JP7427746B1 - Information processing device, information processing method, and information processing program - Google Patents

Abstract

An object of the present invention is to easily determine the timing of switching to manual operation when the accuracy of automatic operation to be operated changes. SOLUTION: An information processing device 100 determines the operation of automatic driving of an operating target when data regarding the accuracy of automatic driving of the operating target satisfies a predetermined condition, and based on the determination result, the information processing device 100 determines whether the automatic driving of the operating target is operating. Stop driving. [Selection diagram] Figure 1

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

A known technology is imitation learning, which uses information about human behavior to create machine learning models that imitate human behavior. For example, supervised learning is known as a method for realizing the imitation learning described above. Furthermore, as a method of supervised learning, a large amount of observed data is accumulated, data in the vicinity of the required point is extracted from the accumulated data, and the extracted data is used to sequentially develop the model. A technology called the Just-In-Time (JIT) method for performing learning is known (see, for example, Non-Patent Document 1).

Furthermore, in recent years, technology has become known for automatically operating equipment, factories, plants, etc. to be operated using learning models that input operating data and the like. For example, as a conventional technology, there is a known technology that uses a learning model that uses acquired data as input to easily and accurately execute optimal control for the real environment in device control (for example, Patent Document 1).

JP2019-067238A

Shigeru Yamamoto, “Just-In-Time Predictive Control: Predictive Control Based on Accumulated Data,” Measurement and Control Vol. 52, No. 10, October 2013 issue (https://www.jstage.jst.go.jp/article /sicejl/52/10/52_878/_pdf/-char/ja)

However, the conventional technology has a problem in that it is difficult to determine the timing to switch to manual operation when the accuracy of automatic operation to be operated changes.

Specifically, the conventional technology presents a recommended value inferred based on a learning model to a user (such as an operator who operates or drives an operation target), and then provides a user with a recommended value for the operation target based on the presented recommended value. performs the operation. Therefore, when the accuracy of the recommended value deteriorates, the user can make the determination himself/herself. However, during automatic operation without user intervention, it is difficult to judge the accuracy of recommended values, and even when the accuracy of automatic operation decreases, it is sometimes difficult to determine whether to switch to manual operation at an appropriate timing.

Therefore, in order to solve the above problems and achieve the purpose, the information processing device of the present invention provides information about the operation of automatic driving to be operated, when data regarding the accuracy of automatic driving to be operated satisfies a predetermined condition. The present invention is characterized in that it has a determination unit that makes a determination, and a stop unit that stops automatic operation of the operation target based on the determination result of the determination unit.

The present invention has the effect of making it easier to determine the timing of switching to manual operation when the accuracy of automatic operation to be operated changes.

FIG. 1 is a diagram illustrating an example of an overview of information processing according to an embodiment. FIG. 2 is a diagram illustrating an example of inference of the learning model according to the embodiment. FIG. 3 is a diagram illustrating an example of the device configuration of the information processing device according to the embodiment. FIG. 4 is a diagram illustrating an example of an overall outline of information processing according to the embodiment. FIG. 5 is a diagram illustrating an example of an automatic driving management screen according to the embodiment. FIG. 6 is a diagram illustrating an example of an automatic driving management screen according to the embodiment. FIG. 7 is a diagram showing an example of an automatic driving management screen according to the embodiment. FIG. 8 is a diagram illustrating an example of an automatic driving management screen according to the embodiment. FIG. 9 is a diagram showing an overview of the abnormality detection algorithm according to the embodiment. FIG. 10 is a diagram showing an overview of the abnormality detection algorithm according to the embodiment. FIG. 11 is a diagram illustrating an example of a flowchart of information processing according to the first embodiment. FIG. 12 is a diagram illustrating an example of a flowchart of information processing according to the second embodiment. FIG. 13 is a diagram illustrating an example of a flowchart of information processing according to the third embodiment. FIG. 14 is a diagram illustrating an example of a flowchart of information processing according to the fourth embodiment. FIG. 15 is a diagram illustrating an example of a flowchart of information processing according to the fifth embodiment. FIG. 16 is a diagram illustrating an example of a flowchart of information processing according to the sixth embodiment. FIG. 17 is a diagram illustrating an example of an overview of information processing in the prior art. FIG. 18 is a diagram illustrating an example of a computer on which the information processing device according to the embodiment is implemented.

