Layered prediction air supply control method for vehicle hydrogen fuel cell

Abstract

The invention provides a layered prediction air supply control method for a vehicle hydrogen fuel cell, and the method comprises the steps: firstly obtaining corresponding fuel cell current prediction information through the collection of urban working condition data, the prediction of a future vehicle speed sequence and the prediction of a driving power demand in cooperation with a designed energy management strategy; the discretized battery current information is filled into the established air supply system state prediction model, and unmeasurable random disturbance is considered in the prediction model; the prediction model simultaneously contains measurable current disturbance information and unmeasurable random disturbance, so that an air supply mode based on improved random model prediction control is obtained, relatively accurate air flow control can be realized, and the output performance of the vehicle hydrogen fuel cell can be remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of fuel cell control, and in particular relates to a stratified prediction air supply control method for a vehicle hydrogen fuel cell. Background Art

[0002] When hydrogen fuel cell vehicles are driving in complex operating environments such as cities, they need to frequently adjust the power distribution between the fuel cell and the power battery, which poses a severe challenge to the air supply system's response to load demand. Existing control strategies for the air supply of hydrogen fuel cells for vehicles are mostly based on control methods such as PID control, fuzzy control, sliding mode control, and model predictive control (MPC). These methods currently have the problem of not being able to adapt well to changes in power demand at future times. At the same time, due to the lack of necessary disturbance information in the modeling of the air supply system, these methods are difficult to accurately predict and adjust the air flow, and cannot ensure the precise match between the transfer process of the reactants inside the fuel cell stack and the requirements of the electrochemical reaction. Therefore, how to provide a fuel cell air flow control strategy that is accurate and efficient and suitable for complex urban driving environments is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0003] In view of this, and in response to the technical problems existing in the art, the present invention provides a method for controlling stratified air supply of a hydrogen fuel cell for a vehicle, which specifically includes the following steps:

[0004] Step 1: Collect vehicle driving data under various urban conditions and establish a training data set for model training;

[0005] Step 2: Build a deep learning short-term operating condition prediction model based on the long short-term memory network. Use the training data set obtained in step 1 for training. After training, use it to output a predicted future speed sequence based on the input historical speed information.

[0006] Step 3: Establish a vehicle longitudinal dynamics model to predict driving power requirements based on vehicle speed and road information.

[0007] Step 4: Consider various vehicle operating modes including brake energy recovery, set the power allocation ratio corresponding to different SOC intervals, and establish a vehicle energy management strategy to predict the hydrogen fuel cell power demand based on the driving power demand;

[0008] Step 5: Establish a hydrogen fuel cell stack potential output model to calculate the load current information prediction value based on the hydrogen fuel cell power demand;

[0009] Step 6: Discretize the load current information in the prediction time domain according to the controller sampling period and obtain the corresponding current disturbance sequence; establish a discretized air supply system state prediction model, and fill the discretized current disturbance sequence into the air supply system state prediction model;

[0010] Step 7: Utilize the air supply system state prediction model after filling current disturbance and the real-time collected vehicle operation information to perform optimization according to the corresponding objective function and output the corresponding air supply system control action.

[0011] Furthermore, in step 1, pure electric buses on urban bus operating routes are specifically used in conjunction with the Npos220 high-precision integrated navigation system to collect various urban working conditions, and data cleaning processing including data filtering and outlier removal is performed to ensure that the working condition data has stable periodic characteristics in the time distribution dimension.

[0012] Furthermore, in step 2, a deep learning short-term operating condition prediction model is established based on a bidirectional long short-term memory network. It takes 10 historical vehicle speed information as input and outputs the predicted vehicle speed sequence for the next 5 moments. The network output calculation process is as follows:

[0013] y k =W y h k +W′ y h′ k +b y

[0014] Where, the weight matrix W y and W′ y Represents the contribution of forward and reverse features respectively; h k and h′ k are the forward layer output and the backward layer output respectively; b y is the bias coefficient.

[0015] Furthermore, in step 3, the driving resistances including rolling resistance, slope resistance, air resistance and acceleration resistance encountered by the vehicle during driving are specifically considered, and the longitudinal dynamic model of the following form is established:

[0016]

[0017] Where r is the wheel radius; η T is the transmission efficiency; i0 is the main reduction ratio; F f 、F i 、F w 、F j They are rolling resistance, slope resistance, air resistance and acceleration resistance, which are calculated using the following formula:

[0018]

[0019] In the formula, G represents the total weight of the car, m represents the total mass of the car, f represents the rolling resistance coefficient of the wheel, C D Indicates the air resistance coefficient, A w represents the frontal area, α represents the road slope, ρ represents the air density, δ is the moment of inertia conversion coefficient, v is the vehicle speed, and t is the time.

