JP5928299B2 - Optimization model construction method and construction apparatus - Google Patents

Description

  The present invention relates to an optimization model construction method and construction apparatus used when optimizing the control performance of a plant to be controlled using a mathematical optimization technique.

  For example, as described in Patent Document 1 (Japanese Patent Laid-Open No. 11-353298), a genetic algorithm is applied to control a target having nonlinear characteristics such as an engine, and evaluation of a plurality of individuals in the genetic algorithm is performed. When performing online, a local preliminary evaluation value is calculated for each subdivided evaluation region, and these are comprehensively evaluated to calculate a comprehensive evaluation value.

Japanese Patent Laid-Open No. 11-353298

  By the way, the present applicant calculates the controller output so as to optimize the control performance of the plant to be controlled using a mathematical optimization method, and uses this controller output to limit the control performance of the plant. We are researching the technology to calculate the controller and design the controller, but the following new problems were found in the research process.

  For example, in the case of a vehicle equipped with an engine, the fuel consumption characteristic of the engine after warm-up is a non-linear function with two inputs and one output that uses two variables of engine torque and engine speed as inputs and outputs fuel consumption. Can be expressed. However, since the characteristics of the two inputs and one output have strong non-linearity, when an optimization model is built by directly formulating a mathematical model, the calculation load (that is, the controller is optimized so as to optimize the control performance). There is a problem that the calculation load (when calculating the output) increases.

  In addition, since the fuel consumption characteristics of the engine are also affected by temperature, when the temperature change is relatively large, such as immediately after the cold start of the engine, the engine torque and the engine rotation speed are two input variables. There is a possibility that the error of optimization is large in the model for use. For this reason, when the temperature change is relatively large, such as immediately after the cold start of the engine, the temperature is also added as an input variable, and an optimization model is constructed that uses engine torque, engine speed, and temperature as input variables. However, if the input variables increase, the nonlinearity becomes stronger and the calculation load when the optimization is further increased. Therefore, the optimization calculation may not be completed in a practical time.

  As a countermeasure for these, there is a method of reducing calculation load by performing rough linear approximation (such as a linear model near the origin). However, if nonlinearity is omitted too much, a correct optimum solution may not be obtained. Even if the control performance is optimized using a heuristic technique such as a genetic algorithm that can be applied to a plant having complicated nonlinearity by using the technique of the above-mentioned Patent Document 1, the local performance is not limited. There is a possibility that only an optimal solution can be obtained.

  Therefore, the problem to be solved by the present invention is to reduce the computational load while ensuring the accuracy when optimizing the control performance of the plant to be controlled using the mathematical optimization method. There is.

In order to solve the above problems, the invention according to claim 1 is a method for constructing an optimization model used when optimizing the control performance of a plant to be controlled using a mathematical optimization method. When the optimization model is constructed by modeling the characteristics of the plant characteristics that use multiple parameters as multiple input variables, the relationship between at least two of the multiple input variables is constrained In this way, the number of input variables is reduced. Alternatively, as in claim 3, the time behavior of at least one of the plurality of input variables is stored or recorded in advance, and the input variable is set to a fixed value and the time behavior of the input variable is set. Based on the above, the number of input variables may be reduced by changing the fixed value at every predetermined timing.

  In this way, by reducing the number of input variables when building the optimization model, the nonlinearity of the optimization model can be moderately weakened, thereby ensuring the accuracy of optimization. However, the calculation load can be reduced while maintaining the optimization accuracy within an allowable range.

FIG. 1 is a diagram showing a schematic configuration of a system in an embodiment of the present invention. FIG. 2 is a flowchart showing the flow of processing when the number of input variables is reduced. FIG. 3 is a diagram for explaining an example of input variable reduction using constraint conditions. FIG. 4 is a diagram for explaining an example of input variable reduction by fixing input variables. FIG. 5 is a diagram illustrating an example of input variable reduction based on time behavior.

Hereinafter, an embodiment embodying a mode for carrying out the present invention will be described.
First, a schematic configuration of the system will be described with reference to FIG.

  A computer 11 is provided for calculating control performance of a plant to be controlled (for example, a vehicle equipped with an engine) and designing a controller. Connected to the computer 11 are an input device 12 such as a keyboard and a mouse, a storage medium drive device 13 for reading and writing data stored in a storage medium such as a DVD and a CD, a display device 14 such as a liquid crystal display, and the like.

