Cooperative Self-Optimization Method for Input Gain and Observation Bandwidth of Multivariable Active Disturbance Rejection Controller in Diesel Engine Air System

Abstract

The present invention discloses a collaborative self-optimizing method for the input gain and observation bandwidth of a multivariable active disturbance rejection controller. Facing the control model of the TVA-VGT-EGR diesel engine air system, a multivariable active disturbance rejection TITO ADRC controller is designed. The multivariable active disturbance rejection TITO ADRC controller includes a tracking differentiator TD, an ESO observer, and an error disturbance control law SEF. Through the convergence of the ESO observer, the control of p 22 , p3 and X EGR is achieved. The output #imgabs0# of the ESO observer is transmitted to the error disturbance control law SEF. By using the A TVA , A VGT and A EGR output by the error disturbance control law SEF respectively, the opening information is calculated to obtain U TVA , U VGT and U EGR , and they are sent into the control model of the TVA-VGT-EGR diesel engine air system. The sensitivity analysis of the input gain and control parameters is carried out, and the optimal values of the parameters to be learned are obtained through the extremum search method ES, so as to solve the complex cross-coupling relationship and mutual coupling influence among the control loops of the air system, and at the same time, the control parameters and model parameters are self-learned online.

Description

Technical Field

[0001] The present invention relates to the technical field of engine air systems, and particularly to a collaborative self-optimizing algorithm for the input gain and observation bandwidth of a multivariable active disturbance rejection controller for a diesel engine air system. Background Art

[0002] The diesel engine air system has multiple state variables and is a complex system with strong nonlinearity, strong coupling, and time-variation. During the operation of the TVA-VGT-EGR diesel engine air system, the opening change of its actuator will cause changes in various variables in the system. At the same time, there are many uncertain mutual influence relationships among these variables, and there is a strong coupling characteristic among the controlled loops. When controlling such a system, it is very difficult to fully decouple the system by directly controlling the boost pressure and the EGR rate, and it is also very difficult to accurately compensate for the uncertainties and various disturbances in the system, and it is not easy to obtain a stable control effect under various working conditions. The control difficulty of the air system is mainly reflected in the complex cross-coupling relationship and the mutual coupling influence among its control loops. Therefore, the key to the control of the air system is to clearly explore the system characteristics, reasonably model for control, and optimally design the control algorithm to achieve its full decoupling.

[0003] At the same time, some controller parameters in the decoupling control algorithm of the air system need to be obtained through tuning. The values of these parameters directly affect the algorithm performance and control effect. The deviation of the values will easily lead to overshoot, oscillation, too long adjustment time, etc. in the control of the boost pressure, and will also affect the control of the EGR rate to a certain extent. Therefore, it is necessary to carry out self-learning on these parameters. For the self-learning of algorithm parameters, reasonably analyzing the system mathematical model, designing the learning algorithm, and ensuring the accuracy of the learned parameters are the key points and difficulties in algorithm development.

[0004] Active Disturbance Rejection Control (ADRC) was formally proposed by Mr. Han Jingqing in 1998. Its uniqueness lies in attributing the uncertain factors acting on the controlled object to "unknown disturbances", and compensating for the unknown disturbances with the input and output data of the object. Its greatest advantage is that it does not require an accurate mathematical model of the controlled object. The research results and engineering application results show that ADRC has good adaptability, strong robustness and anti-interference ability. Its method has essential superiority in theory compared with other control ideas, and can especially effectively adapt to control objects with higher complexity and uncertainty. Therefore, in this study, the development of the control algorithm and the decoupling control of the air system are completed based on the ADRC theory.

[0005] Some researchers have improved it. For example, the literature (Optimization design and application of active disturbance rejection controller based on intelligent algorithm) uses a hybrid algorithm combining fish school algorithm and particle swarm algorithm to optimize the ADRC parameters and verifies it through experiments; the literature (Parameters turning of ADRC based on neural network) combines neural network with ADRC and gives a self-learning method for ADRC parameters. However, the above methods have problems such as complex control, large amount of calculation, and relying on experience to tune parameters, which are not convenient for practical applications. Summary of the Invention

[0006] The purpose of the present invention is to provide a collaborative self-optimizing algorithm for the input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system, so as to solve the complex cross-coupling relationship and mutual coupling influence between the control loops of the air system, and at the same time perform online self-learning on control parameters and model parameters.