[0020] Furthermore, in step 4, a rule-based energy management strategy is specifically established, including five operating modes: starting, fuel cell independent driving, driving charging, combined power supply, and braking with energy recovery function; each operating mode sets several SOC intervals and corresponding power distribution ratios.

[0021] Furthermore, the hydrogen fuel cell stack potential output model established in step 5 specifically adopts a voltage loss model including activation polarization loss, ohmic polarization loss and concentration polarization loss; the fuel cell single output voltage V cell Specifically expressed as:

[0022] V cell =E nernst -V act -V ohm -V con

[0023] Where, E nernst is the open circuit voltage of the fuel cell, V act is the activation polarization voltage, V ohm is the ohmic polarization voltage, V con is the concentration polarization voltage;

[0024] The activation polarization voltage is calculated using the following formula:

[0025]

[0026] Where v0 is the inherent voltage drop at zero current density, v a is the polarization voltage generated by the resistance in the stack, c1 is an empirical constant, which can be taken as c1=10, i fc is the current density;

[0027] The ohmic polarization voltage is calculated according to Ohm's law:

[0028]

[0029] Where, t m is the thickness of the proton membrane, σ m is the membrane conductivity, and its value is related to the stack temperature and membrane water content;

[0030] The concentration polarization voltage is calculated using the following formula:

[0031]

[0032] Where i fc,max is the highest current density, and the coefficient c3 is a system constant related to the stack temperature and oxygen partial pressure, and can be taken as 2.

[0033] Furthermore, the second-level time scale current disturbance prediction value I obtained based on the predicted vehicle speed condition in step 6 is pre (n), using the same sampling period T as the air supply controller mpc The discretization is performed on a synchronized time grid to generate a sequence of equally spaced perturbations I dis :

[0034] I pre (n)→I dis (1:1 / T mpc )

[0035] The established air supply system state prediction model specifically adopts a lumped parameter mechanism model, including the air compressor speed ω cp , Gas pressure in the supply manifold p sm , Oxygen mass in cathode flow channel Nitrogen quality in cathode flow channel Gas pressure in the exhaust manifold p rm These five system state variables; the system continuous dynamic equations are as follows:

[0036]

[0037] Where K t and R cm is the motor parameter, η cm is the motor efficiency, v cp is the air compressor control voltage, the superscript · represents the derivative of the corresponding parameter, and f1…f5 represent the functions corresponding to each state variable;

[0038] The forward Euler method is used to discretize the continuous dynamic equations of the system. Taking into account the additive noise during the operation of the system, the following discrete system state prediction model is obtained:

[0039]

[0040] Where w represents the system noise, T is the discrete time interval, and k represents the system state at time k;

[0041] Based on the measurement signals including sensor output values ​​and controller analysis data, the following system measurement model is established to characterize the relationship between the measurable signals and the system state:

[0042]

[0043] Define the system state vector as The system measurement vector is Z = [z1, z2, z3] T =[z ω ,z sm ,z rm ] T On this basis, the system state prediction equation is transformed into the following state space equation:

[0044]

[0045] Where, the vector field f(·) and B are the state function and input matrix respectively, h(·) is the system measurement equation, the vector field Φ is the disturbance matrix, and w k and v k are the system noise and measurement noise in the form of Gaussian white noise, and the load current I st It is represented as a measurable disturbance, which is mapped to the air supply system state prediction model according to the control time domain dimension. The mapping relationship is specifically expressed as:

[0046] I dis (k:k+T v -1)→d(k:k+T v -1),k=1,2,…,1 / T mpc

[0047] Where, T v is the control time domain of the air supply controller, and d is the disturbance in the prediction model.