Hereinafter, a mode for implementing the present embodiment (hereinafter referred to as "embodiment") will be described with reference to the drawings. Note that this embodiment is not limited to the content described below.

[1. overview〕
First, an overview of information processing by the information processing apparatus 100 in this embodiment will be described using FIG. 1. In the present embodiment, the information processing device 100 stores past historical data Da (for example, temperature, The learning model 20 is trained using the user's operation history (pressure, flow rate, raw material input amount, production amount, etc.) (see (1) in FIG. 1).

Next, the user U (the operator who operates the operation target 10) inputs automatic driving conditions to the information processing device 100 (see (2) in FIG. 1). The automatic driving control unit 138 of the information processing device 100 uses the automatic driving conditions input by the user U and the recommended value RD inferred based on the above-mentioned learning model 20 inputting the driving data Db. (hereinafter, automatic operation for the operation target 10 will be simply referred to as "automatic operation") (see (3) and (4) in FIG. 1).

Then, the information processing device 100 acquires data regarding the accuracy of automatic driving of the operation target 10 (for example, recommended values, explanatory variables, evaluation indicators of the operation target, etc.) during automatic driving, and determines changes in the accuracy of automatic driving. . As a result, if it is determined that the accuracy of automatic driving is below the predetermined allowable range, the information processing device 100 stops automatic driving (see (5) in FIG. 1). Subsequently, the information processing device 100 displays to the user U that automatic driving has stopped (see (6) in FIG. 1). Then, the user U manually operates the operation target 10 after the automatic operation is stopped based on the display from the information processing device 100 (see (7) in FIG. 1).

[1-1. Inference of recommended values using learning model]
Next, the learning model 20 explained in FIG. 1 will be further explained. As shown in FIG. 2, the information processing device 100 receives driving data Db from the operation target 10a. Next, the information processing device 100 performs learning (training) of the learning model 20 for inferring the recommended value RD using the historical data Da similar to the received driving data Db.

The information processing device 100 then infers the recommended value RD based on the learned (trained) learning model 20. Specifically, the information processing device 100 performs imitation learning by learning the learning model 20 using history data Da, which is information such as operations actually performed by the user U on the operation target 10a. As a result, the information processing device 100 realizes automatic driving for the operation target 10b using the recommended value RD inferred by inputting the driving data Db into the learning model 20.

For example, when the operation target 10 is a plant, the learning model 20 learns the input amount of raw materials input by the user U in the past in a specific process. The learning model 20 then outputs the input amount of raw materials as the recommended value RD from the current operation data. By setting the input amount of raw materials according to the recommended value RD inferred based on the learning model 20, the user U can imitate the operations of past users (for example, the user U or an operator other than the user U).

[2. Configuration of information processing device]
From here, the configuration of the information processing device 100 according to this embodiment will be explained using FIG. 3. As shown in FIG. 3, the information processing device 100 includes a communication section 110, a storage section 120, and a control section 130. Although not shown, the information processing device 100 may include an input unit (for example, a keyboard, a mouse, etc.) that accepts various operations, and a display unit (for example, a display) for displaying various information. . Next, detailed functions of each part will be described below.

(Communication Department 110)
The communication unit 110 is realized by a NIC (Network Interface Card) or the like, and controls communication via a telecommunication line such as a LAN (Local Area Network) or the Internet. The communication unit 110 is connected to a network by wire or wirelessly as necessary, and can transmit and receive information in both directions. In this embodiment, it is assumed that communication with an external device (for example, the operation target 10, etc.) is performed via the communication unit 110.

(Storage unit 120)
The storage unit 120 stores data and programs necessary for various processing by the control unit 130. Furthermore, the storage unit 120 includes a history data storage unit 121 , a model storage unit 122 , and a recommended value storage unit 123 . The storage unit 120 is realized by a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk.

(History data storage unit 121)
The history data storage unit 121 stores history data as information regarding past driving of the operation target 10. For example, the history data stored in the history data storage unit 121 includes past explanatory variables of the operation target 10 (for example, sensor data such as time, temperature, pressure, carbon dioxide concentration, etc.) and objective variables (for example, user operation history) and includes history data Da having similar explanatory variables to the driving data Db. Further, the above-mentioned information is just an example, and the historical data storage unit 121 can store any historical data without limitation.

(Model storage unit 122)
The model storage unit 122 stores a learning model 20 that is trained using the history data Da of the operation target 10.