[0007] The technical solution adopted to achieve the purpose of the present invention is as follows:

[0008] A collaborative self-optimizing method for the input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system, comprising the following steps:

[0009] Step 1, establish a control-oriented model of the TVA-VGT-EGR diesel engine air system composed of the core dynamic equation of the air system, the dynamic equation of the pressure before the turbine of the supercharger, and the dynamic equation of the EGR rate, and calculate the boost pressure p 22 , the pressure p 3 before the turbine and the EGR rate X EGR in real time;

[0010] Step 2, in the face of the control-oriented model of the TVA-VGT-EGR diesel engine air system, design a multi-variable active disturbance rejection TITO (Three Input Three Output) ADRC controller. The TITO ADRC controller includes a tracking differentiator TD (Tracking Differentiator), an ESO observer, and a state error feedback control law SEF (States Error Feedback). Through the convergence of the ESO observer, the control of p 22 , p 3 and X EGR is achieved, and the output of the ESO observer Passed to the error feedback control law SEF, and the effective flow area A output by the error feedback control law SEF is utilized TVA , A VGT and A EGR are respectively used to calculate the opening information to obtain U TVA , U VGT and U EGR , and then sent to the TVA-VGT-EGR diesel engine air system control-oriented model;

[0011] Step 3: According to the TVA-VGT-EGR diesel engine air system control-oriented model and the simulation results, perform sensitivity analysis on the model input gain and control parameters. The control parameters include the bandwidth ω c of the error feedback control law SEF 0 and the bandwidth ω c of the ESO observer. From the model input gain, ω 0 , select the parameters that have the most significant impact on the system control and are directly related as the parameters to be learned;

[0012] Step 4: Design an error cost function, and use the extremum seeking method ES (Extremum Seeking) to obtain the optimal values of the parameters to be learned;

[0013] Step 5: Use the forgetting factor recursive least squares FFRLS (Forgetting Factor Recursive Least Squares) method to identify the unknown model parameters in the control loop of the parameters to be learned.

[0014] In the above technical solution, the ESO observer in step 2 is:

[0015]

[0016] In the formula

[0017]

[0018]

[0019]

[0020] η vol is the charging efficiency, V d is the cylinder volume, N Eng is the engine speed, V 22 is the intake manifold volume, p 21 is the pressure in front of the TVA valve, T 22 is the intake temperature, T 21 is the temperature in front of the TVA valve, σ1 is the coefficient to be calibrated, p22 is the pressure after the throttle valve, σ2 is the coefficient to be calibrated, p 3 is the pressure before the turbine, R is the ideal gas constant, T 3 is the temperature before the turbine, V 3 is the volume of the exhaust manifold, is the dynamic change of the fuel mass flow rate, A TVA is the effective flow cross-sectional area of the TVA valve, A EGR is the effective flow cross-sectional area of the EGR valve, T 3 is the temperature before the turbine, p 3 is the pressure before the turbine, A VGT is the effective flow cross-sectional area of the VGT valve, β is the coefficient to be calibrated, ρ 2 is the density of the intake gas in the cylinder;

[0021] are the ESO estimated values of the boost pressure, the pressure before the turbine, and the EGR rate respectively; are p 22 、p 3 and X EGR are the ESO estimated values of the disturbances in the channels; l 1 、l 2 、l 3 、l 4 、l 5 、l 6 are the parameters to be tuned;

[0022] Configure the ESO poles to ω 0 to obtain the parameters to be tuned l 1 、l 2 、l 3 、l 4 、l 5 、l 6 :

[0023]

[0024] In the above technical solution, the error feedback control law SEF is:

[0025]

