Operation support device, operation support method, and program

Abstract

To calculate an amount of operation by which a prescribed value of an evaluation item reaches a target value in consideration of the operation records in the past.SOLUTION: An operation support device comprises: a prediction model creation unit for creating a prediction model for predicting a value of an evaluation item based on a set of actual data of a target, using an amount of operation as input; an initial solution creation unit for creating a solution at n=0; an optimization unit for calculating, as an nth solution, an amount of operation for minimizing an objective function including: an error between a value predicted by the prediction model and a target value for the evaluation item, and a deviation degree representing a deviation between the (n-1)th solution and the distribution of the set of actual data; a deviation degree calculation unit for calculating a deviation degree representing a deviation between the nth solution and the distribution of the set of actual data; an objective function value calculation unit for calculating a value of the objective function including: an error between a predicted value obtained when the amount of operation indicated by the nth solution is input into the prediction model and the target value, and a deviation degree representing a deviation between the nth solution and the distribution of the set of actual data; and an output unit for outputting the amount of operation represented by the solution having the smallest value of the objective function among the solutions from n=0 to n=N.SELECTED DRAWING: Figure 3

Description

[Technical field]

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

[0002] In various plants such as industrial plants, there is a demand to know the manipulated variable that makes the value of a certain evaluation item reach a target value. For example, in a petrochemical plant, there is a demand to know the optimal manipulated variable that makes the measured value of an evaluation item for evaluating the quality of a certain product approach the target value. For this reason, plant operation support technologies have been proposed that calculate the manipulated variable that reaches the target value by combining prediction technology and optimization technology (for example, Patent Documents 1 and 2, etc.). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2006-15180 A [Patent Document 2] International Publication No. 2019 / 208773 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional plant operation support technologies may not be able to calculate appropriate manipulated variables. For example, depending on the search range of the operation to be optimized, the manipulated variables calculated by the optimization technology may deviate from the operation range during normal operation of the plant. On the other hand, for example, if the search range is the range of operations that exist as past operating results, the manipulated variables calculated by the optimization technology may not reach the target value of a predetermined evaluation item.

[0005] The present disclosure has been made in consideration of the above-mentioned points, and has an object to calculate an operation amount at which a predetermined evaluation item value reaches a target value, taking into account past operating records. [Means for solving the problem]

[0006] A driving assistance device according to an aspect of the present disclosure includes a prediction model creation unit that creates a prediction model predicting a value of an evaluation item related to an object based on a set of performance data representing past driving performance of the object, using at least an operation amount for the object as an input; an initial solution creation unit that creates an initial solution representing the n=0th solution; an optimization unit that calculates, as an nth solution, an operation amount that minimizes an objective function including an error between a predicted value predicted by the prediction model and a target value for the evaluation item, and a deviation representing a degree of deviation between the n-1th solution and a distribution of the set of performance data; The prediction model includes a deviation calculation unit that calculates a deviation indicating the degree of deviation from a distribution of a set of actual data, an objective function value calculation unit that calculates a value of an objective function including an error between a predicted value predicted when an operation amount represented by the nth solution is input to the prediction model and a target value for the evaluation item, and a deviation indicating the degree of deviation between the nth solution and the distribution of the set of actual data, and an output unit that outputs at least the operation amount represented by the solution having the smallest value calculated by the objective function value calculation unit among the n=0th solution to the n=Nth solution (where N is a predetermined integer of 1 or more). Effect of the Invention

[0007] Taking into consideration past operating records, it is possible to calculate the amount of operation at which a predetermined evaluation item value reaches a target value. [Brief description of the drawings]

[0008] [Figure 1] FIG. 10 is a diagram illustrating an example in which a solution representing an optimal manipulated variable significantly deviates from the distribution of past performance values. [Diagram 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of the driving assistance device according to the present embodiment. [Diagram 3] FIG. 2 is a diagram illustrating an example of a functional configuration of a driving assistance device according to the present embodiment. [Figure 4] 5 is a flowchart illustrating an example of a driving assistance process according to the present embodiment. [Diagram 5] FIG. 13 is a diagram illustrating an example where Δ2≧Δ1. [Figure 6] FIG. 13 is a diagram showing a schematic example of a case where Δ2<Δ1. [Figure 7] FIG. 13 is a diagram illustrating an example of a case where the deviation becomes 0 because Δ2 is very large. [Figure 8] FIG. 13 is a diagram showing a schematic example of the range of Δ2 after improvement. [Figure 9] FIG. 1 is a diagram illustrating an example of dimensionality reduction. [Figure 10] FIG. 11 is a diagram illustrating a modified example of the functional configuration of the driving assistance device according to the embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] An embodiment of the present invention will be described below. In the following embodiment, an operation support device 10 will be described that can calculate an operation amount at which a predetermined evaluation item value related to a plant reaches a target value, taking into account the past operation record of the plant, for various plants such as industrial plants. By using the operation support device 10 according to this embodiment, a plant operator can know an operation amount at which a predetermined evaluation item value related to the plant reaches a target value, thereby realizing operation support for the plant. However, the evaluation item value reaching the target value does not mean that the evaluation item value strictly matches the target value, but means that the evaluation item value is sufficiently close to the target value (in other words, the difference between the evaluation item value and the target value is within a certain allowable error).

