Operation support device, operation support method, and program

Abstract

To provide a technique for calculating an appropriate manipulated variable of a target in consideration of uncertainty of a prediction value of a simulator.SOLUTION: An operation support device comprises: a simulator construction unit which constructs a simulator predicting a state variable value with an operation variable value and an environment variable value as input using achievement data of a target; a prediction error width estimation model construction unit which constructs a prediction error width estimation model of the simulator using the achievement data; a first scenario creation unit which creates an operation variable scenario using the operation variable value included in the achievement data; a second scenario creation unit which creates a state variable scenario using the simulator, the operation variable scenario, and a prescribed value of an environment variable; a third scenario creation unit which creates a prediction error width scenario using the prediction error width estimation model and the state variable scenario; and an optimization unit which calculates an operation variable optimum value from the prediction error width scenario when the prescribed value of the environment variable and a target value of the state variable are given.SELECTED DRAWING: Figure 6

Description

[Technical field]

[0001] The present disclosure relates to a driving assistance device, a driving assistance method, and a program. [Background technology]

[0002] In recent years, against the backdrop of a shortage of skilled operators in the industrial field, there is an increasing need for plant operation support technology that utilizes AI (Artificial Intelligence), machine learning technology, etc. A known conventional technology for plant operation support is a technology that uses a machine learning model that models a plant as a simulator and combines it with optimization technology to calculate an operation amount that will result in a target plant state (e.g., Patent Documents 1 and 2). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7081728 [Patent Document 2] Patent No. 7207596 Summary of the Invention [Problem to be solved by the invention]

[0004] However, while the simulator's predictions may contain errors (i.e., there is uncertainty in the simulator's predictions), the conventional techniques did not take into account the errors that may be contained in the simulator's predictions.

[0005] The present disclosure has been made in consideration of the above points, and provides a technique for calculating an optimal manipulated variable of a target, taking into account the uncertainty of a simulator's predicted value. [Means for solving the problem]

[0006] A driving assistance device according to one aspect of the present disclosure includes a simulator construction unit that uses a plurality of performance data collected from an object to construct a simulator that predicts values ​​of state variables of the object by inputting values ​​of manipulated variables of the object and values ​​of environmental variables of the object; a prediction error width estimation model construction unit that uses the plurality of performance data to construct a prediction error width estimation model that estimates a prediction error width of values ​​of state variables by the simulator; a first scenario creation unit that uses values ​​of manipulated variables included in the performance data to create manipulated variable scenarios that represent values ​​of manipulated variables that can be realized in the object; and a second scenario creation unit that uses the simulator, the manipulated variable scenario, and predetermined values ​​of given environmental variables to create a manipulated variable scenario that predicts values ​​of the manipulated variables included in the manipulated variable scenario. the predetermined value of the environmental variable and a target value of the state variable are given, the prediction error range scenario is used to calculate, as an optimal manipulated variable value, from among the values ​​of the manipulated variables included in the prediction error range scenario, a value which has a small prediction error range and which brings the value of the state variable of the target close to the target value, when the predetermined value of the environmental variable and a target value of the state variable are given, using the prediction error range scenario. Effect of the Invention

[0007] A technique is provided for calculating an optimal manipulated variable of an object taking into account the uncertainty of a simulator's predicted value. [Brief description of the drawings]

[0008] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of the driving assistance device according to the present embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a functional configuration of a driving assistance device according to the present embodiment. [Diagram 3] FIG. 4 is a diagram showing an example of a driving scenario creation process according to the embodiment; [Figure 4]FIG. 13 is a diagram for explaining an example of creating an manipulated variable scenario. [Diagram 5] FIG. 11 is a diagram for explaining an example of creating a state variable scenario. [Figure 6] FIG. 13 is a diagram for explaining an example of creating a forecast error range scenario. [Figure 7] 5 is a flowchart showing an example of an optimal operation amount calculation process according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] An embodiment of the present invention will be described below. In the following embodiment, a driving support device 10 capable of calculating an optimal operation amount for an object taking into consideration the uncertainty of a predicted value of a simulator will be described. Here, the driving support device 10 according to this embodiment has a driving scenario creation phase in which a driving scenario for the object is created, and a driving support phase in which the optimal operation amount for the object is calculated using this driving scenario and an optimization technique. A driving scenario is information that at least associates an operation amount (operation variable value), an environmental variable value, a state variable value predicted by a simulator using the operation variable value and the environmental variable value as input, and a prediction error width that represents the uncertainty of the state variable value.