[0026] Among them, k 1 、k 2 、k 3 are the parameters to be tuned, is the target value of the controlled variable, X′ des is the feedforward of the change rate of the target value of the controlled variable;

[0027] Configure the poles to the bandwidth ω of the error feedback control lawc to obtain the parameter k to be tuned 1 、k 2 、k 3 :

[0028]

[0029] In the above technical solution, the model input gain in step 3 includes b 1 、b 2 、b 3 、b 4 、b 5 。

[0030] In the above technical solution, in step 4, the parameter to be learned is ω 0 、b 1 、b 3 。

[0031] In the above technical solution, in step 4, the error cost function of ω 0 is:

[0032]

[0033] In the formula, K 1 is the gain coefficient; X des is the parameter target value, that is X is the parameter actual value, that is p 22 、p 3 ; Deriving the output of the ESO observer ; α, β, γ are weight coefficients.

[0034] In the above technical solution, in step 4, the error cost function of b i is:

[0035] J(b i )=-K 2 (X aim -X) 2

[0036] In the formula, K 2 is the gain coefficient, X is the parameter actual value, that is p 22 、p 3 , X aim is the parameter required value X ref filtered by an ideal filter, X ref includes p 22ref 、p 3ref 。

[0037] In the above technical solution, in step 5, the unknown model parameter in the parameter learning control loop is a1 and a 2 。

[0038] In the above technical solution, in step 5, the formula of the forgetting factor recursive least squares method is as follows:

[0039]

[0040]

[0041]

[0042] where λ is the forgetting factor, and 0 ≤ λ ≤ 1;

[0043] Send the learning result b 1 、b 3 into the forgetting factor recursive least squares FFRLS algorithm, so that only the parameter a to be identified 1 、a 2 ;

[0044]

[0045]

[0046] where, is the identification result of a 1 , is the identification result of δ 1 , is the identification result of a 2 , is the identification result of δ 2 .

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. Compared with the traditional control structure, the present invention can improve the control accuracy of key parameters and the anti-interference ability, making the system dynamic response process smoother.

[0049] 2. Compared with the traditional control law, the present invention compensates for the main control parameters and extracts differential signals for feedforward compensation, which not only reduces the overshoot in the dynamic process but also improves the tracking speed of the transient target value.

[0050] 3. Compared with the traditional control mode, the present invention develops a self-learning algorithm for control parameters and model parameters, and optimally estimates the parameters based on the model in the controller. The response time is increased by nearly 1s, and the overshoot is reduced by nearly 15%. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1It is the overall control framework diagram of the TVA-VGT-EGR air system.

[0052] Figure 2 It is the framework diagram of the TITOADRC controller.

[0053] Figure 3 It is the control effect diagram of the FTP75 cycle test.

[0054] Figure 4 It is the overall structure diagram of the multi-variable ES.

[0055] Figure 5 It is the parameter sensitivity analysis curve.

[0056] Figure 6 It is the step test and dynamic test effect curves of the multi-variable ES, where (a) is the step test at 1000 rpm and (b) is the step test at 2200 rpm. Specific implementation manners

[0057] The present invention will be further described in detail below in conjunction with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] Embodiment 1

[0059] A collaborative self-optimizing method for the input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system, comprising the following steps:

[0060] Step 1, establish a control-oriented model of the TVA-VGT-EGR diesel engine air system according to the core dynamic equation of the diesel engine (including the diesel engine supercharging pressure dynamic equation, the supercharger inlet pressure dynamic equation, and the EGR rate dynamic equation), and this model outputs the supercharging pressure p 22 , the inlet pressure p 3 and the EGR rate X EGR in real time, serving the air system decoupling control method.

[0061] According to the ideal gas state equation and the mass conservation principle, establish the diesel engine supercharging pressure dynamic equation:

[0062]

[0063] In the formula, p 22 is the pressure after the throttle valve, represents the dynamic change of the pressure after the throttle valve; R is the ideal gas constant; T 22 is the intake air temperature; V 22 is the intake manifold volume; are respectively the dynamic changes of the throttle valve mass flow rate, the EGR mass flow rate, and the engine intake air volume.