[0048] Furthermore, the system dynamic equations are linearized at the rated operating point using Taylor expansion to obtain a discrete-time state-space model containing measurable disturbances and unmeasurable random disturbances. The specific form is as follows:

[0049]

[0050] Where X(k) is the system state variable at time k, U(k) is the control variable, i.e., the air compressor control voltage, Y(k) is the system output, i.e., the excess oxygen ratio, d(k) represents the measurable disturbance, i.e., the load current information, and w(k) is the unmeasurable random disturbance. C, D u 、D d are all state space equation matrices;

[0051] Rewrite it as an incremental model:

[0052]

[0053] Then the improved prediction model system response filled with current disturbance sequence can be expressed as:

[0054] Y p (k+1|k)=P X ΔX(k)+P U ΔU(k)+P d Δ[d(k)+w(k)]+EY(k)

[0055] Where E is the identity matrix and the equation matrix P is X 、P U 、P d Respectively by C, D u 、D d Calculated;

[0056] Considering the flow control error and the control action increment at the same time, the following objective function is established:

[0057]

[0058] Where R is the reference value of the excess oxygen ratio, γ and κ represent the weight factors of the output error and the control increment, respectively, and T p and T v They are the prediction time domain and the control time domain respectively.

[0059] The above-mentioned stratified predictive air supply control method for vehicle hydrogen fuel cells provided by the present invention first obtains the corresponding fuel cell current prediction information through urban operating condition data collection, future vehicle speed sequence prediction and driving power demand prediction, in conjunction with the designed energy management strategy, and then fills the discretized battery current information into the established air supply system state prediction model, and considers unmeasurable random disturbances in the prediction model, so that the prediction model contains both measurable current disturbance information and unmeasurable random disturbances, thereby obtaining an air supply method based on improved random model predictive control, which can achieve more accurate air flow control and significantly improve the output performance of vehicle hydrogen fuel cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flow chart of the method provided by the present invention;

[0061] Figure 2 This is a diagram of the training and prediction process of the short-term working condition prediction model;

[0062] Figure 3 Schematic diagram of the working condition prediction effect in an example of the present invention

[0063] Figure 4 Schematic diagram of air flow control effect in an example of the present invention. DETAILED DESCRIPTION

[0064] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0065] The present invention provides a method for controlling stratified air supply of a hydrogen fuel cell for a vehicle. Figure 1 As shown, the specific steps include:

[0066] Step 1: Collect vehicle driving data under various urban conditions and establish a training data set for model training;

[0067] Step 2: Build a deep learning short-term operating condition prediction model based on the long short-term memory network. Use the training data set obtained in step 1 for training. After training, use it to output a predicted future speed sequence based on the input historical speed information.

[0068] Step 3: Establish a vehicle longitudinal dynamics model to predict driving power requirements based on vehicle speed and road information.

[0069] Step 4: Consider various vehicle operating modes including brake energy recovery, set the power allocation ratio corresponding to different SOC intervals, and establish a vehicle energy management strategy to predict the hydrogen fuel cell power demand based on the driving power demand;

[0070] Step 5: Establish a hydrogen fuel cell stack potential output model to calculate the load current information prediction value based on the hydrogen fuel cell power demand;

[0071] Step 6: Discretize the load current information in the prediction time domain according to the controller sampling period and obtain the corresponding current disturbance sequence; establish a discretized air supply system state prediction model, and fill the discretized current disturbance sequence into the air supply system state prediction model;

[0072] Step 7: Utilize the air supply system state prediction model after filling current disturbance and the real-time collected vehicle operation information to perform optimization according to the corresponding objective function and output the corresponding air supply system control action.

[0073] In a preferred embodiment of the present invention, in step 1, pure electric buses on urban bus operating lines are specifically used in conjunction with the Npos220 high-precision combined navigation system to collect various urban working conditions, and data cleaning processing including data filtering and outlier removal is performed to make the working condition data have stable periodic characteristics in the time distribution dimension. The final working condition data set is composed of 4 groups of typical working conditions and 4 groups of actual vehicle collection working conditions.

[0074] In a preferred embodiment of the present invention, in step 2, a deep learning short-term operating condition prediction model is established based on a bidirectional long short-term memory network, which takes 10 historical vehicle speed information as input and outputs the predicted vehicle speed sequence of the next 5 moments; Figure 2 As shown, the network output calculation process is:

[0075] y k =W y h k +W′yh′ k +b y

[0076] Where, the weight matrix W y and W′ y Represents the contribution of forward and reverse features respectively; h k and h′ k are the forward layer output and the backward layer output respectively; b y is the bias coefficient.

