Method for model-based control and regulation of a combustion engine

Abstract

A method for model-based control and regulation of a combustion engine (1) is proposed, wherein an injection system target value for actuating an injection system actuator is determined as a function of a target torque by means of a combustion model (4), wherein the combustion model (4) is adapted when the combustion engine (1) is running according to model values, wherein the model values ​​are calculated according to a first Gaussian process model for representing a basic grid and a second Gaussian process model for representing adaptation data points, wherein an optimizer (3) determines a minimized quality measure within a prediction range by varying the injection system target value and, when the minimized quality measure is found, the injection system target value is set to be decisive for setting an operating point of the combustion engine (1). The invention is characterized in that the model value is monitored with respect to a specified monotonicity.

Description

Technical Field

[0001] The invention relates to a method for model-based control and regulation of a combustion engine. Background Art

[0002] The behavior of a combustion engine is largely determined by the engine control unit according to the performance requirements. For this purpose, corresponding characteristic curves and characteristic maps are used in the software of the engine control unit. Using these characteristic curves and characteristic maps, the control variables of the combustion engine, such as the start of injection and the required rail pressure, are calculated according to the performance requirements. These characteristic curves / characteristic maps are provided with data by the manufacturer of the combustion engine during operation on a test bench. However, the coordination complexity caused by the multiple characteristic curves / characteristic maps of these characteristic curves / characteristic maps and the interaction of these characteristic curves / characteristic maps with one another is high.

[0003] Thus, in practice, attempts are made to reduce coordination costs by using mathematical models. Thus, for example, DE102018001727A1 describes a model-based method, in which the injection system target value for manipulating the injection system actuator is calculated by a combustion model according to the target torque and the gas path target value for manipulating the gas path actuator is calculated by a gas path model. Then, the optimizer calculates the quality measure according to the injection system target value and the gas path target value and changes these target values, with the aim of finding the minimum value within the prediction range. Then, in the case of finding the minimum value, the optimizer sets the injection system target value and the gas path target value to be decisive for setting the working point of the combustion engine. In addition, it is known from this reference that the combustion model is adapted according to the model value when the combustion engine is running, wherein the model value is then calculated by a first Gaussian process model for representing the basic grid and by a second Gaussian process model for representing the adaptation data point. In the case of a test bench test, it has now been shown that adaptation under unfavorable operating conditions may result in a local minimum for the optimization. The result of the optimization does not then correspond to a global optimum for the operation of the combustion engine. Summary of the invention

[0004] The object of the present invention is therefore to further develop the above-mentioned method with respect to higher quality.

[0005] This object is achieved by the method according to the invention for model-based control and regulation of a combustion engine. The design is presented in different specific embodiments.

[0006] The present invention proposes a method in which a model value is monitored with respect to a specified monotonicity. The method according to the present invention is a supplement to the practice known from DE 102018001727 A1. The model value is calculated according to a first Gaussian process model for representing a basic grid and a second Gaussian process model for representing an adaptation data point. Monotonicity is defined in the sense of an increasing trend of a positive target gradient for the model value or in the sense of a decreasing trend of a negative target gradient for the model value. The monotonicity is monitored by evaluating the gradient of the model value at the working point. In the case of a monotonic deviation, the monotonicity is corrected by smoothing the data points of the second Gaussian process model to achieve the monotonicity. In other words: the data points stored in the second Gaussian process model are moved by the smoothing until the monotonicity meets the regulations again. In the case of the reverse adaptation of the first Gaussian process model by the second Gaussian process model, the monotonic characteristics of the first Gaussian process model are kept unchanged.

[0007] By monitoring this monotonicity, the influence of measurement errors, i.e. inaccurate data values, for example, is significantly reduced. This ensures that the combustion model is physically correct and well behaved. Since the optimizer uses the combustion model, sufficiently accurate target values ​​for the injection system and the global optimum are guaranteed. In addition, the extrapolation capability of the combustion model remains unchanged. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The preferred embodiment is shown in the accompanying drawings.