[0064] Establish the dynamic equation of the pressure in front of the supercharger turbine:

[0065]

[0066] In the formula, T 3 is the temperature in front of the turbine; p 3 is the pressure in front of the turbine, represents the dynamic change of the pressure in front of the turbine; V 3 is the volume of the exhaust manifold; is the dynamic change of the fuel mass flow rate; is the dynamic change of the VGT mass flow rate, that is, the dynamic change of the turbine mass flow rate.

[0067] The brief descriptions of each mass flow sub-model in Equations (1) and (2) are as follows:

[0068] The throttle mass flow model is shown in Equation (3):

[0069]

[0070] In the formula, A TVA is the effective flow cross-sectional area of the TVA valve; T 21 is the temperature in front of the TVA valve; p 21 is the pressure in front of the TVA valve; σ1 is the coefficient to be calibrated.

[0071] Modeled according to the orifice flow equation, the EGR mass flow model is shown in Equation (4):

[0072]

[0073] In the formula, A EGR is the effective flow cross-sectional area of the EGR valve; T 3 is the temperature in front of the turbine; p 3 is the pressure in front of the turbine; σ2 is the coefficient to be calibrated.

[0074] The engine intake air volume model is established using the speed-density method, and the engine intake air volume model is shown in Equation (5):

[0075]

[0076] In the formula, η vol is the charging efficiency; V d is the cylinder volume; N Eng is the engine speed.

[0077] The VGT mass flow rate is modeled according to the orifice flow equation, and the VGT mass flow model is shown in Equation (6):

[0078]

[0079] In the formula, A vGT is the effective flow cross-sectional area of the VGT valve; β is the coefficient to be calibrated.

[0080] The dynamic equation of the EGR rate is as follows:

[0081]

[0082] In the formula, X EGR is the EGR rate.

[0083] Deriving Equation (7) gives:

[0084]

[0085] Among them, m Eng = ρ 2 V 22 ρ 2 is the density of the intake gas in the cylinder.

[0086] Step 2: Simplify the control-oriented model of the TVA-VGT-EGR diesel engine air system in Step 1. The simplification steps are as follows:

[0087] Simplify the dynamic equation of the diesel engine boost pressure to:

[0088]

[0089] In the formula,

[0090] Simplify the dynamic equation of the pressure in front of the turbine of the supercharger to:

[0091]

[0092] In the formula,

[0093] Simplify the dynamic equation of the EGR rate to:

[0094]

[0095] In the formula,

[0096] Step 3: Design a TITO ADRC controller for the simplified control-oriented model of the TVA-VGT-EGR diesel engine air system obtained in Step 2. The TITO ADRC controller includes a tracking differentiator TD, an ESO observer, and an error feedback control law SEF. The structure diagram of the controller is as shown in Figure 1 and Figure 2 shown.

[0097] According to equations (9), (10), and (11), the system state-space equation can be written as:

[0098]

[0099] where Expand f 1 , f 2 , f 3 into two states, and thus establish an ESO:

[0100]

[0101] where are the ESO estimated values of the supercharging pressure, the pressure before the turbine, and the EGR rate, respectively; are the ESO estimated values of the disturbances in the p 22 , p 3 and X EGR channels; l 1 , l 2 , l 3 , l 4 , l 5 , l 6 are the parameters to be tuned.

[0102] Adopt the pole-placement method to reduce the difficulty of parameter tuning, and place the ESO poles on ω 0 , to obtain the parameters to be tuned l 1 , l 2 , l 3 , l 4 , l 5 , l 6 :

[0103]

[0104] After the ESO converges, approximate p 22 , p 3 and X EGR respectively. Similarly, approximate f 1 , f 2 and f 3 respectively. If f 1 , f 2 and f 3 can be observed in real time, then the basic form of the control law SEF is as follows:

[0105]

[0106] where U 0 is the virtual control quantity and can be expressed in the form of a simplified proportional controller:

[0107] U 0 =K p (X des - X)(16)

[0108] In the formula, k 1 、k 2 、k 3 are parameters to be tuned, is the target value of the controlled variable. X des 、X′ des are the required values of the model parameters generated through the transition of the tracking differentiator (TD).