[0010] The target is an object that is operated by an administrator (or may be called an operator), and examples of such objects include plants, facilities, equipment, and devices. Various sensors are installed on the target, and these sensors measure the state and environment of the target, as well as operations performed on the target. A variable that represents the state of the target is called a "state variable," a variable that represents the environment is called an "environmental variable," and a variable that represents an operation is called an "operational variable." In the following, state variables, environmental variables, and operational variables are collectively referred to simply as "variables."

[0011] Specific examples of state variables include temperature, pressure, and flow rate. Specific examples of environmental variables include outside air temperature and equipment number. Specific examples of operational variables include the opening and closing amounts of air valves and fuel valves. Note that when the target is a plant, these variables are also called process variables, etc., since they represent the plant process.

[0012] In the following, the total number of operational variables is I, and each operational variable is x i (1≦i≦I). Similarly, let the total number of state variables be J, and let each state variable be y j (1≦j≦J). Similarly, let the total number of environment variables be K, and let each environment variable be z k (1≦k≦K). Also, the operational variable x i Let x = (x1, ,x I ) Similarly, the state variable y j Let y = (y1, ,y J ) Similarly, the environment variable z k Let z=(z1, ,z) be a K-dimensional vector with (1≦k≦K) as the k-th element. K ) Note that although x, y, and z are all horizontal vectors, they may also be vertical vectors.

[0013] Also, the value of the operational variable at time t is x i (t), the state variable value at time t is y j (t), the value of the environmental variable at time t is z k (t). Also, x(t) = (x1(t), ,x I (t)), y(t) = (y1(t), ,y J (t)), z(t) = (z1(t), , z K The time t is also called a time step, and if the target control period is Δt, for example, the time is updated to t ← t + Δt every control period Δt.

[0014] <Hardware configuration example of the driving support device 10> An example of a hardware configuration of a driving support device 10 according to this embodiment is shown in Fig. 1. As shown in Fig. 1, the driving support device 10 according to this embodiment has an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a RAM (Random Access Memory) 105, a ROM (Read Only Memory) 106, an auxiliary storage device 107, and a processor 108. Each of these pieces of hardware is connected to each other via a bus 109 so as to be able to communicate with each other.

[0015] The input device 101 is, for example, a keyboard, a mouse, a touch panel, a physical button, etc. The display device 102 is, for example, a display, a display panel, etc. Note that the driving assistance device 10 does not necessarily have to include at least one of the input device 101 and the display device 102, for example.

[0016] The external I / F 103 is an interface with an external device such as a recording medium 103a. Examples of the recording medium 103a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), and a USB (Universal Serial Bus) memory card.

[0017] The communication I / F 104 is an interface for connecting the driving assistance device 10 to a communication network. The RAM 105 is a volatile semiconductor memory (storage device) that temporarily stores programs and data. The ROM 106 is a non-volatile semiconductor memory (storage device) that can store programs and data even when the power is turned off. The auxiliary storage device 107 is a non-volatile storage device such as a hard disk drive (HDD) or a solid state drive (SSD) and stores programs and data. The processor 108 is various types of arithmetic devices such as a central processing unit (CPU) or a graphics processing unit (GPU).

[0018] 1 is an example, and the driving support device 10 may have other hardware configurations. For example, the driving support device 10 may have multiple auxiliary storage devices 107 and multiple processors 108, or may have various hardware other than the hardware shown in the figure.

[0019] <Example of functional configuration of driving assistance device 10> FIG. 2 shows an example of a functional configuration of the driving support device 10 according to the present embodiment. As shown in FIG. 2, the driving support device 10 according to the present embodiment includes an input unit 201, a simulator construction unit 202, a prediction error width estimation model construction unit 203, an operation variable scenario creation unit 204, a state variable scenario creation unit 205, a prediction error width scenario creation unit 206, an optimization unit 207, and an output unit 208. Each of these units is realized, for example, by a process in which one or more programs installed in the driving support device 10 are executed by the processor 108 or the like. The driving support device 10 according to the present embodiment also includes a storage unit 209. The storage unit 209 is realized, for example, by the auxiliary storage device 107 or the like. However, the storage unit 209 may be realized, for example, by a storage device such as a database server connected to the driving support device 10 via a communication network.