Next, the calculation method of the limit value of the control performance of a plant and the design method of a controller are demonstrated.
First, an operator uses a mathematical model to optimize the characteristics of a plant (for example, a vehicle equipped with an engine) (for example, engine fuel consumption characteristics and battery loss characteristics) and constraints (for example, battery remaining capacity constraints). A model is constructed, this optimization model is expressed as a mathematical programming problem (for example, mixed integer programming problem), and control performance (for example, fuel consumption and exhaust gas amount) to be optimized is set as an objective function. The plant load output (for example, traveling load and vehicle speed) and environmental conditions (for example, temperature and wind speed), which are prerequisite conditions, are set. Data relating to these mathematical programming problems and data relating to load output and environmental conditions are input to the computer 11 by the operator by operating the input device 12 or the like (or created by the computer 11).

  Thereafter, the computer 11 calculates the plant controller output (for example, engine output) by solving the mathematical programming problem so as to optimize the objective function while satisfying the plant load output and the environmental conditions.

  Thereafter, when calculating the limit value of the control performance of the plant, load output (for example, traveling load and vehicle speed), environmental conditions (for example, temperature and wind speed), controller output (for example, engine output), etc. are input to the vehicle simulator. Then, the limit value of the control performance of the plant is calculated with this vehicle simulator.

  On the other hand, when designing a controller for a plant, load output (for example, travel load and vehicle speed), environmental conditions (for example, temperature and wind speed), controller output (for example, engine output), and the like are input to the vehicle simulator. A plant state other than the controller (for example, remaining battery level) is calculated by the vehicle simulator, and a plant controller is designed based on the controller output and the plant state.

  In this way, the controller output is calculated so as to optimize the control performance of the plant to be controlled using the mathematical optimization method, and the limit value of the control performance of the plant is calculated using this controller output. When designing the controller, if the number of input variables in the optimization model is large, the nonlinearity becomes stronger and the calculation load when the optimization is performed (that is, the controller output is calculated to optimize the control performance) There is a problem that the calculation load when doing so increases.

  As a countermeasure, in this embodiment, when an optimization model is constructed by modeling a characteristic having a plurality of plant characteristics as input variables, the number of input variables is determined based on the knowledge of the plant. I try to reduce it.

  Hereinafter, the procedure and processing contents when reducing the number of input variables will be described with reference to FIGS. In the present embodiment, the worker executes the processing of steps 101 to 107 in FIG.

  As shown in FIG. 2, first, in step 101, it is determined whether or not there are two or more input variables of the target function (function expressing the plant characteristics), and there are not two or more input variables ( In other words, if it is determined that there is only one input variable), it is determined that there is no need to reduce the number of input variables.

  On the other hand, if it is determined in step 101 that there are two or more input variables, the process proceeds to step 102 to determine whether the relationship between the two input variables can be constrained (for example, the relationship between the two input variables). Whether or not there is a condition for restraining.

  If it is determined in step 102 that the relationship between the two input variables can be constrained, the process proceeds to step 105, where input variable reduction using constraint conditions is executed. In this case, the number of input variables is reduced by constraining the relationship between the two input variables.

  For example, as shown in FIG. 3, when an optimization model is constructed by mathematically modeling fuel consumption characteristics of an engine after warm-up, generally (that is, before the input variable is reduced) The fuel consumption characteristic can be expressed by a function of two inputs and one output in which two variables of engine torque and engine speed are input and fuel consumption is output.

  Here, in a system that can arbitrarily control the engine operating point (combination of engine torque and engine speed), the line connecting the engine operating points at which the fuel consumption is minimized at each engine output is defined as the optimum operating line. There is one that controls the engine on the optimum operating line. On this optimum operating line, the combination of engine torque and engine speed is uniquely determined for each engine output.

  Paying attention to this point, the input of fuel consumption is output with two variables of engine torque and engine rotational speed as inputs, constraining the relationship between engine torque and engine rotational speed with the optimum operating line as the constraint condition The number of input variables is reduced by converting the output function into a 1-input 1-output function in which the engine output is an input and the fuel consumption is an output. An optimization model is constructed by formulating a 1-input 1-output function using the engine output as an input and fuel consumption as an output. In this way, the calculation load for optimization can be reduced by reducing the number of input variables without reducing the accuracy of optimization.

  When building an optimization model by modeling the power consumption characteristics of the motor, a line connecting the motor operating points (combination of motor torque and motor rotation speed) at which the power consumption is minimized at each motor output Is the optimum operation line, and the relation between the motor torque and the motor rotation speed is constrained using the optimum operation line as a constraint condition, and the two variables of the motor torque and the motor rotation speed are input, and the power consumption is output. The number of input variables may be reduced by converting the above function into a one-input one-output function using the motor output as an input and power consumption as an output.

  On the other hand, if it is determined in step 102 in FIG. 2 that the relationship between the two input variables cannot be constrained, the process proceeds to step 103 without executing the processing of step 105, and one input variable is set to a fixed value. When the number of input variables is reduced by setting, it is determined whether or not the optimization accuracy using the optimization model can be ensured (the optimization accuracy can be maintained within an allowable range). In this case, for example, it is determined whether or not optimization accuracy can be ensured by comparing the output of the function when one input variable is set to a fixed value and when it is not set to a fixed value.