[0109] By combining Equation (15) and Equation (16), and introducing the feedforward of the model information and the feedforward of the change rate of the target value of the controlled variable, the complete control law SEF is obtained as shown in Equation (17):

[0110]

[0111] Among them, the X′ des term is the feedforward of the change rate of the target value of the controlled variable, and the - AX des - Q term is the feedforward term of the model information. The parameter K p is tuned using the pole placement method, and the poles are placed on the bandwidth ω c of the error feedback control law, and the parameters k 1 、k 2 、k 3 to be tuned are obtained as follows:

[0112]

[0113] To verify the control effect of the designed TITO ADRC algorithm, step tests of the fuel injection quantity are performed under the working conditions of the engine speed of 1000 rpm and 2200 rpm respectively. The results show that TITO ADRC can effectively decouple the air system.

[0114] Step 4, according to the control-oriented model and simulation results of the TVA-VGT-EGR diesel engine air system( Figure 5 ), determine the parameters to be learned, implement the design of the extremum search method, and optimize the control parameters;

[0115] By analyzing the model parameters (such as b 1 、b 2 、b 3 、b 4 、b 5 ) and the observer bandwidth ω 0Perform sensitivity analysis. According to the different degrees of influence on the overall control effect of the system, select the parameters that have the most significant and direct impact on the system control and are directly related, determine the parameters to be learned, and complete the design of the extreme value search learning algorithm (ES). For the purpose of realizing the online optimization of control parameters, the unknown model parameters in the control loop of the parameters to be learned (such as a 1 and a 2 ) in this example are identified using the method of forgetting factor recursive least squares (FFRLS).

[0116] In the air system decoupling controller, ω 0 and b i (i = 1, 2, 3, 4, or 5) are the two parameters that have the most significant and direct association with the system control. Different values of these two parameters will all have different degrees of influence on the overall control effect of the system.

[0117] To test the sensitivity of the system to the values of ω 0 and b i and confirm the necessity of learning these two parameters, a load step test is carried out at the rated speed of the controlled object. Among them, ω 0 and b i are respectively taken at values deviating from the optimal by 20%, 40%, and 60%, and the control effects of the system are compared respectively. As shown by the parameter sensitivity test result curve. The results show that the parameters that have the most obvious influence on the system control effect are ω 0 , b 1 , b 3 .

[0118] 1) Design the system tracking error cost function and use the extreme value search method to find the optimal parameter values:

[0119] Observe the parameter sensitivity analysis curve. The influence of the observer bandwidth ω 0 and the model parameter b i on the air system control effect is significantly different. Therefore, the optimization learning of the two parameters cannot be completed by designing a single error cost function. For the influence law of the model parameter b i on the control effect, the observer bandwidth ω 0 is more complex and it is more difficult to design the cost function.

[0120] The prerequisite for the extreme value search method to find the optimal parameter values is that there is a relationship similar to a quadratic function between the parameter and the cost function. Then, if a good convex function can be found between the parameter and the cost function (if it is a concave function, a negative sign needs to be added before the cost function), all that needs to be solved is the problem of adjusting the parameters. In this example, the TITO ADRC control simulation results show that as the observation bandwidth increases, the tracking error (X des - X) decreases, and the observation error slightly decreases, while the fluctuation of the observed disturbance becomes larger. Therefore, the observer bandwidth ω 0 optimization takes into account the tracking error, the observation error, and the second derivative of the observed disturbance, and their weighted sum is minimized.

[0121]

[0122] In the formula, K 1 is the gain coefficient; X des is the parameter target value, that is X is the actual parameter value, that is p 22 , p 3 ; Derive the output of the ESO observer; α, β, and γ are weight coefficients.