[0009] Figure 1 A system diagram is shown;

[0010] Figure 2 A block diagram is shown;

[0011] Figure 3 A diagram is shown;

[0012] Figure 4 A table is shown;

[0013] Figure 5 A diagram showing the behavior of the model is shown;

[0014] Figure 6 A block diagram is shown; and

[0015] Figure 7 A program flow chart is shown. DETAILED DESCRIPTION

[0016] Figure 1A system diagram of a model-based, electronically controlled combustion engine 1, such as a diesel engine with a common rail system, is shown. The structure of the combustion engine and the function of the common rail system are known, for example, from DE 102018001727 A1. The input variables of the electronic control device 2 are presented with the reference symbols EIN and MESS. For example, the operator's performance requirements, a library for specifying the IMO's emission level MARPOL (Marine Pollution) or the emission level EU IV / Tier 4final, and the maximum mechanical component load are summarized under the reference symbol EIN. The performance requirements are usually specified as a target torque, a target speed, or an accelerator pedal position. The input variable MESS not only characterizes the directly measured physical quantity but also characterizes the auxiliary variables calculated therefrom. The output variables of the electronic control device 2 are the target values ​​for the lower-level control loop and the injection start SB and injection end SE.

[0017] Arranged within the electronic control device 2 are a combustion model 4, an adaptation 6, a smoothing 7, a gas path model 5 and an optimizer 3. Both the combustion model 4 and the gas path model 5 describe the system behavior of the combustion engine 1 as a mathematical equation.

[0018] The combustion model 4 statically depicts the process during combustion. In contrast, the gas path model 5 depicts the dynamic behavior of the air and exhaust gas guides. The combustion model 4 contains individual models, for example, for NOx and soot formation, for exhaust gas temperature, for exhaust gas mass flow and for peak pressure. These individual models are further specified according to the boundary conditions and injection parameters in the cylinder. The combustion model 4 is determined in the case of a reference combustion engine in a test bench operation, the so-called DoE test bench operation (DoE: Design of Experiments). In the case of the DoE test bench operation, the operating parameters and the control parameters are systematically changed with the aim of depicting the overall behavior of the combustion engine according to the engine parameters and the environmental boundary conditions. The combustion model 4 is supplemented with adaptation 6 and smoothing 7. The purpose of this adaptation is to adapt the combustion model to the actual behavior of the engine system. Smoothing 7 is further used to monitor and maintain monotonicity.

[0019] After activating the combustion engine 1, the optimizer 3 first reads in, for example, the emission level, the maximum mechanical component load and the target torque as performance requirements. The optimizer 3 then evaluates the combustion model 4, more precisely with respect to the target torque, emission limit values, environmental boundary conditions, such as the humidity phi of the charge air, the operating conditions of the combustion engine and the adaptation data points. The operating conditions are defined in particular by the engine speed, the charge air temperature and the charge air pressure. Now, the function of the optimizer 3 is to evaluate the injection system target values ​​for controlling the injection system actuators and the gas circuit target values ​​for controlling the gas circuit actuators. In this case, the optimizer 3 selects the solution in which the quality measure is minimized. The quality measure J is calculated as the integral of the target-actual squared difference within the prediction range. For example in the following form:

[0020] J=∫[w1(NOx(SOLL)-NOx(IST)] 2 +[w2(M(SOLL)-M(IST)] 2 +[w3(....)]+... (1)

[0021] Among them, w1, w2 and w3 represent corresponding weighting factors. As is known, the nitrogen oxide emission NOx is derived from the humidity of the charge air, the charge air temperature, the injection start SB and the rail pressure. Adaptation 9 intervenes in the real actual value, such as the NOx actual value or the exhaust gas temperature actual value. A detailed description of the quality measurement and the interruption criterion can be found in DE102018001727A1.