[0010] The evaluation items are items for evaluating the operation of a plant. Examples of evaluation items include evaluation items for evaluating the quality of products produced or manufactured by the plant. For example, in a pharmaceutical plant, the concentration (content), appearance, hardness, disintegration, and dissolution amount of a certain ingredient in tablets manufactured by the plant may be included.

[0011] In addition, the plant is not limited to a specific plant, and any plant can be the target. Examples of plants include pharmaceutical plants, petrochemical plants, chemical plants, power plants, water treatment plants, food plants, steel plants, and waste treatment plants. However, the operation assistance device 10 according to this embodiment can also be used for facilities, equipment, devices, and the like that perform some kind of processing in response to an operation, in addition to plants.

[0012] Generally, a plant is equipped with measuring devices such as sensors, and these sensors measure the values ​​of evaluation items related to the plant. In addition, there are devices that control or manage the plant, and these devices control or manage the manipulated variables for the plant and the plant's environment. Hereinafter, variables that represent evaluation items related to the plant are called "evaluation target variables," variables that represent the environment are called "environmental variables," and variables that represent operations are called "operational variables."

[0013] In the case of a pharmaceutical plant, specific examples of variables to be evaluated include the concentration (content) of a certain component in tablets manufactured by the plant, appearance, hardness, disintegration property, amount of elution, etc. In the case of a pharmaceutical plant, specific examples of environmental variables include the equipment number of the plant, outside air temperature, etc. In the case of a pharmaceutical plant, specific examples of operational variables include the set temperature of the plant, the opening and closing amount and opening and closing time of a valve for feeding raw materials, etc.

[0014] In the following, the total number of the manipulated variables is I, and the manipulated variables are x = (x1, , x I Similarly, let the total number of variables to be evaluated be J, and let the variables to be evaluated be y = (y1, ,y J On the other hand, we assume that the environmental variables are fixed equipment numbers for each plant, and express them as z = z'. The manipulated variable value corresponds to the manipulated variable.

[0015] <Outline of the method for calculating the optimal operation amount> A set of past operation data {(x (a,m) ,z',y (a,m))|m=1,···,M} and the target value y'=(y1',···,y J ') is given, where M is the number of actual data.

[0016] In this case, a prediction model f = (f1, . . . , f J ), and then |Std(f1(x,z'))-Std(y1')|+···+|Std(f J (x,z'))-Std(y J Consider the case where Std(f ')| is the objective function. j (x,z') is f j This is the value after normalizing (x,z') to a value between 0 and 1. Similarly, Std(y j ') is y j ' is normalized to a value between 0 and 1.

[0017] The search range for the operation to be optimized is set to the operation amount included in the actual data (i.e., {x (a,m) When the objective function is minimized, the optimal manipulated variable may not reach the target value, and the optimal manipulated variable may not be the one that satisfies the objective function minimization requirement. This can occur especially when the total number of variables to be evaluated is 2 or more, i.e., J≧2.

[0018] On the other hand, for example, when a range including operation amounts other than the operation amounts included in the performance data is set as the search range, the operation amount calculated as the optimal solution of the optimization technology may deviate from the operation range during normal operation of the plant. Note that, for example, particle swarm optimization (PSO) is an example of an optimization technology that can obtain an optimal solution for the operation amount by setting a range including operation amounts other than the operation amounts included in the performance data as the search range.

[0019] As an example, Figure 1 shows an example where the manipulated variable calculated as the optimal solution of the optimization technology when I=2 deviates significantly from the distribution of past actual values ​​of the manipulated variable. In the example shown in Figure 1, the manipulated variable calculated as the optimal solution deviates significantly from the distribution of past actual values ​​of the manipulated variable. For this reason, for example, the manipulated variable calculated as the optimal solution cannot be set in the plant, or even if it can be set in the plant, it may result in operation that is far removed from normal operation.

[0020] Therefore, in the driving support device 10 according to the present embodiment, a deviation is introduced as a measure of the degree of deviation between an operation amount that is a candidate for an optimal operation amount and the distribution of actual values ​​of the past operation amounts, and the deviation is added as a penalty term to an objective function for calculating the operation amount by the optimization technique. This makes it possible to prevent an operation amount that deviates significantly from the distribution of actual values ​​of the past operation amounts from being calculated as an optimal solution, while setting a range including operation amounts other than the operation amount included in the actual data as the search range.