[0020] In the operation scenario creation phase, the input unit 201 acquires past operation data (hereinafter also referred to as actual data) used for creating an operation scenario from the storage unit 209. The operation data is data consisting of operation variable values, state variable values, and environmental variable values ​​at each time t. That is, operation data at a certain time t is expressed as (x(t), y(t), z(t)), for example. Hereinafter, in the operation scenario creation phase, an actual data set D={(x(t), y(t), z(t))|t∈T} is acquired from the storage unit 209. Hereinafter, T is a time set of actual data used for creating an operation scenario. However, each actual data included in the actual data set D is assumed to be operation data during normal operation of the target, operation data when the target is being operated appropriately, operation data when exemplary operation is being performed, etc. Hereinafter, each x(t) will be referred to as an operation variable actual value, each y(t) will be referred to as a state variable actual value, and each z(t) will be referred to as an environmental variable actual value.

[0021] In addition, in the driving assistance phase, the input unit 201 acquires the environmental variable values ​​and the target values ​​of the state variables at the current time t=t0 from the storage unit 209. That is, the target values ​​of the state variables at time t are expressed as y′(t)=(y1′(t), . . . , y J If z(t0) is the current time t=t0, the input unit 201 acquires the environmental variable value z(t0) and the target value y'(t0) of the state variable at the current time t=t0 from the storage unit 209.

[0022] The simulator construction unit 202 constructs a target simulator using the result data set D acquired by the input unit 201 in the operation scenario creation phase. Here, the simulator is a model expressed by a function that receives manipulated variable values ​​and environmental variable values ​​as inputs and outputs state variable values ​​(i.e., a model that predicts state variable values ​​from manipulated variable values ​​and environmental variable values). Specifically, a model expressed as y=f(x,z) is the simulator. For f, for example, a physical model, a statistical model, a machine learning model, etc. can be used. For example, when f is a statistical model or a machine learning model, regression analysis, SVR (Support Vector Regression), neural network, etc. can be used. Note that f is expressed as f=(f1,...,f J ) (where y j =f j (x,z)). In the following, as an example, f is expressed as f=(f1, ,f J ) is the model represented as

[0023] The prediction error range estimation model construction unit 203 constructs a prediction error range estimation model using the actual data set D acquired by the input unit 201 in the operation scenario creation phase. Here, the prediction error range estimation model is a model expressed by a function that inputs manipulated variable values ​​and environmental variable values ​​and outputs a prediction error range of a value predicted by simulator f (i.e., a model that estimates a prediction error range from manipulated variable values ​​and environmental variable values). The prediction error is the difference between the actual state variable value and the predicted value by simulator f, that is, y j (err) (t):=y j (t)-f j This refers to (x(t),z(t)).

[0024] State variable y j The prediction error for y j (err) As, y (err) =(y1 (err) ,···,y J (err) ) Also, y (err)= g(x,z), where g = (g1, ,g J ) and each g j is a model realized by a decision tree. In this case, the prediction error range estimation model construction unit 203 uses the performance data set D to construct a prediction error estimation model g=(g1, . . . , g J ) and then each decision tree g j The prediction error y j (err) The average μ j (i) , standard deviation σ j (i) This calculates the decision tree g j Let T be the set of terminal node numbers of j The prediction error range estimation model (g,{{(μ j (i) ,σ j (i) )|i∈T j}|j=1,···,J}) is obtained.

[0025] Each prediction error y j (err) Assuming that the distribution of is normal, when a certain manipulated variable value x(t) and an environmental variable value z(t) are given, the state variable y j The upper limit of the forecast error band for y j (err_upper) and lower limit y j (err_lower) can be calculated as follows, respectively.

