JP4186785B2 - Driver model construction method, operation prediction method, and operation prediction apparatus - Google Patents

Description

  The present invention relates to a driver model construction method for constructing a driver model capable of estimating a driver (driver) operation, and more specifically, a driver model capable of estimating an operation based on a judgment of a driver situation. The present invention relates to a driver model construction method for construction, an operation prediction method using the driver model, or an operation prediction apparatus using the driver model.

  As a conventional driver model for estimating a driver's operation, for example, Patent Document 1 calculates a deceleration start inter-vehicle distance for following a preceding vehicle under a constant deceleration condition, that is, traveling following the preceding vehicle. A driver model for estimating at which timing a driver performing a deceleration operation is disclosed. This driver model is a model that follows a preceding vehicle by a deceleration operation based on input information from each sensor.

  As described above, in the prior art, a technique for constructing a “single driver model” for estimating a driver operation based on input information from each sensor has been established.

JP-A-8-132931 (paragraph numbers 17 to 26, FIGS. 2 and 3)

  However, the driving characteristics of the driver originally consist of a driver's individual situation determination mechanism and steering and accelerator / brake operations. Specifically, the above conventional driver model accurately controls the operation timing so that a driver traveling following the preceding vehicle performs a deceleration operation (braking operation) following the start of deceleration of the preceding vehicle. Although it is estimated, for example, it is not a “multi-driver model”, that is, even if another operation due to other factors is required, it is not possible to shift to a driver model that performs another operation by situation judgment This makes it impossible to predict operation with higher accuracy.

  As described above, the conventional driver model can accurately predict the operation when it is only intended to follow the preceding vehicle, but the driver model is not configured to switch a plurality of driver models according to the situation determination. There was a problem that it was not possible to accurately predict the behavior of the.

  The present invention has been made in view of the above, and a driver's action is composed of a plurality of driver models (corresponding to sub driver models of embodiments described later), and includes a sub driver model switching determination condition. An object of the present invention is to provide a driver model construction method for constructing a driver model, an operation prediction method and an operation prediction device using the driver model construction method or the driver model.

The operation prediction method according to the present invention constructs a driver model by combining a plurality of motion models (each represented as a sub-driver model) representing the operation of the driver, and further, the individual sub-driver is added to the driver model. An operation prediction method using a driver model construction method including a model switching determination condition, for example, based on predetermined driving data obtained as information recognized by a driver to perform an operation, An auxiliary theoretical variable (binary variable) and auxiliary continuous variables are introduced by a polynomial structure determination step (corresponding to step S1) for determining the structure of each polynomial constituting the system theory based on MLDS (Mixed Logical Dynamical System). The switching judgment condition and the sub-driver model are A linearization step expressed by an equation and a linear state equation (corresponding to step S2 and step S3), a driver operation based on the linear inequalities and the linear state equations, and further, the constraints on the auxiliary theoretical variables A parameter that defines the switching judgment condition and a kinematic parameter that constitutes the polynomial are mixed integers that minimize the integral value of the difference between the output of the polynomial representing the operation actually observed and A parameter identification step for searching by linear programming (corresponding to step S4), predicting the timing of the sub-driver model based on the parameter for defining the switching judgment condition, and simultaneously calculating the polynomial of the sub-driver model an operation amount of the driver of the operation based on the kinematic parameters constituting A prediction step of measuring (corresponding to step S12), and comprising a.

  According to the present invention, the sub-driver model switching timing and the driver's action (operation) are predicted while the driver model is updated in time series. Thereby, the operation of the driver can be predicted more accurately.