[0022] The mass measure is minimized by the optimizer 3 calculating a first mass measure at a first point in time, then changing the injection system target value and the gas circuit target value and predicting a second mass measure within the prediction range as a function of these target values. Subsequently, as a function of the deviation of the two mass measures from one another, the optimizer 3 specifies a minimum mass measure and sets it to be decisive for the combustion engine. For the example shown in the figure, this is the target rail pressure pCR(SL), the injection start SB and the injection end SE for the injection system. The target rail pressure pCR(SL) is the command variable for the lower rail pressure control loop 8. The manipulated variable of the rail pressure control loop 8 corresponds to the PWM signal for loading the intake throttle valve. The injector is loaded directly with the injection start SB and the injection end SE. For the gas circuit, the optimizer 3 determines the gas circuit target value indirectly. In the example shown, this is the lambda target value LAM(SL) and the AGR target value AGR(SL) for the lower lambda control loop 9 and the lower AGR control loop 10. In the case of using variable valve control, the gas path target value is adapted accordingly. The manipulated variables of the two control loops 9 and 10 correspond to the signal TBP for controlling the turbine bypass, the signal AGR for controlling the AGR actuator, and the signal DK for controlling the throttle valve. The returned measured variables MESS are read in by the electronic control device 2. These measured variables MESS should be understood as directly measured physical quantities and auxiliary variables calculated therefrom. In the example shown, the actual Lambda value and the actual AGR value are read in.

[0023] Figure 2The interaction of two Gaussian process models for adapting the combustion model and for specifying the model value E[X] is shown in a block diagram. Gaussian process models are known to those skilled in the art, for example from DE 102014225039 A1 or DE 102013220432 A1. Quite generally, Gaussian processes are defined by a mean function and a covariance function. The mean function is usually assumed to be zero or a linear / polynomial curve is used. The covariance function describes the relationship of any points. The first functional block 11 contains DoE data (DoE: Experimental Design) for the entire engine. These data are determined for a reference combustion engine during the operation of the test bench, in that all changes of the input variables within the range of the static driving of the combustion engine are determined within its entire adjustment range. These data characterize the behavior of the combustion engine within the range of the static driving with high precision. The second functional block 12 contains data obtained on a single-cylinder test bench. In the case of a single-cylinder test bench, those operating ranges that cannot be checked during the operation of the DoE test bench, such as high ground altitudes or extreme temperatures, can be set. These small amounts of measured data serve as the basis for parameterizing a physical model that reproduces the global behavior of combustion approximately correctly. The physical model approximately represents the behavior of the combustion engine under extreme boundary conditions. The physical model is completed by extrapolation so that the normal operating range is approximately correctly described. Figure 2 In the diagram, the model with extrapolation capability is denoted by reference numeral 13. From this model, a first Gaussian model 14 (GP1) for representing the basic grid is generated.

[0024] The merging of the two data point sets forms a second Gaussian process model (GP2) 15. The operating range of the fuel engine described by the DoE data is thereby also specified by these values, and the operating range for which no DoE data exist is reproduced by the data of the physical model. Since the second Gaussian process model 15 is adapted during operation, it is used to represent the adaptation point. Quite generally, i.e., see reference numeral 16, for the model value E[X] it applies:

[0025] E[X]=GP1+GP2 (2)

[0026] In this case, GP1 corresponds to a first Gaussian process model for representing the basic grid, GP2 corresponds to a second Gaussian process model for representing the adapted data points, and the model value E[X] corresponds to an input variable both for the smoothing and for the optimizer, such as an actual NOx value or an actual exhaust gas temperature value. Two information paths are shown by the double arrows in the figure. The first information path characterizes the provision of data from the first Gaussian process model 14 to the model value 16 of the basic grid. The second information path characterizes the reverse adaptation of the first Gaussian process model 14 by the second Gaussian process model 15.