[0021] <Hardware configuration example of the driving support device 10> An example of the hardware configuration of the driving support device 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the hardware configuration of the driving support device 10 according to this embodiment.

[0022] 2, the driving assistance device 10 according to this embodiment includes an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a random access memory (RAM) 105, a read only memory (ROM) 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.

[0023] 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.

[0024] 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.

[0025] The communication I / F 104 is an interface for the driving assistance device 10 to communicate with other devices and other apparatuses. 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), a solid state drive (SSD), or a flash memory, 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).

[0026] The driving assistance device 10 according to this embodiment can realize the driving assistance process described below by having the hardware configuration shown in Fig. 2. However, the hardware configuration shown in Fig. 2 is only an example, and the hardware configuration of the driving assistance device 10 is not limited to this. For example, the driving assistance device 10 may have multiple auxiliary storage devices 107 and multiple processors 108, may not have some of the hardware shown in the figure, or may have various hardware other than the hardware shown in the figure.

[0027] <Example of functional configuration of driving assistance device 10> An example of the functional configuration of the driving support device 10 according to this embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the functional configuration of the driving support device 10 according to this embodiment.

[0028] As shown in FIG. 3, the driving support device 10 according to the present embodiment includes a performance data acquisition unit 201, a prediction model creation unit 202, an initial solution creation unit 203, an update control unit 204, an optimization unit 205, a deviation calculation unit 206, an objective function value calculation unit 207, and an optimal solution 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 performance data storage unit 209, a prediction model storage unit 210, and an optimal solution candidate storage unit 211. Each of these storage units is realized, for example, by a storage area of ​​the auxiliary storage device 107 or the like. However, for example, at least a part of the storage units may be realized by a storage area of ​​a storage device or the like that is communicably connected to the driving support device 10.

[0029] The performance data acquisition unit 201 acquires a set of performance data (hereinafter referred to as performance data set) stored in the performance data storage unit 209. Performance data is data that indicates past operating performance, and is composed of performance values ​​of operation variables, performance values ​​of environmental variables, and performance values ​​of evaluation target variables. Hereinafter, the performance data set is defined as D={(x (a,m) ,z',y (a,m) )|m=1, ,M}, where M is the number of pieces of performance data that make up the performance data set D.

[0030] The prediction model creation unit 202 creates a prediction model f that predicts the evaluation target variable value from the manipulated variables and the environmental variable values, using the performance data set D acquired by the performance data acquisition unit 201. That is, the prediction model creation unit 202 creates a prediction model f in which the manipulated variables and the environmental variables are explanatory variables and the evaluation target variable is a response variable. At this time, the prediction model creation unit 202 creates a prediction model f in which each evaluation target variable y jFor the evaluation target variable y j A predictive model that predicts the value of f j Alternatively, we can create the evaluation target variables y1, , y J A prediction model f may be created to predict the value of each evaluation target variable y j For the evaluation target variable y j A predictive model that predicts the value of f j are created, respectively, and f = (f1, ,f J ) will be expressed as

[0031] The prediction model creation unit 202 may create the prediction model f by using a method such as machine learning including deep learning, or may create a physical model based on physical and chemical characteristics of a process executed by the plant as the prediction model f, or may create the prediction model f by an ensemble of these. Regardless of the method used to create the prediction model f, the prediction model creation unit 202 calculates the operation amount x (a,m) And the variable value y to be evaluated from the environmental variable value z' (a,m) Create a prediction model f that can predict with high accuracy.

[0032] Furthermore, the prediction model creation unit 202 stores the created prediction model f in the prediction model storage unit 210.

[0033] The initial solution creation unit 203 creates an initial solution for the operation to be optimized. Hereinafter, the initial solution is denoted by x (0) Let us express this as: Also, this initial solution x (0) The solution after updating x (n) The initial solution generating unit 203 may, for example, generate a random value as the initial solution x (0) Alternatively, a certain actual value x included in the actual data set D may be created as (a,n) Let x be the initial solution. (0) Alternatively, each performance value x included in the performance data set D may be (a,n) The average value of the initial solution x (0) It is also possible to use the following.

[0034] The update control unit 204 updates the solution x (n) More specifically, the update control unit 204 initializes a variable n that indicates the number of times the solution has been updated, and updates the solution x in accordance with the value of the variable n. (n) It determines whether to update or not.