  If it is determined in step 103 that the accuracy of optimization can be ensured even if one input variable is set to a fixed value and the number of input variables is reduced, the process proceeds to step 106 and input by fixing the input variable is performed. Perform variable reduction. In this case, the number of input variables is reduced by setting one input variable to a fixed value.

  For example, as shown in FIG. 4, when an optimization model is constructed by modeling a battery loss characteristic as a mathematical model (in general, before the input variable is reduced), the battery loss characteristic is represented by battery voltage and battery current. These two variables can be expressed as a function of two inputs and one output with battery loss as an output.

  Here, in the system that controls the charge / discharge amount of the battery so that the remaining battery level is within the predetermined usage range, the battery voltage changes within the usage range of the remaining battery level. Get smaller.

  Focusing on this point, the average voltage Vave, which is the average value of the battery voltage within the usage range of the remaining battery power, is obtained, and the battery voltage, which is one of the input variables, is set to the average voltage Vave (fixed value). A 2-input 1-output function that outputs battery loss as input with two variables of battery voltage and battery current, 1-input 1-output that outputs battery loss with battery power (= battery current x average voltage Vave) as input The number of input variables is reduced by approximating the function to An optimization model is constructed by formulating a function of one input and one output having the battery power as an input and the battery loss as an output. In this way, it is possible to easily reduce the number of input variables and reduce the calculation load of optimization while ensuring the accuracy of optimization (while maintaining the accuracy of optimization within an allowable range). .

  On the other hand, if it is determined in step 103 of FIG. 2 that the accuracy of optimization cannot be ensured by setting one input variable to a fixed value and reducing the number of input variables, the processing of step 106 is executed. Without proceeding to step 104, it is determined whether or not the time behavior of one input variable is available at the time of optimization. In this case, for example, the time behavior of one input variable is measured in advance and stored or recorded several times, and whether or not the time behavior is stable every time (whether or not the same behavior is shown every time). To determine whether it can be used during optimization.

  If it is determined in step 104 that the time behavior of one input variable can be used at the time of optimization, the process proceeds to step 107, and input variable reduction based on the time behavior is executed. In this case, the number of input variables can be reduced by setting one input variable for which time behavior is stored or recorded in advance to a fixed value and changing the fixed value at each predetermined timing based on the time behavior of the input variable. To do.

  For example, as shown in FIG. 5, when an optimization model is constructed by mathematically modeling fuel consumption characteristics of an engine immediately after a cold start, generally (that is, before reducing input variables), The fuel consumption characteristics of the engine can be expressed by a three-input one-output engine model in which three variables of engine torque, engine rotation speed, and engine coolant temperature are input and fuel consumption is output.

  Here, the time behavior of the engine coolant temperature immediately after the cold start has a characteristic of showing almost the same behavior every time. Paying attention to this point, the time behavior of the engine coolant temperature immediately after the cold start is measured in advance and stored or recorded (for example, a map that defines the relationship between the elapsed time after the engine start and the engine coolant temperature is stored in the computer) 11 ROM).

  When the optimization is actually performed, at the first optimization calculation timing T1, the engine cooling water temperature corresponding to the calculation timing T1 is set to a fixed value by referring to the time behavior of the engine cooling water temperature stored or recorded in advance. C1 is set, the current engine coolant temperature is set to a fixed value C1, and the fuel consumption characteristic of the engine at the engine coolant temperature C1 is expressed as a function of one input and one output with the engine output as an input and the fuel consumption as an output. Thus, the number of input variables is reduced. An optimization model is constructed by formulating a 1-input 1-output function using the engine output as an input and fuel consumption as an output.

  At the calculation timing T2 of the second optimization, the engine cooling water temperature corresponding to the calculation timing T2 is set to a fixed value C2 with reference to the time behavior of the engine cooling water temperature stored or recorded in advance, and the current engine cooling water temperature is set to a fixed value. The number of input variables is reduced by expressing the fuel consumption characteristic of the engine at the engine cooling water temperature C2 as a function of one input and one output with the engine output as an input and the fuel consumption as an output. An optimization model is constructed by formulating a 1-input 1-output function using the engine output as an input and fuel consumption as an output.

  Similarly, at the calculation timing Tn for the n-th optimization, the engine cooling water temperature corresponding to the calculation timing Tn is set to a fixed value Cn with reference to the time behavior of the engine cooling water temperature stored or recorded in advance, and the current engine cooling is performed. By setting the water temperature to a fixed value Cn and expressing the fuel consumption characteristics of the engine at the engine cooling water temperature Cn as a function of one input and one output with the engine output as an input and the fuel consumption as an output, the number of input variables Reduce. An optimization model is constructed by formulating a 1-input 1-output function using the engine output as an input and fuel consumption as an output. In this way, it is possible to reduce the number of input variables and reduce the calculation load of optimization while reducing the decrease in optimization accuracy compared to the case where the engine coolant temperature is always set to a fixed value. it can.