[0077] In a preferred embodiment of the present invention, in step 3, the driving resistance including rolling resistance, slope resistance, air resistance and acceleration resistance encountered by the vehicle during driving is specifically considered, and a longitudinal dynamic model of the following form is established:

[0078]

[0079] Where r is the wheel radius; η T is the transmission efficiency; i0 is the main reduction ratio; F f 、F i 、F w 、F j They are rolling resistance, slope resistance, air resistance and acceleration resistance, which are calculated using the following formula:

[0080]

[0081] In the formula, G represents the total weight of the car, m represents the total mass of the car, f represents the rolling resistance coefficient of the wheel, C D Indicates the air resistance coefficient, A w represents the frontal area, α represents the road slope, ρ represents the air density, δ is the moment of inertia conversion coefficient, v is the vehicle speed, and t is the time.

[0082] In a preferred embodiment of the present invention, a rule-based energy management strategy is specifically established in step 4, including five operating modes: starting, fuel cell driving alone, driving and charging, combined power supply, and braking with energy recovery function; each operating mode is respectively set with several SOC intervals and corresponding power distribution ratios. For example, the fuel cell output can be limited in the high SOC interval to avoid overcharging of the power battery and extend its service life; when the SOC is lower than the lower limit threshold, the fuel cell is allowed to bear all driving needs to prevent deep discharge damage to the power battery.

[0083] In a preferred embodiment of the present invention, the hydrogen fuel cell stack potential output model established in step 5 specifically adopts a voltage loss model including activation polarization loss, ohmic polarization loss and concentration polarization loss; the fuel cell single output voltage V cell Specifically expressed as:

[0084] V cell =E nernst -V act -V ohm -V con

[0085] Where, E nernst is the open circuit voltage of the fuel cell, V act is the activation polarization voltage, V ohm is the ohmic polarization voltage, V con is the concentration polarization voltage;

[0086] The activation polarization voltage is calculated using the following formula:

[0087]

[0088] Where v0 is the inherent voltage drop at zero current density, v a is the polarization voltage generated by the resistance in the stack, c1 is an empirical constant, which can be taken as c1=10, i fc is the current density;

[0089] The ohmic polarization voltage is calculated according to Ohm's law:

[0090]

[0091] Where, t m is the thickness of the proton membrane, σ m is the membrane conductivity, and its value is related to the stack temperature and membrane water content;

[0092] The concentration polarization voltage is calculated using the following formula:

[0093]

[0094] Where i fc,max is the highest current density, and the coefficient c3 is a system constant related to the stack temperature and oxygen partial pressure, and can be taken as 2.

[0095] In a preferred embodiment of the present invention, the second-level time scale current disturbance prediction value I is obtained based on the predicted vehicle speed condition in step 6. pre (n), using the same sampling period T as the air supply controller mpc The discretization is performed on a synchronized time grid to generate a sequence of equally spaced perturbations I dis :

[0096] I pre (n)→I dis (1:1 / T mpc )

[0097] The established air supply system state prediction model specifically adopts a lumped parameter mechanism model, including the air compressor speed ω cp , Gas pressure in the supply manifold p sm , Oxygen mass in cathode flow channel Nitrogen quality in cathode flow channel Gas pressure in the exhaust manifold p rm These five system state variables; the system continuous dynamic equations are as follows:

[0098]

[0099] Where K t and R cm is the motor parameter, η cm is the motor efficiency, v cp is the air compressor control voltage, the superscript · represents the derivative of the corresponding parameter, and f1…f5 represent the functions corresponding to each state variable;

[0100] The forward Euler method is used to discretize the continuous dynamic equations of the system. Taking into account the additive noise during the operation of the system, the following discrete system state prediction model is obtained:

[0101]

[0102] Where w represents the system noise, T is the discrete time interval, and k represents the system state at time k;

[0103] Based on the measurement signals including sensor output values ​​and controller analysis data, the following system measurement model is established to characterize the relationship between the measurable signals and the system state:

[0104]

[0105] Define the system state vector as The system measurement vector is Z = [z1, z2, z3] T =[z ω ,z sm ,z rm ] T On this basis, the system state prediction equation is transformed into the following state space equation:

[0106]

[0107] Where, the vector field f(·) and B are the state function and input matrix respectively, h(·) is the system measurement equation, the vector field Φ is the disturbance matrix, and w k and v k are the system noise and measurement noise in the form of Gaussian white noise, and the load current I st It is represented as a measurable disturbance, which is mapped to the air supply system state prediction model according to the control time domain dimension. The mapping relationship is specifically expressed as:

[0108] I dis (k:k+T v -1)→d(k:k+T v -1),k=1,2,…,1 / T mpc

[0109] Where, T v is the control time domain of the air supply controller, and d is the disturbance in the prediction model.