[0035] If n>0, the optimization unit 205 optimizes the previous solution x (n-1) And the solution x (n-1) The following solution x is chosen to minimize the objective function that includes the degree of deviation between and the distribution of the actual data set D as a penalty term. (n) At this time, the optimization unit 205 calculates, for example, the previous solution x (n-1) Using existing optimization techniques such as particle swarm optimization with the initial value x, the solution x is obtained within the optimization algorithm of the optimization technique. (n-1) By repeatedly updating from the initial value, the next solution x (n) For details of particle swarm optimization, see, for example, Reference 1.

[0036] The deviation calculation unit 206 calculates the solution x (n) The degree of discrepancy d between the distribution of the actual data set D and (n) Calculate.

[0037] The objective function value calculation unit 207 calculates the solution x (n) At this time, the objective function value calculation unit 207 calculates an objective function value for the solution x (n) Objective function E for (n) After creating (x,z), E (n) (x (n) ,z') to find the solution x (n) Objective function value E for (n) (x (n) ,z') is calculated. Here, the solution x (n) The objective function for E (n) (x,z)=E(x,z)+α d (n) Here, E(x,z) is the objective function used in existing optimization techniques, for example, E(x,z)=|Std(f1(x,z))-Std(y1')|+···+|Std(f J(x,z))-Std(y J ')| can be used. J ') is the target value for the variable to be evaluated and is given in advance. Also, Std(f j (x,z) is f j This is the value after normalizing (x,z) to a value between 0 and 1. Similarly, Std(y j ') is y j ' is normalized to a value between 0 and 1. Furthermore, α is the penalty term d (n) is a hyperparameter that controls the strength of the α, and a value of α≧1 is set. In particular, α may be 1.

[0038] The objective function E used in existing optimization techniques is E(x,z)=|Std(f1(x,z))-Std(y1')|+···+|Std(f J (x,z))-Std(y J Using |f1(x,z),...,f J (x,z)) and y' = (y1', ,y J Any objective function can be used as E as long as it represents the error between

[0039] In addition, the objective function value calculation unit 207 calculates the solution x (n) and its objective function value E (n) (x (n) , z') is stored in the optimum solution candidate storage unit 211 as an optimum solution candidate.

[0040] The optimal solution output unit 208 outputs the optimal solution candidate (x (n) ,E (n) (x (n) ,z')), the solution x with the smallest objective function value is (n') is output as a final optimal solution to a predetermined output destination. As a result, the manipulated variable represented by this final optimal solution is obtained as the optimal manipulated variable.

[0041] The predetermined output destination is not limited to a specific output destination, but may be, for example, a terminal used by a plant operator or the like. This enables the plant operator to know the optimal manipulated variable that allows the evaluation target variable to reach the target value, and for example, by referring to the optimal manipulated variable, the plant operator can set an appropriate manipulated variable for a control device or the like that controls the plant. However, using a terminal used by a plant operator or the like as the output destination is just one example, and other than this, the control device that controls the plant or a monitoring device that monitors the plant may be the output destination.

[0042] The performance data storage unit 209 stores the performance data set D. Note that the performance data set D is stored in the performance data storage unit 209 in advance.

[0043] The prediction model storage unit 210 stores the prediction model f created by the prediction model creation unit 202.

[0044] The optimal solution candidate storage unit 211 stores optimal solution candidates (x (n) ,E (n) (x (n) ,z') where N is the solution x (n) is a hyperparameter that represents the number of updates.

[0045] <Driving assistance processing> An example of the driving support processing according to this embodiment will be described with reference to Fig. 4. Fig. 4 is a flowchart showing an example of the driving support processing according to this embodiment.

[0046] First, the performance data acquisition unit 201 acquires the performance data set D stored in the performance data storage unit 209 (step S101).

[0047] Next, the prediction model creation unit 202 creates a prediction model f from the result data set D acquired in the above step S101, and stores the prediction model f in the prediction model storage unit 210 (step S102).

[0048] Next, the initial solution generating unit 203 generates the initial solution x (0) is created (step S103).

[0049] Next, the update control unit 204 initializes the value of a variable n, which indicates the number of times the solution has been updated, to 0 (that is, initializes n←0) (step S104).

[0050] Next, the update control unit 204 determines whether the value of the variable n is not 0 (that is, whether n≠0) (step S105).

[0051] If it is determined in step S105 above that the value of the variable n is not 0 (i.e., n>0), the optimization unit 205 calculates the solution x (n-1) And the solution x (n-1) The degree of discrepancy d between the distribution of the actual data set D and (n-1) The objective function E includes the penalty term (n) (x,z)=E(x,z)+α d (n-1) Using and, the following solution x (n) (Step S106). That is, the optimization unit 205 calculates, for example, the solution x (n-1) Using existing optimization techniques such as particle swarm optimization with the initial value x, the solution x is obtained within the optimization algorithm of the optimization technique. (n-1) By repeatedly updating from the initial value, the next solution x (n) Calculate.