[0123] For the model input gain b i consider that the deviation between the ideal target trajectory and the actual trajectory is minimized, and the target trajectory is obtained through an ideal filter. The error cost function is designed as:

[0124] J(b i ) = -K 2 (X aim - X) 2 (20)

[0125] In the formula, K 2 is the gain coefficient, X is the actual parameter value, that is p 22 , p 3 . X aim is the required parameter value X ref filtered by an ideal filter, and X ref includes p 22ref , p 3ref .

[0126] 2) Determine the frequency of the high-frequency disturbance signal of this channel. The TITO ADRC parameter tuning problem can be simplified to a bandwidth tuning problem. In this system, the control law bandwidth ω c = 10Hz. The system response time τ = 1.33s. And the forced frequency ω is specified to be lower than the important system dynamics frequency and higher than the cut-off frequencies of the high-pass filter and the low-pass filter.

[0127] 3) Determine the amplitudes of the demodulation signal and the modulation signal. The amplitude b of the modulation signal should meet the requirements of exciting the dynamic change of the object model. At the same time, the amplitude of the demodulation signal is higher than that of the modulation signal by a >> b; the demodulation and modulation phases meet the conditions

[0128] Step 5, the unknown model parameters in the control loop of the parameter to be learned (for example, a in this example 1, a 2 ) Identify using the method of forgetting factor recursive least squares (FFRLS).

[0129] 1) FFRLS estimates the true output at the current moment based on the data generated by the system's past operations, and sends the extreme value search learning result b i into FFRLS to estimate the parameter a at the current moment i . The formula for the forgetting factor recursive least squares method is as follows:

[0130]

[0131]

[0132]

[0133] where λ is the forgetting factor, 0 ≤ λ ≤ 1, which is used to reduce the influence of the "old data" at the front of the input data on the current parameter estimation. The larger the forgetting factor, the greater the influence of the "old data" on the current estimation result; I is the identity matrix; is the estimate of the parameter matrix to be learned obtained through recursive calculation; Y is the input of FFRLS; k is the sampling number at the current moment.

[0134] 2) The value of the forgetting factor λ in the forgetting factor recursive least squares parameters is generally in the range of [0.98, 1], which is directly related to the stability of parameter identification. Although the corresponding results do not change much for different values, it can be shown from the results that the value of the forgetting factor does affect the model accuracy. The larger the forgetting factor, the more stable the parameter identification result, but the slower the response speed, the poor real-time performance, and the model accuracy is not the best; the smaller the forgetting factor, the greater the fluctuation of the parameter identification result, but the faster the response speed and the higher the model accuracy.

[0135] b 1 , b 3 The control loops where b 1 , b 3 are located are shown in formula (20). Send the learning results b 1 , a 2 into the FFRLS algorithm, so that only the parameters a

[0136]

[0137] Therefore, the FFRLS parameter learning expression is:

[0138]

[0139]

[0140] Among them, is the recognition result of a 1 , is the recognition result of δ 1 , is the recognition result of a 2 , is the recognition result of δ 2 . Considering the stability, real-time performance, and high precision of the model of the system, the forgetting factor is adjusted. In the example, λ = 0.995.

[0141] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system, characterized in that, it includes the following steps: Step 1, establish a control-oriented model of the TVA-VGT-EGR diesel engine air system composed of the core dynamic equation of the air system, the dynamic equation of the pressure before the turbine of the supercharger, and the dynamic equation of the EGR rate, and calculate the boost pressure p 22 , the pressure before the turbine p 3 and the EGR rate X EGR ; Step 2, facing the TVA-VGT-EGR diesel engine air system-oriented control model, design a multivariable active disturbance rejection TITO ADRC controller. The multivariable active disturbance rejection TITO ADRC controller includes a tracking differentiator TD, an ESO observer, and an error feedback control law SEF. Through the convergence of the ESO observer, the control of p 22 , p 3 and X EGR is achieved. The output of the ESO observer is transmitted to the error feedback control law SEF. Using the effective flow area A TVA , A VGT and A EGR respectively, the opening information is calculated to obtain U TVA , U VGT and U EGR , and they are sent to the TVA-VGT-EGR diesel engine air system-oriented control model; Step 3: According to the TVA-VGT-EGR diesel engine air system control-oriented model and the simulation results, perform a sensitivity analysis on the model input gain and control parameters. The control parameters include the bandwidth ω of the error feedback control law SEF c and the bandwidth ω of the ESO observer 0 , and select the parameters that have the most significant impact on system control and are directly related from the model input gain, ω c , ω 0 as the parameters to be learned; Step 4, design an error cost function, and use the extremum seeking method ES to obtain the optimal value of the parameter to be learned; Step 5, use the forgetting factor recursive least squares FFRLS method to identify the unknown model parameters in the control loop of the parameter to be learned.

2. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 1, characterized in that, the ESO observer in the said step 2 is: In the formula η vol is the charging efficiency, V d is the cylinder volume, N Eng is the engine speed, V 22 is the intake manifold volume, p 21 is the pressure in front of the TVA valve, T 22 is the intake air temperature, T 21 is the temperature in front of the TVA valve, σ1 is the coefficient to be calibrated, p 22 is the pressure behind the throttle valve, σ2 is the coefficient to be calibrated, p 3 is the pressure in front of the turbine, R is the ideal gas constant, T 3 is the temperature in front of the turbine, V 3 is the exhaust manifold volume, is the dynamic change of the fuel mass flow rate, A TVA is the effective flow cross-sectional area of the TVA valve, A EGR is the effective flow cross-sectional area of the EGR valve, T 3 is the temperature in front of the turbine, p 3 is the pressure in front of the turbine, A vGT is the effective flow cross-sectional area of the VGT valve, β is the coefficient to be calibrated, ρ 2 is the density of the intake gas in the cylinder; They are the ESO estimated values of the supercharging pressure, the pressure before the turbine, and the EGR rate respectively; They are p 22 , p 3 and the ESO estimated values of the disturbances in the X EGR channel; l 1 , l 2 , l 3 , l 4 , l 5 , l 6 are the parameters to be tuned; Configure the ESO poles to ω 0 to obtain the parameters l to be tuned 1 、l 2 、l 3 、l 4 、l 5 、l 6 :

3. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 2, characterized in that, the error feedback control law SEF is: Among them, k 1 and k 2 and k 3 are parameters to be tuned, is the target value of the controlled variable, is the feedforward of the change rate of the target value of the controlled variable; Configure the pole to the bandwidth ω of the error feedback control law c to obtain the parameters k 1 , k 2 , k 3 :

4. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 3, characterized in that, In step 3, the model input gain includes b 1 , b 2 , b 3 , b 4 , b 5 .

5. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 3, characterized in that, In step 4, the parameter to be learned is ω 0 , b 1 , b 3 .

6. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 1, characterized in that, In step 4, ω 0 has an error cost function as follows: where K 1 is the gain coefficient; X des is the parameter target value, that is X is the actual parameter value, that is p 22 , p 3 ; obtained by differentiating the output of the ESO observer ; α, β, and γ are weight coefficients.

7. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 1, characterized in that, In step 4, b i has an error cost function as follows: J(b i ) = -K 2 (X aim - X) 2 where K 2 is the gain coefficient, X is the actual value of the parameter, i.e., p 22 , p 3 , and X aim is the required value of the parameter X ref obtained by filtering through an ideal filter, and X ref includes p 22ref , p 3ref .

8. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 1, characterized in that, In step 5, the unknown model parameters in the parameter control loop to be learned are a 1 and a 2 .

9. The collaborative self-optimizing method for input gain and observation bandwidth of a multi-variable active disturbance rejection controller for a diesel engine air system according to claim 8, characterized in that, In step 5, the formula of the forgetting factor recursive least squares method is as follows: where λ is the forgetting factor, 0 ≤ λ ≤ 1; Send the learning result b 1 , b 3 into the forgetting factor recursive least squares FFRLS algorithm, so that only the parameter a to be identified 1 , a 2 ; Among them, is the recognition result of a 1 , is the recognition result of δ 1 , is the recognition result of a 2 , is the recognition result of δ 2 .

Images