[0110] In a preferred embodiment of the present invention, the system dynamic equation is linearized at the rated operating point using Taylor expansion to obtain a discrete-time state-space model containing measurable disturbances and unmeasurable random disturbances, which is specifically in the following form:

[0111]

[0112] Where X(k) is the system state variable at time k, U(k) is the control variable, i.e., the air compressor control voltage, Y(k) is the system output, i.e., the excess oxygen ratio, d(k) represents the measurable disturbance, i.e., the load current information, and w(k) is the unmeasurable random disturbance. C, D u 、D d are all state space equation matrices;

[0113] Rewrite it as an incremental model:

[0114]

[0115] Then the improved prediction model system response filled with current disturbance sequence can be expressed as:

[0116] Yp (k+1|k)=P X ΔX(k)+P U ΔU(k)+P d Δ[d(k)+w(k)]+EY(k)

[0117] Where E is the identity matrix and the equation matrix P is X 、P U 、P d Respectively by C, D u 、D d Calculated;

[0118] Considering the flow control error and the control action increment at the same time, the following objective function is established:

[0119]

[0120] Where R is the reference value of the excess oxygen ratio, γ and κ represent the weight factors of the output error and the control increment, respectively, and T p and T v They are the prediction time domain and the control time domain respectively.

[0121] Figure 3 and 4 The prediction effect of the working condition and the air flow control effect based on the example of the present invention are respectively shown, and a comparison of the control effect of the present invention and the traditional method is also provided, which fully demonstrates the advantages of the present invention over the existing technology.

[0122] It should be understood that the size of the serial numbers of the steps in the embodiment of the present invention does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0123] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for controlling stratified air supply in a hydrogen fuel cell for a vehicle, characterized by: The specific steps include: Step 1: Collect vehicle driving data under various urban conditions and establish a training data set for model training; Step 2: Build a deep learning short-term operating condition prediction model based on the long short-term memory network. Use the training data set obtained in step 1 for training. After training, use it to output a predicted future speed sequence based on the input historical speed information. Step 3: Establish a vehicle longitudinal dynamics model to predict driving power requirements based on vehicle speed and road information. Step 4: Consider various vehicle operating modes including brake energy recovery, set the power allocation ratio corresponding to different SOC intervals, and establish a vehicle energy management strategy to predict the hydrogen fuel cell power demand based on the driving power demand; Step 5: Establish a hydrogen fuel cell stack potential output model to calculate the load current information prediction value based on the hydrogen fuel cell power demand; Step 6: Discretize the load current information in the prediction time domain according to the controller sampling period and obtain the corresponding current disturbance sequence; establish a discretized air supply system state prediction model, and fill the discretized current disturbance sequence into the air supply system state prediction model; Step 7: Utilize the air supply system state prediction model after filling current disturbance and the real-time collected vehicle operation information to perform optimization according to the corresponding objective function and output the corresponding air supply system control action.

2. The method according to claim 1, wherein: In step 1, pure electric buses on urban bus operating routes are used in conjunction with the Npos220 high-precision integrated navigation system to collect various urban operating conditions, and data cleaning processing including data filtering and outlier removal is performed to ensure that the operating condition data has stable periodic characteristics in the time distribution dimension.

3. The method according to claim 1, wherein: In step 2, a deep learning short-term operating condition prediction model is established based on a bidirectional long short-term memory network. It takes 10 historical vehicle speed information as input and outputs the predicted vehicle speed sequence for the next 5 moments. The network output calculation process is as follows: y k =W y h k +W′ y h′ k +b y Where, the weight matrix W y and W′ y Represent the contribution of forward and reverse features respectively; h k and h′ k are the forward layer output and the backward layer output respectively; b y is the bias coefficient.

4. The method according to claim 1, wherein: In step 3, the driving resistance of the vehicle, including rolling resistance, slope resistance, air resistance and acceleration resistance, is specifically considered, and the longitudinal dynamic model of the following form is established: Where r is the wheel radius; η T is the transmission efficiency; i0 is the main reduction ratio; F f 、F i 、F w 、F j They are rolling resistance, slope resistance, air resistance and acceleration resistance, which are calculated using the following formula: In the formula, G represents the total weight of the car, m represents the total mass of the car, f represents the rolling resistance coefficient of the wheel, C D Indicates the air resistance coefficient, A w represents the frontal area, α represents the road slope, ρ represents the air density, δ is the moment of inertia conversion coefficient, v is the vehicle speed, and t is the time.

