:

[0084] S41: In order to enhance the adaptability of the controller and obtain a more accurate decoupling coefficient, 10 equilibrium operating points are selected for system identification within the range of compressor speed from 0 to 90000 RPM;

[0085] S42: Considering the dynamic characteristics of the air supply system of the proton exchange membrane fuel cell and the principle of minimum order selection, the input and output relationship of the system is described as a first-order inertia link. Therefore, the model structure of the air system is expressed by the following formula:

[0086]

[0087] in is the increment of air flow, Δp is the increment of pressure, Δn is the increment of compressor speed, and Δγ is the increment of back pressure valve opening.

[0088] S43: Identify the dual-input and dual-output transfer function of the air system through a data-driven approach.

[0089] S44: By adding matrices to the control system, a diagonal matrix decoupling controller is proposed. Multiply with the object feature matrix The formed generalized object matrix becomes a diagonal matrix, thus achieving system decoupling.

[0090]

[0091] Assuming that the mathematical model matrix of the controlled object is non-singular, Therefore, the decoupling compensation matrix is ​​rewritten as:

[0092]

[0093] S5: The predicted value of short-term power is used to obtain the required air compressor speed and back pressure valve opening by table lookup method;

[0094] S6: The decoupling effect of the air system can be analyzed through the relative control errors of pressure and flow, and the fluctuation of the actuator can be analyzed through the total change value of the actuator (air compressor speed and back pressure valve opening).

[0095] In order to quantitatively analyze the simulation results, the square of the mean absolute error of flow and pressure is used. The total change of the actuator including the air compressor and the back pressure valve is analyzed at the same time, as shown in the following formula.

[0096]

[0097] Where Pressure(t) and Flow(t) represent the pressure and flow at a certain moment, and Pressure_MAE, Flow_MAE, and Pressure_MAE represent the average values ​​of pressure and flow. The better the decoupling effect of the decoupling strategy, the smaller the Flow_MAE and Pressure_MAE, the smaller the Tv, and the better the stability and anti-interference performance.

[0098] In the present invention, for the coupling of air system pressure and flow, according to the pressure-flow-speed table and pressure-flow-opening table tested on the bench, a feedforward control table is obtained based on the table lookup, as shown in FIG. Figure 1 As shown in the figure, the speed of the air compressor and the opening of the back pressure valve under different powers can be obtained, which is conducive to improving the dynamic response of the system; Figure 2 The timing prediction algorithm shown in the figure can realize the short-term power prediction of the fuel cell engine, which is conducive to the advance control of the air system and can effectively avoid the phenomenon of fuel shortage. The change data of the air system pressure and flow rate under the back pressure valve opening and air compressor speed within a certain range are obtained through bench test data, and then the double-input and double-output transfer function of the air system is identified. The decoupling controller designed is as follows Figure 3 As shown, the above method can realize the decoupled control of pressure and flow of the air system.

[0099] A person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing related hardware through a program. The program can be stored in a computer-readable storage medium. When the program is executed, the steps of the method can be implemented. The storage medium, such as ROM / RAM, a disk, an optical disk, etc.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A decoupling method for advance control of a fuel cell air system, characterized in that: The following steps are involved: S1: Design a timing prediction algorithm to achieve short-term power prediction of fuel cell engines; The sliding window analysis is used to perform feature engineering on the power demand time series data. The input and output vector spaces of the recursive prediction model based on recursive least squares support vector machine are constructed by combining the phase space reconstruction theory. The autocorrelation analysis is used to determine the embedding dimension of the phase space reconstruction, and a short-term power prediction algorithm is proposed. S2: Test the required speed of the air compressor and the required opening of the back pressure valve under different pressures and flows on the test bench to provide data support for feedforward control; S3: Through bench experiments, obtain the change data of air system pressure and flow rate under the back pressure valve opening and air compressor speed within a certain range to provide data for decoupling control; S4: Using a data-driven approach, the transfer function of the air system is identified, and the corresponding solver is designed in combination with the diagonal matrix decoupling method; Step S4 specifically includes the following steps: S41: within a certain range of compressor speed, select multiple equilibrium operating points for system identification; S42: The input-output relationship of the system is described as a first-order inertia link. The model structure of the air system is expressed by the following formula: in is the increment of air flow, Δp is the increment of pressure, Δn is the increment of compressor speed, and Δγ is the increment of back pressure valve opening; S43: Identify the dual-input and dual-output transfer function of the air system through a data-driven method; S44: By adding matrices to the control system, a diagonal matrix decoupling controller is constructed, including: Multiply with the object feature matrix The generalized object matrix formed becomes a diagonal matrix, so that the generalized object matrix formed by the product of the matrix and the object feature matrix becomes a diagonal matrix, thereby realizing system decoupling: When the mathematical model matrix of the controlled object is non-singular, The decoupling compensation matrix is ​​written as: S5: The predicted value of short-term power is used to obtain the required air compressor speed and back pressure valve opening by table lookup method; S6: The decoupling effect of the air system can be analyzed through the relative control errors of pressure and flow, and the fluctuation of the actuator can be analyzed through the total change value of the actuator.