[0027] exist Figure 3 In FIG. 1 , a first Gaussian process model for a single reservoir pressure pES is shown in a diagram, which is normalized to the maximum pressure pMAX. The measured NOx values ​​are plotted on the ordinate. Within the diagram, the DoE data values ​​determined at the full engine are characterized by a cross. The data points from the first Gaussian process model are presented as circles. These data points are generated in the following way: the trend is determined based on the data of the single-cylinder test bench and the DoE data is well mapped. For example, these are three data values ​​for points A, B and C. In the first step, the orientation of these data values ​​relative to each other, i.e., the trend information, is determined. Since a higher NOx actual value is obtained from the data value at point B than at point A, the function is monotonic within this range. This applies similarly to the data value at point C, that is, the NOx actual value at point C is higher than at point B. Thus, the trend information obtained for data values ​​A to C is: monotonic and linearly increasing. Then, in the second step, the deviation (model error) of these data values ​​relative to the DoE data is minimized. In other words: determine a mathematical function that maps the DoE data values ​​as well as possible taking into account the trend information. For data values ​​A, B and C this is a monotonic, linearly increasing function F1. Function F2 is characterized as monotonic only by data values ​​A, D and E. Function F3 maps by data values ​​A, F and G. See Figure 4 The measured variables single reservoir pressure pES, fuel mass mKrSt, injection start SB, rail pressure pCR and charge air temperature TLL shown by way of example are represented in accordance with function F1, i.e. monotonically and linearly increasing. The measured variable engine speed nIST is represented in accordance with function F3, i.e. unrestricted. "Unrestricted" means that there is no trend information for this measured variable. The charge air pressure pLL is represented in a monotonically decreasing manner. As can be seen from Figure 3 As can be seen, intermediate values, such as data value H, can be extrapolated. That is, the model has extrapolation capability ( Figure 2 : 13). The determination of the first Gaussian process model is carried out automatically, that is to say without expert knowledge. The automatic extrapolation capability of the model thus ensures a high degree of robustness, since in the unknown range the model is not susceptible to extreme or sudden reactions based on trend information.

[0028] Figure 5 A diagram is shown for the behavior of a combustion model. In this diagram, a first variable X is shown on the abscissa, for example the single reservoir pressure ( Figure 4:pES). A second variable Y, for example a NOx value, is shown on the ordinate. The curve of the first Gaussian process model GP1, i.e. the basic grid, depending on the first variable X and the second variable Y is shown as a dotted line with reference numeral 17. The dotted line 18 represents the curve of the model value E[X] in the initial state, i.e. without smoothing. The model value E[X] is calculated based on the sum of the first Gaussian process model and the second Gaussian process model. The solid line 19 represents the smoothed curve of the model value E[X]. The operating point of the variable X, i.e. the abscissa value AP, is plotted as a line 20 parallel to the ordinate.

[0029] about Figure 5 The further elaboration is based on the monotonicity with a positive increasing trend and a positive target gradient in the first Gaussian process model. In addition, it is stipulated for the model behavior that the monotonic characteristics of the first Gaussian process model must not be changed by the second Gaussian process model and that the monotonic characteristics at the current operating point, i.e. the operating point, are guaranteed. After the current operating point AP is detected, the model value E[X] corresponding to this operating point is calculated, here: E(AP). The model curve E[X] at the operating point E(AP) is then evaluated. At the operating point AP, the model value curve 18 shows a decreasing trend with a negative actual gradient. This behavior is caused by the local maximum of the model, which in turn causes a local minimum when calculating the quality measure. Therefore, the optimizer then calculates a manipulated variable that is inappropriate for the subordinate control loop based on the model value. In other words: the model value E[X] calculated according to the first Gaussian process model and the second Gaussian process model contradicts the required monotonic characteristics, so that the optimizer does not set the optimal operating point of the combustion engine.

[0030] The method according to the invention now provides that the monotonicity of the model values ​​is monitored and the combustion model is smoothed when a violation of the monotonicity is found. Specifically, this is achieved by changing the adapted data values ​​of the second Gaussian process model. As shown in the figure, the stored data points YD with coordinates (xD / yD) are thus changed in the direction of the basic grid (line 17). In this example, the abscissa values ​​remain constant. The changes relative to the original data points YD are implemented as small as possible. This can be described as the minimization of the squared difference of the smoothed data points in the following form:

[0031] minYG S(YD(i)-YG(i)) 2 (3) Considering the monotonic characteristics

[0032] In it, YD represents the stored data point, i represents the control variable and YG represents the smoothed data point at the position xD. That is, by means of relation (3), the stored data point YD and thereby the model value curve 18 for achieving the specified monotonic behavior are shifted in the direction of the curve 17 of the first Gaussian process model. In order to ensure that the predictions before and after smoothing are the same, an offset is used. See the figure for this.