The operation predicting device according to the present invention constructs a driver model by combining a plurality of motion models (each represented as a sub-driver model) representing a driver's operation, and further, An operation prediction apparatus using a driver model construction process including a sub-driver model switching determination condition, for example, based on predetermined operation data obtained as information recognized by a driver to perform an operation An auxiliary theoretical variable (2) is determined by a polynomial structure determining means for determining the structure of each polynomial constituting the model (corresponding to a polynomial structure determining unit 11 in an embodiment described later), and a system theory based on MLDS (Mixed Logical Dynamical System). Value variable) and auxiliary continuous variables, the switching judgment condition and the support variable are introduced. Linearizing means for expressing the driver model with linear inequalities and linear state equations (corresponding to the linear inequalities / equation generation unit 12), the linear inequalities and the linear state equations, and further the constraints on the auxiliary theoretical variables Based on the parameter for defining the switching judgment condition and the kinematics constituting the polynomial, which minimize the integral value of the difference between the output of the polynomial representing the operation of the driver and the actually observed operation At the same time as predicting the timing of the sub-driver model based on the parameter identification means for searching for a specific parameter by the mixed integer linear programming (corresponding to the parameter identification unit 13) and the parameter for defining the switching judgment condition , Based on kinematic parameters that make up the polynomial of the subdriver model Characterized in that and an operation prediction unit for predicting an operation amount of the driver operation.

  According to the present invention, the driver model is updated in time series, and the switching timing of the sub driver model and the behavior (operation) of the driver are predicted. Thereby, the operation prediction apparatus which can predict a driver's operation more correctly can be provided.

  In the driver model construction method according to the present invention, the driver model is constructed by combining a plurality of sub-driver models, and further, the driver model includes the sub-driver model switching judgment condition. It is possible to accurately predict the conditions for switching the sub-driver model and the behavior (operation) of the driver. That is, it is possible to provide a driver model that more accurately reflects the driver's behavior.

  Embodiments of a driver model construction method, an operation prediction method using the built driver model, and an operation prediction device according to the present invention will be described below in detail with reference to the drawings. Note that the present invention is not limited to the embodiments.

  FIG. 1 is a diagram showing a configuration of a driver model construction unit for realizing a driver model construction method according to the present invention. In the driver model construction unit 1 shown in FIG. 1, a driver's behavior, that is, a driver model is constructed by combining a plurality of motion models (each of which is expressed as a sub-driver model in the following), and the driver model Are characterized by including switching determination conditions (driver status determination) of individual sub-driver models. In order to realize the construction of such a driver model, the driver model construction unit 1 according to the present invention calculates the structure of each polynomial constituting the sub-driver model accompanied by the situation determination based on predetermined operation data received from the sensor. A polynomial structure determination unit 11 to determine, a linear inequalities / equation generation unit 12 that expresses the subdriver model accompanied by the above-described situation determination with linear inequalities and linear state equations, and a mixed integer linear programming method. A parameter identification unit 13 for simultaneously determining a parameter indicating switching and a kinematic parameter;

  Here, the operation of the driver model construction unit 1 configured as described above will be described in detail with reference to the drawings. FIG. 2 is a flowchart showing a driver model construction method according to the present invention.

  First, in the polynomial structure determination unit 11, each polynomial constituting a sub-driver model accompanied by a situation determination based on predetermined driving data (various information recognized by the driver for operation, such as distance, speed, and acceleration). Is determined (step S1). FIG. 3 is a diagram showing a sub-driver model (corresponding to the equation (1)) accompanied by a situation determination, which is expressed in a PWPS (Piece-wise Polynomial Systems) format. PWPS is a kind of hybrid system and represents a system in which a polynomial (corresponding to a sub-driver model) and a logical condition (corresponding to a judgment condition) are integrated.

In addition, y _{i, k} constituting the polynomial of the equation (1) shown in FIG. 3 represents an output (the amount of operation that acts on the vehicle) indicating the driver's action (operation), and i is the type of output (Accelerator operation, brake operation, steering operation, etc.), k represents the passage of time, x ₁ , x ₂ ,... Are variable driving data depending on the type of i, and the driver performs the operation in detail. Various information to be recognized (for example, if the action is “stop”, the distance to the stop line, the speed, the acceleration, the distance to the vehicle ahead, the deviation from the center line, etc.), a _{0 to} a ₃ , b _{0 to} b ₃ , c _{0 to} c ₃ ,... represent kinematic parameters. However, in the above polynomial, it is necessary to determine in advance which one is to be used as the operation data (x ₁ , x ₂ ,...). The number of the parameters a, b, c... Is arbitrarily determined according to the polynomial structure.