[0026] y j (err_upper) =f j (x(t),z(t))+μ j (i') +kσ j (i') (1) y j (err_lower) =f j (x(t),z(t))+μ j (i') -kσ j (i') (2) Here, μ j (i')is a decision tree g j When x(t) and z(t) are input to the terminal node i' to which x(t) and z(t) belong, the prediction error y j (err) The mean of σ j (i') is a decision tree g j When x(t) and z(t) are input to the terminal node i' to which x(t) and z(t) belong, the prediction error y j (err) The standard deviation of the prediction error range dy j :=y j (err_upper) -y j (err_lower) is obtained.

[0027] Note that in the above, each prediction error y j (err) We have assumed that the distribution of is normal, but other distributions may be assumed. In this case, the prediction error y j (err) It is only necessary to calculate appropriate statistics according to the distribution that the distribution of follows, and then calculate the upper and lower limits of the prediction error band from the statistics.

[0028] In the operation scenario creation phase, the manipulated variable scenario creation unit 204 creates a manipulated variable scenario from the manipulated variable actual values ​​{x(t)|t∈T} included in the actual data set D acquired by the input unit 201. The manipulated variable scenario refers to S (where S≧N) manipulated variable values ​​having the same distribution as N manipulated variable actual values ​​{x(t)|t∈T}, where N=|T|. The manipulated variable scenario may be created by using, for example, a method similar to the method described in Patent Document 2.

[0029] In the operation scenario creation phase, the state variable scenario creation unit 205 creates a state variable scenario using the manipulated variable scenario created by the manipulated variable scenario creation unit 204, the simulator f constructed by the simulator construction unit 202, and any predetermined environmental variable value. The state variable scenario is information in which the predicted value (state variable value) of the simulator f when the manipulated variable value is input to the manipulated variable scenario is associated with the environmental variable value. As a method for creating the state variable scenario, for example, a method similar to the method described in Patent Document 2 may be used.

[0030] In the operation scenario creation phase, the prediction error range scenario creation unit 206 uses the state variable scenario created by the state variable scenario creation unit 205 and the prediction error range estimation model (g,{{(μ j (i) ,σ j (i) )|i∈T j}|j=1, ,J}) is used to create a prediction error range scenario as an operation scenario. A prediction error range scenario is information that associates a prediction error range with a state variable scenario. In this way, a prediction error range scenario is created as an operation scenario.

[0031] In the driving assistance phase, the optimization unit 207 calculates optimal manipulated variables for bringing the target state variable values ​​closer to the target values, using the driving scenario created by the prediction error range scenario creation unit 206 and the environmental variable values ​​and the target values ​​of the state variables at the current time t=t0 acquired by the input unit 201. That is, the optimization unit 207 extracts an optimal driving scenario from the environmental variable values ​​and the target values ​​of the state variables at the current time t=t0, and then calculates the manipulated variable values ​​included in the optimal driving scenario as optimal manipulated variables.

[0032] The output unit 208 outputs the optimal operation amount calculated by the optimization unit 207 to a predetermined output destination. Such an output destination may be, for example, the display device 102 or a terminal connected via a communication network (particularly, a terminal used by an operator).

[0033] The storage unit 209 stores at least an actual data set D and environmental variable values ​​(or information indicating the range of values ​​that the environmental variables can take, conditions, etc.) used to create the state variable scenario in the operation scenario creation phase. Each actual data included in the actual data set D is collected from a target via a communication network, for example, at each time t.

[0034] The storage unit 209 may further store data (hereinafter also referred to as auxiliary data) including information such as the target operation date / stop date, normal range / abnormal range of state variable values, etc. These auxiliary data may be input by a user via the input device 101 or the like, or may be received from a user terminal or the like via a communication network.

[0035] Furthermore, the storage unit 209 stores at least the environmental variable value z(t0) at the current time t=t0 collected from the target during the driving assistance phase, and a set of target values ​​{y'(t)} of the state variables of the target at each time t. The environmental variable value z(t0) at the current time t=t0 is acquired from the target via a communication network, for example. The state variable value y(t0) at the current time t=t0 may be further acquired and stored in the storage unit 209. Meanwhile, the set of target values ​​of the state variables at each time t is created or determined in advance and stored in the storage unit 209.

[0036] <Driving scenario creation process> The driving scenario creation process executed in the driving scenario creation phase will be described below with reference to Fig. 3. Note that the driving scenario creation process is generally executed when the target is offline, but is not limited to this. For example, after being executed once offline, the process may be periodically executed in the background when online.