[0052] If it is determined in step S105 that the value of the variable n is 0 (that is, n=0), or following step S106, the deviation calculation unit 206 calculates the solution x (n) The degree of discrepancy d between the distribution of the actual data set D and (n) (Step S107). Here, the deviation d (n) As for the solution x (n) Any scale can be used as long as it measures the degree of deviation between the distribution of the result data set D and the distribution of the result data set D. As an example, the deviation calculation unit 206 calculates the deviation d by any one of the following deviation calculation methods 1 to 3. (n)It is sufficient to calculate

[0053] ·Difference calculation method 1 A predetermined distance function that measures the distance between two manipulated variables is Δ(·,·). As the distance function Δ(·,·), for example, a distance function that measures distance such as Euclidean distance or Mahalanobis distance can be used.

[0054] The actual value x of the manipulated variable included in the actual data set D (a,m) In (m∈{1, ,M}), the solution x (n) Distance Δ(x (n) ,x (a,m) ) is the smallest actual value x (a,m') is called the nearest point, and r:=x (a,m') In addition, Δ1:=Δ(x (n) ,r).

[0055] Furthermore, the actual value x of the operation amount included in the actual data set D (a,m) Among (m∈{1, ,M}\{m'}), the distance Δ(r,x (a,m) ) is the smallest actual value x (a,m'') is called the comparison point, and s:=x (a,m'') Let Δ2:=Δ(r,s).

[0056] In this case, if Δ2 ≥ Δ1, then d (n) = 0, if Δ2<Δ1 then d (n) = (Δ1 / Δ2)-1, the deviation d (n) Here, the state when Δ2 ≧ Δ1 is shown in FIG. 5, and the state when Δ2 < Δ1 is shown in FIG. 6. As shown in FIG. 5, when Δ2 ≧ Δ1, the solution x (n) The distance Δ1 between the nearest point r and the comparison point s is less than or equal to the distance Δ2 between the nearest point r and the comparison point s. Therefore, if Δ2 ≥ Δ1, the solution x (n) is assumed to be consistent with the distribution of the actual data set D, and d (n) On the other hand, as shown in FIG. 6, if Δ2<Δ1, the solution x (n)The distance Δ1 between the nearest point r and the comparison point s is greater than the distance Δ2 between the nearest point r and the comparison point s. Therefore, if Δ2<Δ1, the solution x (n) is assumed to deviate from the distribution of the actual data set D, and d (n) =(Δ1 / Δ2)-1.

[0057] ·Difference calculation method 2 k is a preset integer equal to or greater than 1. In this case, Δ(x (n) ,x (a,1) ), ,Δ(x (n) ,x (a,M) ) in ascending order of value, and the average value of these k distances is set as Δ1. Δ2 may be the same as in deviation calculation method 1, or Δ(r, x (a,1) ), ,Δ(r,x (a,M) ) in ascending order of value, and the average value of these k distances may be set as Δ2.

[0058] In this case, if Δ2 ≥ Δ1, then d (n) = 0, if Δ2<Δ1 then d (n) = (Δ1 / Δ2)-1, the deviation d (n) In this deviation calculation method 2, k actual values ​​x including the nearest point r are calculated. (a,m) In order to set the average value of the distance to Δ1, it is possible to calculate the deviation taking into account the density of the performance data set D, particularly by setting the value of k to a large value.

[0059] ·Difference calculation method 3 β:=1-1 / ln(M+e). In this case, in the above deviation calculation method 1 or 2, if Δ2<Δ1, d (n) = β × ((Δ1 / Δ2) - 1). β approaches 0 as the number M of actual data included in the actual data set D decreases, while it approaches 1 as the value of M increases. Therefore, β is a function of the deviation d (n) It can be said that this represents the reliability of the performance data set D when calculating. Therefore, by multiplying this β by (Δ1 / Δ2)-1, the degree of deviation d (n)can be calculated.

[0060] ·Difference calculation method 4 In the above deviation calculation method 1, the solution x (n) The magnitude of Δ2 varies depending on which actual value is the closest point. Therefore, depending on the actual value that is the closest point, Δ2 can become very large, and as a result, the solution x (n) Even if x is far from the distribution of the performance data set D, the deviation may be 0. For example, in the example shown in Figure 7, (n) Although is far from the distribution of the actual data set D, Δ2 is very large, so Δ2 ≧ Δ1, and d (n) =0.

[0061] Therefore, in deviation calculation method 4, the solution x (n) The magnitude of Δ2 is the same regardless of which actual value is the closest to the actual value. (a,m) And its actual value x (a,m) and the closest other actual value x (a,m') The distance between (m) In this case, Δ2=(Δ (1) +···+Δ (M) ) / M. That is, the actual value x of the manipulated variable included in the actual data set D (a,m) And its actual value x (a,m) The other actual value x that is closest to (a,m') The average distance between x and x is Δ2. As a result, the solution x (n) It is possible to make Δ2 the same size regardless of which actual value is closest to x. (n) Whatever actual value is closest to, Δ1 can be compared with Δ2 of the same magnitude.