Further, d ₁ and d ₂ constituting the judgment condition of the expression (1) shown in FIG. 3 represent parameters indicating switching of the sub driver model, and M _d represents the maximum value of the parameter indicating switching of the sub driver model. M _d represents the minimum value of the parameter indicating the switching of the sub-driver model, ε represents a small integer, and x _{j, k} represents operation data serving as a determination criterion for making a situation determination. The operation data x _{j, k} may be a single operation data or a composite value of the operation data.

  Further, in this embodiment, as an example, a driver model capable of switching three sub-driver models according to the situation is constructed. However, the present invention is not limited to this, and two or four or more sub-driver models are included. It is good also as constructing the driver model which can be switched according to a situation.

  Next, the linear inequality / equation generation unit 12 converts the judgment condition part of the equation (1) shown in FIG. 3 into a linear inequality (step S2), and further configures each sub-driver model shown in FIG. The piecewise polynomial system is converted to a linear polynomial system with inequality constraints with binary variables (step S3). In the following, using system theory based on MLDS (Mixed Logical Dynamical System), in detail, auxiliary theory variables (corresponding to binary variables described later): δ∈ {0, 1} and auxiliary continuous variables (described later) The case where the sub-driver model accompanied by the situation determination shown in FIG. 3 is expressed by a linear inequality and a linear state equation will be specifically described.

  Note that the procedure for converting the judgment condition part into a linear inequality will be described for the sake of simplicity. First, the case where the two judgment conditions are expressed by a linear inequality will be described, and then (1) shown in FIG. A case where the judgment condition part of the expression is expressed by a linear inequality will be described.

For example, when the determination condition part (if) shown in the following expression (2) is expressed by a linear inequality, the linear inequality / equation generation unit 12 introduces a binary variable δ, and the determination condition part is expressed by the following expression (3). Express as follows.
a ₀ x _{1, k} + a ₁ x _{2, k} , if x _{j, k} ≧ d ₁
b ₀ x _{1, k} + b ₁ x _{2, k} , if x _{j, k} ≦ d ₁ −ε
m _d ≦ x _{j, k} ≦ M _d
... (2)
[X _{j, k} ≧ d ₁ ] ⇔ [δ = 1]
[X _{j, k} ≦ d ₁ −ε] ⇔ [δ = 0]
... (3)
The above equation (3) can be expressed as shown in FIG. 4 when the vertical axis is δ and the horizontal axis is x _{j, k} .

The linear inequality / equation generator 12 formulates a region sandwiched by straight lines ss ′ and tt ′ as shown in FIG. 5 under the constraint that δ = {1, 0}. The above equation (3) is replaced with the linear inequality shown in the following equations (4) and (5). Note that if 0 or 1 is substituted for δ in the following equations (4) and (5), it can be easily confirmed that they are equivalent to the above equation (3).
−x _{j, k} + (d ₁ −m _d ) δ + m _d ≦ 0 (4)
x _{j, k} − (M _d −d ₁ + ε) δ−d ₁ + ε ≦ 0 (5)

Further, in the linear inequality / equation generation unit 12, for example, even when the determination condition shown in the equation (1) of FIG. 3 is expressed by a linear inequality, the binary variables δ _{1, k} , δ _{2, k} are the same as above. And the judgment condition part is expressed as the following equation (6).
[D ₁ −ε ≧ x _{j, k} ≧ m _d ] ⇔ [δ _{1, k} = 0, δ _{2, k} = 0]
[D ₂ −ε ≧ x _{j, k} ≧ d ₁ ] ⇔ [δ _{1, k} = 1, δ _{2, k} = 0]
[M _d ≧ x _{j, k} ≧ d ₂ ] ⇔ [δ _{1, k} = 0, δ _{2, k} = 1]
(6)