2. The decoupling method for advance control of a fuel cell air system according to claim 1, characterized in that: Step S1 specifically includes the following steps: S11: Assume that the time series of the fuel cell vehicle power demand with N data points is [p1, p2, p3, …, p N ] T , select the embedding dimension m and time delay τ, and construct the input vector x t =[p t-(m-1)τ ,…,p t-2τ ,p t-τ ,p t ]∈R m ; The embedding dimension refers to the number of historical time series data points input, and the time delay refers to the time interval between the input historical time series data points; the corresponding predicted output variable is y t+h =[p t+h ]∈R 1 , where h is the prediction step length, called h-step forward prediction. The demand power time series recursive prediction model of h-step forward prediction is as follows: Where f(·) is a generalized smooth nonlinear mapping; S12: Based on the phase space reconstruction method, the demand power time series prediction input and output sample pairs are established when the sliding window width L = 9, the embedding dimension m = 5, the time delay τ = 2 and the sliding period T = 4, as shown in the following formula: In the formula, are the input matrix and the expected output matrix after reconstructing the phase space, respectively. The required power time series is [p1, p2, p3, …, p N ] T Represents; using the input sample matrix after reconstructing the phase space as the input of the prediction model F(·), the predicted data value at the future moment is estimated; S13: The formula for calculating the autocorrelation coefficient is defined as: Wherein, the lag number k = 1, 2, ...; the autocorrelation coefficient is the time series data y = [y1, y2, y3, ..., y N ] T Two sample point sequences y constructed according to the lag number k t =[y 1+k ,y 2+k ,y 3+k ,…,y N ] T and t-k =[y1,y2,y3,…,y N-k ] T degree of relevance.

3. The decoupling method for advance control of a fuel cell air system according to claim 2, characterized in that: An iterative learning framework is constructed to replace the one-time prediction method of completing a large time span prediction by several iterative predictions with shorter time steps. Based on the theory of reconstructed embedded phase space, the training data sets of the learning elements used to predict power and power change rate in the xth iteration are constructed respectively. and As shown in the following formula: [P(t+β);P(t+2β);…;P(t+xβ);P′(t+β);P′(t+2β);…;P′(t+xβ)]=f(P(t),P′(t),P″(t))# Where F(·) represents the abstract nonlinear mapping of the trained ILF iterative learning framework.

4. The decoupling method for advance control of a fuel cell air system according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31: Build and debug the test bench; S32: Setting the amplitude range of the air compressor speed M sequence and the back pressure valve opening; S33: Considering the identification algorithm, the sampling period of the control system and the sensitivity of the sensor at the same time, the sampling period of the M sequence is selected and the sampling time is set.

5. The decoupling method for advance control of a fuel cell air system according to claim 4, characterized in that: The step S31 of constructing and debugging the test bench specifically includes: (1) Determine the topology of the test bench, determine the component selection and pipeline design; (2) Make mechanical and electrical connections according to the topology and test the mechanical and electrical connections; (3) Calibrate each sensor; (4) Test the status of the air compressor and back pressure valve; (5) Calibrate the opening of each ball valve according to the pressure drop of the fuel cell stack, intercooler, and humidifier; (6) Bench test the system under the full range of operating conditions.

6. The decoupling method for advance control of a fuel cell air system according to claim 4, characterized in that: In step S33, the sampling time is set to: T0 is the sampling period, T 95 It is the response time for the controlled object to reach 95% of the target value. The denominator is set to any constant in the range of 5 to 15 according to the actual situation.

7. The decoupling method for advance control of a fuel cell air system according to claim 1, characterized in that: Step S6 specifically includes: The total variation of the actuators including the air compressor and the back pressure valve is analyzed simultaneously using the mean square of the absolute error of flow and pressure as shown below: Among them, Pressure(t) and Flow(t) represent the pressure and flow at a certain moment, Pressure_MAE, Flow_MAE, and Pressure_MAE represent the average values ​​of pressure and flow; the better the decoupling effect of the decoupling strategy, the smaller the Flow_MAE and Pressure_MAE, the smaller the Tv, and the better the stability and anti-interference performance.

Images