[0062] ·Difference calculation method 5 In deviation calculation method 5, the calculation method of Δ2 in deviation calculation method 2 is changed to the solution x (n) Regardless of which actual value is closest to x, the magnitude of Δ2 is the same. (a,m) ∈D and x (a,m') ∈D\{x (a,m)} is the distance Δ(m,m') As a result, Δ (m,m') (m'∈{1, ,M}\{m}) is used to find the k' distances in ascending order of value, and the average of these k' distances is calculated as Δ (m) In this case, Δ2=(Δ (1) +···+Δ (M) ) / M. This gives the solution x (n) It is possible to calculate the degree of deviation from the data density around the actual value and the data density around each actual value.

[0063] Returning to the description of FIG. 4, following step S107, the objective function value calculation unit 207 calculates the solution x (n) Objective function value E for (n) (x (n) , z') (step S108). That is, the objective function value calculation unit 207 calculates the solution x (n) Objective function E for (n) (x,z)=E(x,z)+α d (n) After creating the E (n) (x (n) ,z') to find the solution x (n) Objective function value E for (n) (x (n) ,z') is calculated.

[0064] Next, the objective function value calculation unit 207 calculates the solution x (n) and the objective function value E calculated in step S108 above. (n) (x (n) , z') is stored as an optimal solution candidate in the optimal solution candidate storage unit 211 (step S109).

[0065] Next, the update control unit 204 adds 1 to the value of the variable n (that is, n←n+1) (step S110).

[0066] Next, the update control unit 204 determines whether the value of the variable n is greater than N (that is, whether n>N) (step S111).

[0067] If it is not determined in step S111 that the value of the variable n is greater than N (that is, n≦N), the update control unit 204 returns to step S105. As a result, steps S105 to S110 are repeatedly executed until it is determined in step S111 that the value of the variable n is greater than N.

[0068] On the other hand, if it is determined in step S111 that the value of the variable n is greater than N (i.e., n>N), the optimal solution output unit 208 outputs the optimal solution candidate (x (n) ,E (n) (x (n) ,z')), the solution x with the smallest objective function value is (n') The optimal solution output unit 208 outputs the solution x (n') as well as one or more other solutions x (n'') In particular, the optimal solution output unit 208 may output L solutions in ascending order of objective function value, where L is a predetermined integer of 2 or more.

[0069] <Modification> When the correlation between the manipulated variables to be input to the prediction model is known in advance, dimensionality reduction may be applied to those manipulated variables. Dimensionality reduction makes it possible to reduce the dimensions while maintaining the information of the original dimensions. This makes it possible to map the manipulated variables onto a space (hereinafter also referred to as a compressed space) that well represents the relationships between the manipulated variables.

[0070] When the solution of the optimization calculation is updated for the operational variables after the dimensionality reduction, the solution moves only in the compressed space, so the solution can be optimized while keeping the original data distribution well represented. For example, the solution x (n) Solution x (n+1) As shown in Figure 9, when the solution x (n) Solution x (n+1) Go to.

[0071] In addition, since the number of variables to be optimized is reduced by the dimensionality reduction, it becomes possible to efficiently search for an optimal solution. The number of dimensions after the reduction may be determined, for example, by a correlation coefficient between the operational variables.

[0072] An example of a functional configuration of the driving support device 10 when the operation amount to be input to the prediction model is dimensionally compressed is shown in Fig. 10. As shown in Fig. 10, the driving support device 10 when the operation amount to be input to the prediction model is dimensionally compressed has a dimensionality compression unit 212 and a dimensionality restoration unit 213 in addition to the units described in Fig. 3. These units are 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.

[0073] The dimension reduction unit 212 applies a predetermined compression map to the actual values ​​of the operational variables included in the actual data set D acquired by the actual data acquisition unit 201, and reduces the dimensions of the actual values ​​of the operational variables. After that, in steps S102 to S112 of FIG. 4, a prediction model f is created using the actual values ​​of the operational variables after the dimension reduction, and an initial solution x (0) or solution x (n) etc. are also created and updated in the compressed space and the optimal solution is calculated in the compressed space.

[0074] The dimension restoration unit 213 applies the inverse mapping of the compression mapping to the solution (optimal solution) output by the optimal solution output unit 208 to restore the dimension. This allows an optimal solution in the original dimension to be obtained. The optimal solution after being restored to the original dimension is output by the optimal solution output unit 208 to a predetermined output destination.