[0037] First, the input unit 201 acquires the result data set D used for generating an operation scenario from the storage unit 209 (step S101).

[0038] Next, the simulator construction unit 202 constructs a target simulator f by using the result data set D acquired in the above step S101 (step S102). That is, the simulator construction unit 202 constructs a simulator f such that f(x(t),z(t)) simulates (reproduces) y(t) for each time t∈T.

[0039] When constructing the simulator f in step S102, for example, auxiliary data may also be used to construct the simulator f. This is expected to result in a more accurate simulator.

[0040] Next, the prediction error width estimation model construction unit 203 constructs a prediction error width estimation model using the performance data set D acquired in the above step S101 (step S103). That is, the prediction error width estimation model construction unit 203 constructs a prediction error width estimation model g=(g1, . . . , g J ) and then each decision tree g j The prediction error y j (err) The average μ j (i) , standard deviation σ j (i) This calculates the prediction error range estimation model (g,{{(μ j (i) ,σ j (i) )|i∈T j}|j=1, ,J}) is obtained. Note that for each decision tree g j can be constructed using existing machine learning methods.

[0041] Next, the manipulated variable scenario creation unit 204 creates a manipulated variable scenario using the manipulated variable actual values ​​{x(t)|t∈T} included in the actual data set D acquired in the above step S101, in a manner similar to the method described in Patent Document 2 (step S104).

[0042] In the following, as an example, I=2, J=1, and K=2 are mainly assumed. At this time, the manipulated variable scenario creation unit 204 creates S (where S≧N) manipulated variable scenarios having the same distribution as the distribution of N=|T| manipulated variable actual values. As a result, for example, as shown in FIG. 4, S manipulated variable scenarios are created from N manipulated variable actual values. However, the distribution of the S manipulated variable scenarios does not need to be completely identical to the distribution of the N manipulated variable actual values, and a certain degree of identity or similarity is sufficient. For example, it is sufficient that the similarity between the distribution of the S manipulated variable scenarios and the distribution of the N manipulated variable actual values ​​is equal to or less than a predetermined threshold value using an index value that can evaluate the similarity between the distributions (for example, a known index value such as KL divergence).

[0043] Hereafter, the sth (where 1≦s≦S)th instrumental variable scenario is called x (s) In the example shown in Figure 4, (s) =(x1 (s) ,x2 (s) )(s = 1, ,S).

[0044] Examples of methods for creating manipulated variable scenarios include the following manipulated variable scenario creation method 1 to creation method 3. However, these are all merely examples, and the method for creating manipulated variable scenarios is not limited to these, and any method can be used as long as it can create S manipulated variable scenarios with the same or similar distribution from N manipulated variable actual values. In the following, N manipulated variable actual values ​​will be written as {x(n)|n=1, ,N}.

[0045] How to create an instrumental variable scenario: Part 1 Let α [%] be a predetermined value. At this time, x(n) is added with ±α [%] (i.e., x(n)+(x(n)×α) and x(n)-(x(n)×α)). Then, for n=1,...,N, x(n), x(n)+(x(n)×α), and x(n)-(x(n)×α)) are set as manipulated variable scenarios, respectively. In this creation method, for example, two manipulated variable values ​​are obtained from one manipulated variable actual value x(n), so S=N+2N=3N. This makes it possible to obtain manipulated variable scenarios that do not deviate from the distribution region of N manipulated variable actual values ​​x(n) (i.e., manipulated variable values ​​that can actually operate the target, taking into account the values ​​that each manipulated variable can take and the combinations that can be implemented between the manipulated variables, etc.).

[0046] How to create an instrumental variable scenario #2 Multidimensional normal random numbers are used, which can take into account the correlation between the actual values ​​of the manipulated variables. Multidimensional normal random numbers are random numbers that follow a multidimensional normal distribution. Let Σ be the variance-covariance matrix of the actual values ​​of the manipulated variables {x(n)|n=1, ,N}, μ be the mean vector of the actual values ​​of the manipulated variables {x(n)|n=1, ,N}, C be a lower triangular matrix, and R be a standard normal random number (random number that follows a standard normal distribution). In this case, the manipulated variable scenario x is calculated as follows: (s) Generate.