Then, the linear inequality / equation generator 12 converts the above equation (6) into the following (7) under the constraint that δ _{1, k} = {0,1}, δ _{2, k} = {0,1}. ) And inequality shown in equation (8).
x _{j, k} ≦ d ₁ (1− (δ _{1, k} + δ _{2, k} )) + d ₂ δ _{1, k} −ε + δ _{2, k} (M _d + ε)
... (7)
x _{j, k} ≧ (1− (δ _{1, k} + δ _{2, k} )) m _d + d ₁ δ _{1, k} + d ₂ δ _{2, k}
(8)

If a product term δf (x) appears in the process of model construction, it is converted into a linear expression by introducing an auxiliary variable z = δf (x). That is, the relationship z = δf (x) can be converted into a logical expression such as the following expression (9), and finally, the following expressions (10), (11), (12), (13) It can be converted into the linear inequality shown in the equation.
[Δ = 0] ⇒ [z = 0]
[Δ = 1] ⇒ [z = f (x)]
... (9)
z ≦ Mδ (10)
−z ≦ −mδ (11)
z ≦ f (x) −m (1−δ) (12)
−z ≦ −f (x) + M (1−δ) (13)

Specifically, in the above equations (7) and (8), the auxiliary variable z is defined as the following equations (14), (15), (16), and (17), and the above (7 ) And (8) are substituted into the following expressions (18) and (19).
z _{9, k} = δ _{1, k} f _{9, k} = δ _{1, k} d ₁ (14)
z _{10, k} = δ _{1, k} f _{10, k} = δ _{1, k} d ₂ (15)
z _{11, k} = δ _{2, k} f _{11, k} = δ _{2, k} d ₁ (16)
z _{12, k} = δ _{2, k} f _{12, k} = δ _{2, k} d ₂ (17)
x _{j, k} ≦ d ₁ −z _{9, k} + z _{10, k} −z _{11, k} −ε + δ _{2, k} (M _d + ε) (18)
x _{j, k} ≧ (1- (δ _{1, k} + δ _{2, k} )) m _d + z _{9, k} + z _{12, k} (19)

  After executing the processing in step S2, the linear inequality / equation generator 12 then converts the piecewise polynomial system constituting each sub-driver model shown in FIG. 3 into an inequality-constrained linear polynomial system with binary variables. (Step S3).

Specifically, the above formula (1) can be expressed as the following formula (20) by using the above formula (6).
y _{i, k} = (a ₀ x _{1, k} + a ₁ x _{2, k} + a ₂ (x _{1, k} ) ² + a ₃ (x _{2, k} ) ² )
× (1- (δ _{1, k} + δ _{2, k} ))
+ (B ₀ x _{1, k} + b ₁ x _{2, k} + b ₂ (x _{1, k} ) ² + b ₃ (x _{2, k} ) ² ) δ _{1, k}
+ (C ₀ x _{1, k} + c ₁ x _{2, k} + c ₂ (x _{1, k} ) ² + c ₃ (x _{2, k} ) ² )
... (20)

When the above equation (20) is expanded, the following equation (21) is obtained.
y _{i, k} = a ₀ x _{1, k} + a ₁ x _{2, k} + a ₂ (x _{1, k} ) ² + a ₃ (x _{2, k} ) ²
+ (B ₀ −a ₀ ) x _{1, k} δ _{1, k} + (b ₁ −a ₁ ) x _{2, k} δ _{1, k}
+ (B ₂ −a ₂ ) (x _{1, k} ) ² δ _{1, k} + (b ₃ −a ₃ ) (x _{2, k} ) ² δ _{1, k}
+ (C ₀ −a ₀ ) x _{1, k} δ _{2, k} + (c ₁ −a ₁ ) x _{2, k} δ _{2, k}
+ (C ₂ −a ₂ ) (x _{1, k} ) ² δ _{2, k} + (c ₃ −a ₃ ) (x _{2, k} ) ² δ _{2, k}
... (21)