[0075] For example, when there is a linear relationship between the manipulated variables, the compression map and its inverse map may be generated using, for example, PCA (Principal Component Analysis). On the other hand, when there is a nonlinear relationship between the manipulated variables, the compression map and its inverse map may be generated using, for example, AE (Autoencoder).

[0076] <Summary> As described above, the driving support device 10 according to the present embodiment introduces the degree of deviation as a measure of the degree of deviation between a solution that is a candidate for the optimal manipulated variable and the distribution of the actual values ​​of the past manipulated variables, and calculates a solution that is a candidate for the optimal manipulated variable by optimization technology using an objective function that includes the degree of deviation as a penalty term. This makes it possible to prevent a manipulated variable that is significantly deviated from the distribution of the actual values ​​of the past manipulated variables from being calculated as the optimal solution while setting a range that includes manipulated variables other than the manipulated variables included in the actual data as the search range. Therefore, the driving support device 10 according to the present embodiment makes it possible to calculate a manipulated variable that will reach a target value for a predetermined evaluation item within a range that does not deviate from the operating range during normal operation of the plant.

[0077] 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.

[0078] [References] Reference 1: J. Kennedy and R. Eberhart, "Particle Swarm Optimization", IEEE International Conference on Neural Network(ICNN), 1995. [Explanation of symbols]

[0079] 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 Performance Data Acquisition Department 202 Prediction Model Creation Department 203 Initial solution creation part 204 Update control section 205 Optimization Department 206 Deviation Calculation Unit 207 Objective function value calculation unit 208 Optimal solution output unit 209 Performance data storage unit 210 Prediction model storage unit 211 Optimal solution candidate memory unit 212 Dimensional Compression Unit 213 Dimensional Restoration Department

Claims

1. a prediction model creation unit that creates a prediction model that predicts a value of an evaluation item related to an object by inputting at least an operation amount for the object based on a set of performance data representing a past operation performance of the object; an initial solution generating unit that generates an initial solution representing the n=0th solution; an optimization unit that calculates, as an n-th solution, a manipulation amount that minimizes an objective function that includes an error between a predicted value predicted by the prediction model and a target value for the evaluation item, and a deviation degree that represents a degree of deviation between the (n-1)-th solution and a distribution of the set of performance data; a deviation calculation unit that calculates a deviation indicating a degree of deviation between an n-th solution and a distribution of the set of performance data; an objective function value calculation unit that calculates a value of an objective function including an error between a predicted value predicted when an operation amount represented by the n-th solution is input to at least the prediction model and a target value for the evaluation item, and a deviation representing a degree of deviation between the n-th solution and a distribution of the set of performance data; an output unit that outputs at least an operation amount represented by a solution having a smallest value calculated by the objective function value calculation unit among the n=0th solution to the n=Nth solution (where N is a predetermined integer of 1 or more); A driving assistance device having the above configuration.

2. 2. The driving assistance device according to claim 1, wherein, for each of n = 1, ..., N, the optimization unit calculates the nth solution, the deviation calculation unit calculates a deviation, and the objective function value calculation unit calculates an objective function value.

3. The optimization unit is 3. The driving assistance device according to claim 1, wherein a manipulation amount that minimizes the objective function including, as a penalty term, a deviation degree that represents a degree of deviation between the (n-1)th solution and a distribution of the set of performance data is calculated as the nth solution.

4. The deviation calculation unit a first distance representing a distance between an operation amount represented by the n-th solution and a first operation amount actual value representing an operation amount actual value that is the nearest to the operation amount represented by the n-th solution among the operation amount actual values ​​included in the set of performance data; 2. The driving assistance device according to claim 1, wherein the deviation degree is calculated to be 0 when the first distance is equal to or less than the second distance by using a second distance representing a distance between the first operation amount actual value and a second operation amount actual value representing an operation amount actual value that is closest to the first operation amount actual value among the operation amount actual values ​​included in the set of actual data, and the deviation degree is calculated to be a positive value based on a ratio of the first distance to the second distance when the first distance is greater than the second distance.

5. The deviation calculation unit a first distance representing an average of distances between the operation amount represented by the n-th solution and each of top k (where k is a predetermined integer of 1 or more) operation amount actual values ​​that are closest to the operation amount represented by the n-th solution among the operation amount actual values ​​included in the set of performance data; 2. The driving assistance device according to claim 1, further comprising: a first operation amount actual value representing an operation amount actual value that is nearest to the operation amount represented by the n-th solution among the operation amount actual values ​​included in the set of performance data; and a second operation amount actual value representing an operation amount actual value that is nearest to the first operation amount actual value among the operation amount actual values ​​included in the set of performance data; and a second distance representing a distance between the first distance and the second distance, the deviation degree is calculated to be 0 if the first distance is equal to or smaller than the second distance, and the deviation degree is calculated to be a positive value based on a ratio of the first distance to the second distance if the first distance is greater than the second distance.