[0033] Figure 6 The method is again shown in a block diagram. Here, the input variable is the variable MESS, which characterizes the current operating point. The output variable corresponds to the manipulated variable SG for the subordinate control loop. In the function block adaptation 6, the model value E[X] is calculated from the variable MESS and the already stored data points. This model value is determined by a first Gaussian process model for representing the basic grid and by a second Gaussian process model for calculating the adaptation data values. Figure 5 , in which the set of data points yD, the set of abscissa values ​​xD and the inverse covariance matrix inv(KD) are transferred from the adaptation 6 to the smoothing 7. By means of the smoothing 7, the specified monotonicity is monitored in accordance with the target gradient at the operating point and the combustion model is smoothed if a violation of the monotonicity is found. Subsequently, the smoothing 7 transfers the smoothed values ​​yG, the smoothed values ​​xG, the associated inverse covariance matrix inv(KG) and the corresponding offsets to the combustion model 4 and thereby to the optimizer 3.

[0034] exist Figure 7 In the present invention is shown in a program flow chart. This program flow chart is a supplement to the program flow chart known from DE 102018001727 A1. The measured value MESS is read in at S1, and the model value E[X] is calculated by the first Gaussian process model and the second Gaussian process model at S2, here: the model value E(AP) at the operating point. Then, the actual gradient at the operating point is determined at S3. Monotonicity is then checked based on the comparison of the target gradient and the actual gradient at S4. In the case of the same sign, the branch returns to point A. If a violation of monotonicity is identified at S4, the stored data point YD is changed to a smoothed data point YG at S5 for the purpose of making the gradient have the same sign and in the case of maintaining monotonicity by relation (3). Then, the offset is calculated at S6 and then a smoothed combustion model is generated at S7. The smoothed combustion model is then the input parameter of the optimizer, that is, it returns to the main program.

[0035] Reference numerals

[0036] 1 Combustion engine

[0037] 2 Electronic control equipment

[0038] 3 Optimizer

[0039] 4 Combustion model

[0040] 5 Gas path model

[0041] 6 Adaptation

[0042] 7 Smoothing

[0043] 8 Rail pressure regulation circuit

[0044] 9 Lambda Regulation Loop

[0045] 10 AGR regulation loop

[0046] 11 First functional block (DoE data)

[0047] 12 Second function block (data single cylinder)

[0048] 13 Model, with extrapolation capability

[0049] 14 The first Gaussian process model (GP1)

[0050] 15 Second Gaussian Process Model (GP2)

[0051] 16 Model Value

[0052] 17 Curve GP1

[0053] 18 Curve model value, initial state

[0054] 19 Curve model value, smoothed

[0055] 20 lines

Claims

1. A method for model-based control and regulation of a combustion engine (1), wherein an injection system target value for controlling an injection system actuator is determined based on a target torque by means of a combustion model (4), wherein the combustion model (4) is adapted when the combustion engine (1) is running based on a model value E[X], wherein the model value E[X] is calculated based on a first Gaussian process model (14) for representing a basic grid and a second Gaussian process model (15) for representing adaptation data points, wherein an optimizer (3) determines a minimized quality measure within a prediction range by changing the injection system target value and, when the minimized quality measure is found, the injection system target value is set to be decisive for setting the operating point of the combustion engine (1), characterized in that The model value E[X] is monitored with respect to a specified monotonicity, wherein the gradient of the model value E[X] at an operating point is evaluated in order to monitor the monotonicity.

2. The method according to claim 1, characterized in that The monotonicity is specified in the sense of having an increasing trend of positive target gradients with respect to the model value E[X].

3. The method according to claim 1, characterized in that The monotonicity is specified in the sense of having a decreasing trend of negative target gradients with respect to the model value E[X].

4. The method according to any one of claims 1 to 3, characterized in that In case a deviation from monotonicity is found, the monotonicity is corrected by smoothing the data points of the second Gaussian process model (15) to achieve a specified monotonicity.

5. The method according to any one of claims 1 to 3, characterized in that In addition to the monotonicity, the linear dependence of the input variables of the combustion model (4) on the model value E[X] is monitored.

6. The method according to any one of claims 1 to 3, characterized in that When the first Gaussian process model (14) is back-adapted by the second Gaussian process model (15), the monotonic properties of the first Gaussian process model (14) are maintained unchanged.

Images