Finally, in the linear inequality / equation generation unit 12, the auxiliary variable z is defined as in the following formulas (22) to (29) and substituted into the above formula (21) in the same manner as described above. 30) to obtain the formula.
z _{1, k} = δ _{1, k} f _{1, k} = δ _{1, k} (b ₀ −a ₀ ) (22)
z _{2, k} = δ _{1, k} f _{2, k} = δ _{1, k} (b ₁ −a ₁ ) (23)
z _{3, k} = δ _{1, k} f _{3, k} = δ _{1, k} (b ₂ −a ₂ ) (24)
z _{4, k} = δ _{1, k} f _{4, k} = δ _{1, k} (b ₃ −a ₃ ) (25)
z _{5, k} = δ _{1, k} f _{5, k} = δ _{2, k} (c ₀ −a ₀ ) (26)
z _{6, k} = δ _{1, k} f _{6, k} = δ _{2, k} (c ₁ −a ₁ ) (27)
z _{7, k} = δ _{1, k} f _{7, k} = δ _{2, k} (c ₂ −a ₂ ) (28)
z _{8, k} = δ _{1, k} f _{8, k} = δ _{2, k} (c ₃ −a ₃ ) (29)
y _{i, k} = a ₀ x _{1, k} + a ₁ x _{2, k} + a ₂ (x _{1, k} ) ² + a ₃ (x _{2, k} ) ²
_{+ X 1, k z 1,} k + x 2, k z 2, k + (x 1, k) 2 z 3, k
+ (X _{2, k} ) ² z _{4, k} + x _{1, k} z _{5, k} + x _{2, k} z _{6, k}
+ (X _{1, k} ) ² z _{7, k} + (x _{2, k} ) ² z _{8, k}
... (30)

That is, in the above (1), the processing of step S2 and step S3 is performed by the expressions (10), (11), (12), (13), (18), (19), (30). It can be expressed by linear inequalities and linear equations. In the present embodiment, as an example, the case where there are one type of binary variable (δ) and the case of two types (δ _{1, k} , δ _{2, k} ) has been described. The variable δ is appropriately set according to the number of sub driver models constituting the driver model according to the present invention.

Next, in the parameter identification unit 13, parameters d ₁ and d ₂ representing switching of judgment by the mixed integer linear programming, kinematic parameters a _{0 to} a ₃ , b _{0 to} b ₃ , c _{0 to} c ₃ The binary variables δ _{1, k} and δ _{2, k} are simultaneously determined (step S4). Specifically, the above expressions (10), (11), (12), (13), (18), (19), (30), and δ _{1, k} = { 0, 1}, δ _{2, k} = {0, 1}, 0 ≦ δ _{1, k} + δ _{2, k} ≦ 1, δ _{2, k} ≦ δ _{2, k + 1,} etc. Optimal parameters d ₁ , d ₂ , a _{0 to} a ₃ , b ₀ that minimize the integral value of the difference between the output y _{i, k} indicating the action (operation) and the actually observed y ′ _{i, k.} ˜b ₃ , c _{0 to} c ₃ , δ _{1, k} , δ _{2, k} are searched by mixed integer linear programming.

FIG. 6 is a diagram showing an example of the parameters d ₁ and d ₂ , that is, the conditions for switching the sub driver model, predicted by executing the driver model construction method.

  As described above, in this embodiment, the driver's behavior, that is, the driver model according to the present invention is constructed by combining a plurality of sub-driver models, and further, switching determination of the sub-driver model is performed on the driver model. Conditions were included. Accordingly, it is possible to accurately predict the conditions for switching the sub-driver model and the behavior (operation) of the driver. That is, it is possible to provide a driver model that more accurately reflects human behavior.