[0047] x (s) =C R + μ Σ=C C τ Here, τ represents the transpose of a matrix. Note that Σ=C C τ If Σ is a positive definite value, C can be obtained by Cholesky decomposition.

[0048] In this method, for any S number of instrumental variable scenarios x (s) (s=1, ,S). As a result, as with the above creation method 1, it is possible to obtain an instrumental variable scenario that does not deviate from the distribution region of the N instrumental variable actual values ​​x(n).

[0049] How to create an instrumental variable scenario: Part 3 We use a generative model that can generate data with a distribution close to the distribution of N actual values ​​of manipulated variables {x(n)|n=1, , N}. In other words, this generative model generates data for any S scenarios of manipulated variables x (s) (s=1, ,S). Such a generative model can be constructed, for example, by a method such as GAN (Generative Adversarial Network). This makes it possible to obtain an manipulated variable scenario that does not deviate from the distribution region of the N manipulated variable actual values ​​x(n), just like in the above creation method 2.

[0050] Returning to the explanation of Fig. 3, following step S104, the state variable scenario creation unit 205 creates the manipulated variable scenario x (s) Using (s=1, . . . , S) and a predetermined arbitrary environmental variable value z=z0, a state variable scenario is created by a method similar to that described in Patent Document 2 (step S105). That is, the state variable scenario creation unit 205 creates the manipulated variable scenario x (s) The state variable value y is calculated by the simulator f using the environmental variable value z=z0 as input. (s) =f(x (s) , z0) and then use the instrumental variable scenario x (s) and the environment variable value z=z0 and the state variable value y (s) For example, as shown in Figure 5, for s=1, ,S, the manipulated variable scenario x (s) and the environment variable value z=z0 and the state variable value y (s) =f(x (s) , z0) and the associated state variable scenario (x (s) ,z0,y (s) ) is obtained.

[0051] Then, the forecast error range scenario creation unit 206 calculates the state variable scenario (x (s) ,z0,y (s) ) (s=1, ,S) and the prediction error band estimation model (g,{{(μ j(i) ,σ j (i) )|i∈T j}|j=1, . . . , J}) to generate a prediction error range scenario as an operation scenario (step S106). That is, the prediction error range scenario generating unit 206 generates a state variable scenario (x (s) ,z0,y (s) )(s=1, ,S) j (err_upper)(s) and the forecast error range lower limit y j (err_lower)(s) After calculating these forecast error limits, j (err_upper)(s) and the forecast error range lower limit y j (err_lower)(s) From the prediction error range dy j (s) =y j (err_upper)(s) -y j (err_lower)(s) After calculating the forecast error band dy (s) =(dy1 (s) ,···,dy J (s) ) into the state variable scenario (x (s) ,z0,y (s) ) to create a forecast error range scenario. For example, as shown in Figure 6, for s=1, ,S, we create a forecast error range scenario by associating the (s) and the environment variable value z=z0 and the state variable value y (s) and the prediction error band dy (s) The forecast error range scenario (x (s) ,z0,y (s) ,dy (s) ) is obtained as the driving scenario.

[0052] From the above, the driving scenario (x (s) ,z0,y (s) ,dy (s) )(s=1, ,S). In the following, we consider these driving scenarios (x (s) ,z0,y (s) ,dy (s) )(s=1, ,S) is also called the set of driving scenarios, and {(x(s) ,z0,y (s) ,dy (s) )|s=1, ,S}. The set of driving scenarios {(x (s) ,z0,y (s) ,dy (s) )|s=1, . . . , S} is stored in, for example, the storage unit 209.

[0053] In addition, when the target can take various environmental variable values, various environmental variable values ​​z are given in the above steps S105 to S106. For example, when the target can take M environmental variable values ​​z0, z1, . . . , z M For each environment variable z∈{z0,z1,...,z M}, the above steps S105 to S106 are executed. As a result, for example, a set of M driving scenarios {{(x (s) ,z m ,y (s) ,dy (s) ) |s=1, . . . , S}|m=1, . . . , M} are obtained, and a set of M driving scenarios is stored in the storage unit 209.