  Note that if the time behavior of the engine coolant temperature stored or recorded in advance is significantly different from the time behavior of the engine coolant temperature when the optimization is actually performed, the correct optimum solution may not be obtained. If the time behavior of the engine cooling water temperature is updated after optimization and the optimization is repeatedly performed again, the optimal optimal solution can be approached.

  On the other hand, if it is determined in step 104 that the time behavior of one input variable cannot be used during optimization (that is, if all the determinations in steps 102 to 104 are “No”), the number of input variables It is judged that it cannot be reduced, and the process ends.

  In the present embodiment described above, when an optimization model is constructed by mathematically modeling characteristics (for example, engine fuel consumption characteristics and battery loss characteristics) having a plurality of plant characteristics as input variables. Since the number of input variables is reduced based on plant knowledge (eg whether input variables can be constrained or fixed, temporal behavior, etc.), the nonlinearity of the optimization model should be moderately weakened As a result, the calculation load can be reduced while ensuring the accuracy in optimization (while maintaining the optimization accuracy within an allowable range). In addition, by using a method that can obtain a global optimal solution (such as an interior point method or a branch-and-bound method) for solving a mathematical optimization problem (mathematical programming problem), it is possible to prevent a local optimal solution from being obtained. .

  In the above embodiment, the worker executes the processing of steps 101 to 107 in FIG. 2. However, the present invention is not limited to this, and the computer 11 executes the processing of steps 101 to 107 in FIG. 2. May be. In this case, the processing in steps 105 to 107 serves as an input variable reduction means in the claims, and the processing in step 103 serves as determination means in the claims. Further, it is not always necessary to execute all the processes of steps 101 to 107 on the computer 11, and a part of the processes (for example, the processes of steps 101 and 102) may be executed by the worker.

  The plant to which the present invention can be applied is not limited to a vehicle using only an engine as a power source. For example, the present invention is applied to a hybrid vehicle using an engine and a motor as power sources and an electric vehicle using only a motor as a power source. It may be applied. Further, the present invention may be applied to plants other than vehicles (automobiles), for example, ships, construction machines, railway vehicles, cogeneration, and the like.

  DESCRIPTION OF SYMBOLS 11 ... Computer (input variable reduction means, determination means), 12 ... Input device, 13 ... Storage medium drive device, 14 ... Display device

Claims (5)

A method for constructing an optimization model used when optimizing the control performance of a plant to be controlled using a mathematical optimization method,
In constructing the model for optimization plurality of properties that the plurality of input variables parameters and mathematical expression model among the characteristics of the plant, at least two input variables relationship among the plurality of input variables A method for constructing an optimization model, characterized in that variable reduction processing is performed to reduce the number of input variables by restraining . After reducing the number of the input variables by the variable reduction process, the time behavior of at least one input variable of the plurality of input variables is stored or recorded in advance, and the input variable is set to a fixed value. 2. The variable reduction processing according to claim 1, further comprising: performing another variable reduction process for reducing the number of the input variables by setting and changing the fixed value at each predetermined timing based on the time behavior of the input variables. How to build a model for optimization. A method for constructing an optimization model used when optimizing the control performance of a plant to be controlled using a mathematical optimization method,
The time behavior of at least one input variable among the plurality of input variables when the optimization model is constructed by modeling the characteristics having the plurality of parameters of the plant characteristics as the plurality of input variables. the previously stored or recorded, a variable reduction to reduce the number of the input variables by changing the fixed value for each predetermined timing based on the time behavior of the input variables and sets the input variable to a fixed value method for constructing a model for optimal reduction you and performs processing. In terms of a reduced number of the input variables by the variable reduction process, further, another variable to reduce the number of the input variables by setting at least one of the input variables of the plurality of input variables to a fixed value method for constructing a model for optimization of any one of claims 1 to 3, characterized in that the reduction process. An apparatus for constructing an optimization model for use in optimizing the control performance of a plant to be controlled using a mathematical optimization method,
An input for reducing the number of input variables based on the knowledge of the plant when the optimization model is constructed by modeling the characteristics having a plurality of parameters among the characteristics of the plant as a plurality of input variables. Variable reduction means (11) ;
Determination means (11) for determining whether or not the accuracy of optimization using the optimization model can be maintained within an allowable range even if the number of input variables is reduced,
The input variable reduction means (11) reduces the number of input variables when the determination means (11) determines that the accuracy of the optimization can be maintained within an allowable range. Model building device.

Images