FIG. 7 is a diagram illustrating a configuration of an operation prediction device using the driver model construction method in the first embodiment. The operation prediction device 21a includes the driver model construction unit 1 in the first embodiment, and a driver. A model switching point prediction unit 32 that predicts / outputs a switching point (timing) of the sub-driver model based on parameters d ₁ and d ₂ that are outputs of the model construction unit 1, and a parameter a that is an output of the driver model construction unit 1 An operation predicting unit 31 that predicts / outputs a driver operation (brake operation, accelerator operation, steering operation) based on _{0 to} a ₃ , b _{0 to} b ₃ , and c _{0 to} c ₃ .

  Here, the operation prediction method in a present Example is demonstrated in detail using drawing. FIG. 8 is a flowchart showing the operation prediction method of the present embodiment. In addition, the same code | symbol is attached | subjected about the process similar to Example 1 demonstrated previously, and the description is abbreviate | omitted.

For example, in the driver model construction unit 1, operation data x ₁ , x ₂ ,... Sent from a predetermined sensor as needed (sensor values: various information recognized by the driver for operation, such as distance, speed, acceleration, etc.) Is received (step S11), the processing of steps S1 to S4 is executed each time.

Then, the model switching point prediction unit 32 predicts the switching point (switching time, etc.) of the sub-driver model based on, for example, the parameters d ₁ and d ₂ that are the outputs of the driver model construction unit 1, and the prediction result is obtained. Output (step S12). FIG. 9 is a diagram illustrating a specific example of the switching time of the sub-driver model. At the same time, in the operation prediction unit 31, for example, based on the parameters a _{0 to} a ₃ , b _{0 to} b ₃ , c _{0 to} c ₃ , d ₁ , and d ₂ that are outputs of the driver model construction unit 1, A behavior is predicted (step S12).

  As described above, in this embodiment, the driver model construction method in the first embodiment is used to predict the switching timing of the sub-driver model and the driver's action (operation) while updating the driver model in time series. It was decided. As a result, the driver's operation can be predicted more accurately. For example, it is possible to control the driver according to the situation.

  FIG. 10 is a diagram showing a configuration of an operation prediction device using the driver model constructed in the first embodiment, and the operation prediction device 21b includes a driver model 41 constructed in the first embodiment. To do. Further, as described above, the driver model 41 is constructed by combining a plurality of sub-driver models, and further includes switching determination conditions (logical conditions) for individual sub-driver models. That is, the driver model 41 includes a logical condition unit 42 indicating switching determination conditions for individual sub-driver models, and a sub-driver model unit 43 indicating a plurality of sub-driver models, and is an operation sent from a sensor. Based on the data, the switching point (timing) of the sub-driver model and the driver operations (brake operation, accelerator operation, steering operation) are predicted / output.

  Here, the operation prediction method in a present Example is demonstrated in detail using drawing. FIG. 11 is a flowchart showing the operation prediction method in the present embodiment.

  For example, the operation predicting device 21b receives driving data (sensor values: various information recognized by the driver for operation, such as distance, speed, acceleration, etc.) sent from a predetermined sensor at any time (step S11). Each time, the driver model 41 predicts the switching point of the sub-driver model and the driver's behavior.

Specifically, in the driver model 41, operating data x _j as a criterion for performing status _{determination, k} (operation data x _{j, k} may be a single operating data, operating data If a composite value may be present), a switching point (switching time, etc.) of the sub-driver model is predicted based on the switching determination condition indicated in the logic condition unit 41, and the prediction result is output (step S21). A specific example of the switching time of the sub driver model is the same as that in FIG. 9 described above. Thereafter, the driver model 41 predicts the driver's action based on the switching information notified from the logical condition unit 41 and the driving data x ₁ , x ₂ ,... Sent from the sensor (step S22).

  As described above, in this embodiment, the driver model constructed in the first embodiment is used to predict the switching timing of the sub-driver model and the driver's action (operation). That is, since the driver model is updated in time series and the sub-driver model switching timing and the driver's behavior (operation) are not predicted, the calculation process can be greatly reduced as compared with the process of the second embodiment.