6. The deviation calculation unit a first distance representing a distance between an operation amount represented by the n-th solution and a first operation amount actual value representing an operation amount actual value that is the nearest to the operation amount represented by the n-th solution among the operation amount actual values ​​included in the set of performance data; 2. The driving assistance device according to claim 1, further comprising: a second distance representing an average of distances between an operation amount actual value included in the set of performance data and another operation amount actual value that is closest to the operation amount actual value; and when the first distance is equal to or less than the second distance, the deviation is calculated to be 0; and when the first distance is greater than the second distance, the deviation is calculated to be a positive value based on a ratio of the first distance to the second distance.

7. The deviation calculation unit The operation amount represented by the n-th solution and the top k operation amount actual values ​​included in the set of actual data that are close to the operation amount represented by the n-th solution １ (However, k １ a first distance representing an average of distances between each of the first and second manipulated variable actual values ​​(each of which is a predetermined integer equal to or greater than 1); Among the operation amount actual values ​​included in the set of performance data, the top k closest to each operation amount actual value ２ (However, k ２ 2. The driving assistance device according to claim 1, wherein the deviation degree is calculated to be 0 when the first distance is equal to or less than the second distance, and the deviation degree is calculated to be a positive value based on a ratio of the first distance to the second distance when the first distance is greater than the second distance, using a second distance that represents an average value of distances to a predetermined number of other operation amount actual values ​​(wherein x is a predetermined integer of 1 or more).

8. The driving assistance device according to claim 4 , wherein the distance is a Euclidean distance or a Mahalanobis distance.

9. a dimensionality reduction unit that reduces the dimension of an operation amount for the object by a predetermined compression mapping; A dimension restoration unit that restores the operation amount after the dimensional compression by inverse mapping of the compression mapping, The prediction model creation unit, The prediction model is created by using as an input at least an operation amount obtained by dimensionally compressing an operation amount for the target by the dimensionality compression unit; The initial solution generating unit creating the initial solution within a space to which the dimensionally reduced operation variable belongs; The optimization unit is Calculating the n-th solution in a space to which the dimensionally reduced operation variable belongs; The deviation calculation unit Calculating the degree of deviation within a space to which the dimensionally reduced operation amount belongs; The output unit is The driving support device according to claim 1 , further comprising: a control unit for controlling a driving force of the vehicle that is a vehicle that is driven by the vehicle speed controller;

10. A prediction model creation step of creating a prediction model for predicting a value of an evaluation item related to an object by inputting at least an operation amount for the object based on a set of performance data representing a past operation performance of the object; An initial solution generation procedure for generating an initial solution representing the n=0th solution; an optimization procedure for calculating, as an n-th solution, a manipulation amount that minimizes an objective function including an error between a predicted value predicted by the prediction model and a target value for the evaluation item, and a deviation degree that represents a degree of deviation between the (n-1)-th solution and a distribution of the set of performance data; a deviation calculation step of calculating a deviation representing a degree of deviation between the n-th solution and the distribution of the set of performance data; an objective function value calculation step of calculating a value of an objective function including an error between a predicted value predicted when an operation amount represented by the n-th solution is inputted at least to the prediction model and a target value for the evaluation item, and a deviation representing a degree of deviation between the n-th solution and a distribution of the set of performance data; an output step of outputting at least an operation amount represented by a solution having a smallest value calculated in the objective function value calculation step among the n=0th solution to the n=Nth solution (where N is a predetermined integer of 1 or more); A driving assistance method implemented by a computer.

11. A prediction model creation step of creating a prediction model for predicting a value of an evaluation item related to an object by inputting at least an operation amount for the object based on a set of performance data representing a past operation performance of the object; An initial solution generation procedure for generating an initial solution representing the n=0th solution; an optimization procedure for calculating, as an n-th solution, a manipulation amount that minimizes an objective function including an error between a predicted value predicted by the prediction model and a target value for the evaluation item, and a deviation degree that represents a degree of deviation between the (n-1)-th solution and a distribution of the set of performance data; a deviation calculation step of calculating a deviation representing a degree of deviation between the n-th solution and the distribution of the set of performance data; an objective function value calculation step of calculating a value of an objective function including an error between a predicted value predicted when an operation amount represented by the n-th solution is inputted at least to the prediction model and a target value for the evaluation item, and a deviation representing a degree of deviation between the n-th solution and a distribution of the set of performance data; an output step of outputting at least an operation amount represented by a solution having a smallest value calculated in the objective function value calculation step among the n=0th solution to the n=Nth solution (where N is a predetermined integer of 1 or more); A program that causes a computer to execute the following.

Images