[0054] <Optimal operation amount calculation process> The optimal operation amount calculation process executed in the driving assistance phase will be described below with reference to FIG. 7. The optimal operation amount calculation process is repeatedly executed, for example, every control period Δt. In the following, it is assumed that the environmental variable value z(t0) at the current time t=t0 is acquired from the target and stored in the storage unit 209. In the following, a set of M driving scenarios {{(x (s) ,z m ,y (s) ,dy (s) )|s=1, . . . , S}|m=1, . . . , M} are stored in the storage unit 209.

[0055] First, the input unit 201 acquires the environmental variable value z(t0) at the current time t=t0 and the target value y′(t0) of the state variable from the storage unit 209 (step S201).

[0056] Next, the optimization unit 207 calculates z(t0) by using the environmental variable value z(t0) and the target value y'(t0) obtained in step S201. m =z(t0) = {(x (s) ,z m ,y (s) )|s=1,···,S m}, the optimal driving scenario is extracted (step S202). m =z(t0) = {(x (s) ,z m ,y (s) )|s=1,···,S m} in |y1 (s) -y1'(t0)|+···+|y J (s) -y J '(t0)|+|dy1 (s) |+···+|dy J (s) The driving scenario (x (s) ,z m ,y (s) This means that the driving scenario that minimizes the sum of the difference with the target value y'(t0) and the prediction error range is the optimal driving scenario. (s) =(y1 (s) ,···,y J (s) ), dy (s) =(dy1 (s) ,···,dy J (s) ) The optimal driving scenario extracted in this step is defined as (x (s ' ) ,z m ,y (s ' ) )

[0057] In the above, the absolute value of the difference from the target value y'(t0) and the absolute value of the prediction error band are used, but the square of the difference may be used instead of the absolute value.

[0058] Next, the optimization unit 207 optimizes the optimal driving scenario (x (s ') ,z m ,y (s ' ) ) the instrumental variable value x (s ' ) is set as the optimal manipulated variable (step S203).

[0059] Then, the output unit 208 outputs the optimal operation amount x obtained in step S203 above. (s ' ) to a predetermined output destination (step S204). This allows an operator to operate a target such as a plant using this optimal manipulated variable (or a manipulated variable determined with reference to this optimal manipulated variable).

[0060] <Summary> As described above, the driving support device 10 according to this embodiment creates a driving scenario that takes into account the prediction error range of the simulator f (i.e., the uncertainty of the simulator f) by using the actual driving data (performance data) of the target, and can support optimal driving of the target by using the driving scenario. Therefore, for example, an operator of a target such as a plant can realize optimal driving of the target by optimal operation amount that also takes into account the uncertainty of the simulator f.

[0061] In the above embodiment, the operation amount calculated by the optimum operation amount calculation process is presented to the operator, but the present invention is not limited to this, and for example, the optimum operation amount may be output to a control device that controls the operation of the target. This makes it possible to automatically operate the target using the operation amount.

[0062] The present invention is not limited to the above-described embodiments specifically disclosed, and various modifications, changes, combinations with known technologies, etc. are possible without departing from the scope of the claims. [Explanation of symbols]

[0063] 10 Driving assistance devices 101 Input Device 102 Display device 103 External I / F 103a Recording media 104 Communication I / F 105 RAM 106 ROM 107 Auxiliary storage 108 processors 109 Bus 201 Input section 202 Simulator Construction Department 203 Forecast Error Range Estimation Model Construction Unit 204 Manipulated Variable Scenario Creation Section 205 State variable scenario creation section 206 Forecast Error Range Scenario Creation Department 207 Optimization Department 208 Output section 209 Memory section

Claims

1. a simulator construction unit that constructs a simulator that predicts values ​​of state variables of the target by inputting values ​​of manipulated variables of the target and values ​​of environmental variables of the target using a plurality of performance data collected from the target; a prediction error range estimation model constructing unit that constructs a prediction error range estimation model for estimating a prediction error range of a value of a state variable by the simulator, using the plurality of pieces of performance data; a first scenario creation unit that creates a manipulated variable scenario representing values ​​of manipulated variables that can be realized in the target, using values ​​of manipulated variables included in the performance data; a second scenario creation unit that creates a state variable scenario by using the simulator, the manipulated variable scenario, and a given predetermined value of an environmental variable, in which values ​​of state variables when the values ​​of the manipulated variables and the predetermined value of the environmental variable included in the manipulated variable scenario are input to the simulator are associated with the manipulated variable scenario; a third scenario creation unit that uses the prediction error range estimation model and the state variable scenario to create a prediction error range scenario in which a prediction error range of a value of a state variable included in the state variable scenario corresponds to the state variable scenario; an optimization unit that, when a predetermined value of the environmental variable and a target value of the state variable are given, calculates, using the prediction error range scenario, a value of the manipulated variable included in the prediction error range scenario, which has a small prediction error range and causes the value of the target state variable to approach the target value, as an optimal manipulated variable value; A driving assistance device having the above configuration.