  As described above, the driver model construction method according to the present invention is useful as a method for constructing a driver model for predicting driver behavior, and in particular, a driver model capable of predicting an operation based on a situation determination of the driver. Suitable as a method for building.

It is a figure which shows the structure of the driver model construction part for implement | achieving the driver model construction method concerning this invention. It is a flowchart which shows the driver model construction method concerning this invention. It is a figure which shows the subdriver model accompanied by a situation judgment. It is a figure which shows the case where a condition judgment part is represented by the vertical axis | shaft (delta) and horizontal axis _{xj, k} . It is a figure which shows the case where the condition judgment part of FIG. 4 is replaced by the linear inequality. Is a diagram illustrating an example of the predicted parameters d _1, d ₂ I by executing the driver model building method according to the present invention. It is a figure which shows the structure of the operation prediction apparatus of Example 2 using the driver model construction method in Example 1. FIG. 10 is a flowchart illustrating an operation prediction method according to the second embodiment. It is a figure which shows the specific example of the switching time of a subdriver model. It is a figure which shows the structure of the operation prediction apparatus of Example 3 using the driver model constructed | assembled in Example 1. FIG. 10 is a flowchart illustrating an operation prediction method according to the third embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Driver model construction part 11 Polynomial structure determination part 12 Linear inequalities / equation generation part 13 Parameter identification part 21a, 21b Operation prediction apparatus 31 Operation prediction part 32 Model switching point prediction part 41 Driver model 42 Logical condition part 43 Sub driver model part

Claims (2)

A driver model is constructed by combining a plurality of motion models (each represented as a sub-driver model) representing the operation of the driver, and the driver model further includes a switching judgment condition for each sub-driver model. In the operation prediction method using the model construction method,
A polynomial structure determining step for determining a structure of each polynomial constituting the sub-driver model based on predetermined operation data obtained as information recognized by the driver for operation;
By introducing auxiliary theoretical variables (binary variables) and auxiliary continuous variables based on a system theory based on MLDS (Mixed Logical Dynamical System), the switching judgment condition and the sub-driver model are converted into linear inequalities and linear state equations. A linearization step expressed by
Based on the linear inequality and the linear equation of state, and further, the constraint on the auxiliary theoretical variable, the integral value of the difference between the output of the polynomial representing the operation of the driver and the operation actually observed is minimized. A parameter identification step of searching for a parameter for defining the switching judgment condition and a kinematic parameter constituting the polynomial by a mixed integer linear programming;
Based on the parameter for defining the switching judgment condition, the switching timing of the sub-driver model is predicted, and at the same time, the operation amount of the driver's operation is calculated based on the kinematic parameters constituting the polynomial of the sub-driver model. A prediction step to predict;
An operation prediction method characterized by comprising: A driver model is constructed by combining a plurality of motion models (each represented as a sub-driver model) representing the operation of the driver, and the driver model further includes a switching judgment condition for each sub-driver model. In the operation prediction device using the model building process,
Polynomial structure determining means for determining the structure of each polynomial constituting the sub-driver model based on predetermined operation data obtained as information recognized by the driver for operation, and based on MLDS (Mixed Logical Dynamical System) A linearization means for expressing the switching judgment condition and the sub-driver model by a linear inequality and a linear state equation by introducing an auxiliary theoretical variable (binary variable) and an auxiliary continuous variable according to the system theory;
Based on the linear inequality and the linear equation of state, and further, the constraint on the auxiliary theoretical variable, the integral value of the difference between the output of the polynomial representing the operation of the driver and the operation actually observed is minimized. A parameter identifying means for searching for a parameter for defining the switching judgment condition and a kinematic parameter constituting the polynomial by a mixed integer linear programming;
Based on the parameter for defining the switching judgment condition, the switching timing of the sub-driver model is predicted , and at the same time, the operation amount of the driver is determined based on the kinematic parameters constituting the polynomial of the sub-driver model. An operation prediction means for predicting;
An operation prediction apparatus comprising:

Images