5. The method according to claim 1, wherein: In step 4, a rule-based energy management strategy is specifically established, including five operating modes: starting, fuel cell driving alone, driving charging, combined power supply, and braking with energy recovery function; each operating mode sets several SOC intervals and corresponding power allocation ratios.

6. The method according to claim 1, wherein: The hydrogen fuel cell stack potential output model established in step 5 specifically adopts a voltage loss model including activation polarization loss, ohmic polarization loss and concentration polarization loss; the fuel cell single output voltage V cell Specifically expressed as: V cell =E nernst -V act -V ohm -V con Where, E nernst is the open circuit voltage of the fuel cell, V act is the activation polarization voltage, V ohm is the ohmic polarization voltage, V con is the concentration polarization voltage; The activation polarization voltage is calculated using the following formula: Where v0 is the inherent voltage drop at zero current density, v a is the polarization voltage generated by the resistance in the battery stack, c1 is an empirical constant, i fc is the current density; The ohmic polarization voltage is calculated according to Ohm's law: Where, t m is the thickness of the proton membrane, σ m is the membrane conductivity, and its value is related to the stack temperature and membrane water content; The concentration polarization voltage is calculated using the following formula: Where i fc,max is the maximum current density, and the coefficient c3 is a system constant related to the stack temperature and oxygen partial pressure.

7. The method according to claim 1, wherein: In step 6, the second-level time scale current disturbance prediction value I is obtained based on the predicted vehicle speed condition. pre (n), using the same sampling period T as the air supply controller mpc The discretization is performed on a synchronized time grid to generate a sequence of equally spaced perturbations I dis : I pre (n)→I dis (1:1 / T mpc ) The established air supply system state prediction model specifically adopts a lumped parameter mechanism model, including the air compressor speed ω cp , Gas pressure in the supply manifold p sm , Oxygen mass in cathode flow channel Nitrogen quality in cathode flow channel Gas pressure in the exhaust manifold p rm These five system state variables; the system continuous dynamic equations are as follows: Where K t and R cm is the motor parameter, η cm is the motor efficiency, v cp is the air compressor control voltage, the superscript · represents the derivative of the corresponding parameter, and f1…f5 represent the functions corresponding to each state variable; The forward Euler method is used to discretize the continuous dynamic equations of the system. Taking into account the additive noise during the operation of the system, the following discrete system state prediction model is obtained: Where w represents the system noise, t is the discrete time interval, and k represents the system state at time k; Based on the measurement signals including sensor output values ​​and controller analysis data, the following system measurement model is established to characterize the relationship between the measurable signals and the system state: Define the system state vector as The system measurement vector is Z = [z1, z2, z3] T =[z ω ,z sm ,z rm ] T On this basis, the system state prediction equation is transformed into the following state space equation: Where, the vector field f(·) and B are the state function and input matrix respectively, h(·) is the system measurement equation, the vector field Φ is the disturbance matrix, and w k and v k are the system noise and measurement noise in the form of Gaussian white noise, and the load current I st It is represented as a measurable disturbance, which is mapped to the air supply system state prediction model according to the control time domain dimension. The mapping relationship is specifically expressed as: I dis (k:k+T v -1)→d(k:k+T v -1),k=1,2,…,1 / T mpc Where, T v is the control time domain of the air supply controller, and d is the disturbance in the prediction model.

8. The method according to claim 7, wherein: The system dynamic equations are linearized at the rated operating point using Taylor expansion to obtain a discrete-time state-space model containing measurable disturbances and unmeasurable random disturbances. The specific form is as follows: Where X(k) is the system state variable at time k, U(k) is the control variable, i.e., the air compressor control voltage, Y(k) is the system output, i.e., the excess oxygen ratio, d(k) represents the measurable disturbance, i.e., the load current information, and w(k) is the unmeasurable random disturbance. C, D u 、D d are all state space equation matrices; Rewrite it as an incremental model: Then the improved prediction model system response filled with current disturbance sequence can be expressed as: Y p (k+1|k)=P X ΔX(k)+P U ΔU(k)+P d Δ[d(k)+w(k)]+EY(k) Where E is the identity matrix and the equation matrix P is X 、P U 、P d Respectively by C, D u 、D d Calculated; Considering the flow control error and the control action increment at the same time, the following objective function is established: Where R is the reference value of the overoxygen ratio, γ and k represent the weight factors of the output error and the control increment, respectively, and T p and T v They are the prediction time domain and the control time domain respectively.

Images