2. The driving support device according to claim 1 , wherein the prediction error width estimation model is a model including a decision tree that estimates a prediction error of a value of a state variable by the simulator.

3. The prediction error range estimation model is 3. The driving support device according to claim 2, wherein the model estimates the prediction error range using values ​​of state variables calculated by the simulator and statistics of terminal nodes of the decision tree to which values ​​of operational variables and environmental variables input to the simulator belong.

4. The driving assistance device according to claim 3 , wherein the statistics are a mean and a standard deviation of prediction errors belonging to the terminal node.

5. The optimization unit is 5. The driving assistance device according to claim 1, wherein the optimum manipulated variable value is a value among values ​​of the manipulated variable included in the prediction error range scenario, which minimizes a sum of an absolute value of the prediction error range and an absolute value of a difference between a value of the target state variable and the target value.

6. A simulator construction step for constructing a simulator that predicts values ​​of state variables of an object by inputting values ​​of manipulated variables of the object and values ​​of environmental variables of the object using a plurality of performance data collected from the object; a prediction error range estimation model construction step of constructing a prediction error range estimation model for estimating a prediction error range of a value of a state variable by the simulator, using the plurality of pieces of performance data; a first scenario creation step of creating a manipulated variable scenario representing values ​​of manipulated variables that can be realized by the target using values ​​of manipulated variables included in the performance data; a second scenario creation step of creating a state variable scenario by using the simulator, the manipulated variable scenario, and a given predetermined value of an environmental variable, in which values ​​of state variables when the values ​​of the manipulated variables and the predetermined value of the environmental variable included in the manipulated variable scenario are input to the simulator are associated with the manipulated variable scenario; a third scenario creation step of creating a prediction error range scenario by using the prediction error range estimation model and the state variable scenario, in which a prediction error range of a value of a state variable included in the state variable scenario is associated with the state variable scenario; an optimization procedure for calculating, when a predetermined value of the environmental variable and a target value of the state variable are given, as an optimal manipulated variable value, from among values ​​of the manipulated variables included in the forecast error range scenario, a value that has a small forecast error range and that brings the value of the target state variable close to the target value, using the forecast error range scenario; A driving assistance method implemented by a computer.

7. A simulator construction step for constructing a simulator that predicts values ​​of state variables of an object by inputting values ​​of manipulated variables of the object and values ​​of environmental variables of the object using a plurality of performance data collected from the object; a prediction error range estimation model construction step of constructing a prediction error range estimation model for estimating a prediction error range of a value of a state variable by the simulator, using the plurality of pieces of performance data; a first scenario creation step of creating a manipulated variable scenario representing values ​​of manipulated variables that can be realized by the target using values ​​of manipulated variables included in the performance data; a second scenario creation step of creating a state variable scenario by using the simulator, the manipulated variable scenario, and a given predetermined value of an environmental variable, in which values ​​of state variables when the values ​​of the manipulated variables and the predetermined value of the environmental variable included in the manipulated variable scenario are input to the simulator are associated with the manipulated variable scenario; a third scenario creation step of creating a prediction error range scenario by using the prediction error range estimation model and the state variable scenario, in which a prediction error range of a value of a state variable included in the state variable scenario is associated with the state variable scenario; an optimization procedure for calculating, when a predetermined value of the environmental variable and a target value of the state variable are given, as an optimal manipulated variable value, from among values ​​of the manipulated variables included in the forecast error range scenario, a value that has a small forecast error range and that brings the value of the target state variable close to the target value, using the forecast error range scenario; A program that causes a computer to execute the following.

Images