(Recommended value storage unit 123)
The recommended value storage unit 123 stores the recommended value RD inferred by the learning model 20.

(Control unit 130)
The control unit 130 includes an acquisition unit 131, a learning unit 132, an update unit 133, an inference unit 134, a determination unit 135, a stop unit 136, a display unit 137, and an automatic driving control unit 138. The control unit 130 has an internal memory for temporarily storing programs and processing data that define various processing procedures, and includes electronic circuits such as a CPU (Central Processing Unit) and an MPU (Micro Processing Unit). , is realized by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

(Acquisition unit 131)
The acquisition unit 131 acquires data regarding the operation of the operation target 10. Specifically, the acquisition unit 131 acquires driving data Db of the operation target 10 (for example, sensor data collected by the operation target 10 that is used as an explanatory variable) as data related to the operation of the operation target 10. do.

Furthermore, the acquisition unit 131 uses a history search key (current sensor data of the operation target 10, hereinafter simply referred to as "history search key") to obtain data related to the operation of the operation target 10. History data Da (including similar sensor data and operation history by the user) similar to the driving data Db at the time of driving is acquired from the history data storage unit 121. In addition to the information described above, the acquisition unit 131 can acquire information without limitation as long as it is in the category of data related to the operation of the operation target 10.

(Learning section 132)
The learning unit 132 uses the history data Da of the operation target 10 acquired by the acquisition unit 131 to learn (train) the learning model 20 using explanatory variables (sensor data) and objective variables (operation history by the user) as learning data. conduct. For example, when the operation target 10 is a plant, the learning unit 132 can learn the learning model 20 using the input amount of raw materials input by the user U in the past in a specific situation as the history data Da.

Further, when the learning model 20 is an ensemble model, the learning unit 132 may perform learning using techniques such as bagging, boosting, and stacking.

(Update unit 133)
The updating unit 133 updates the learning model 20 based on the learning by the learning unit 132. Furthermore, the updating unit 133 updates the plurality of learning models 20 based on the learning by the learning unit 132. In this embodiment, the learning model 20 is an ensemble model, and the information processing apparatus 100 will be described on the assumption that the information processing apparatus 100 has a plurality of learning models 20.

(Inference unit 134)
The inference unit 134 infers the recommended value RD based on the learning model 20 that receives as input the driving data Db (for example, explanatory variables, etc.) that the acquisition unit 131 acquires from the operation target 10. Note that in the case of an ensemble model in which a plurality of learning models 20 exist, the inference unit 134 may infer the recommended value RD to be inferred by a majority vote of the plurality of learning models 20, an average, or the like.

(Determination unit 135)
The determining unit 135 determines whether the automatic driving of the operating target 10 is performed when the data regarding the accuracy of the automatic driving of the operating target 10 satisfies a predetermined condition. Specifically, the determination unit 135 makes the determination using an explanatory variable, a recommended value RD, a prediction variance, and an evaluation index of the operation target as data regarding the accuracy of automatic driving of the operation target 10. Note that details of the determination performed by the determination unit 135 will be explained for each embodiment in the following items.

(Stop part 136)
The stopping unit 136 stops automatic operation of the operation target 10 based on the determination result of the determining unit 135. Specifically, the stop unit 136 stops automatic operation performed by an automatic operation control unit 138, which will be described later.

(Display section 137)
The display unit 137 displays to the user U that the automatic operation of the operation target 10 has been stopped when the stop unit 136 has stopped the automatic operation of the operation target 10. Note that the display unit 137 may use a method that the user U can perceive, such as text or audio, as a method of displaying the information to the user U, for example.

Further, the display unit 137 may display information regarding the automatic driving stop based on the decrease in the accuracy of the automatic driving to the user U even when the automatic driving is not stopped automatically by the stopping unit 136. For example, the display unit 137 may display content such as "Due to a decrease in the accuracy of automatic driving, it is recommended that automatic driving be stopped."

(Automatic operation control unit 138)
The automatic driving control unit 138 performs automatic driving on the operation target 10 using the recommended value RD inferred based on the learning model 20 inputting the driving data Db (explanatory variable).

[3. Overall image of information processing according to embodiment]
From here, the overall image of information processing performed by the information processing apparatus 100 in this embodiment will be described using FIG. 4. In FIG. 4, the information processing device 100 uses the methods of Embodiments 1 to 6, which will be described later, based on data regarding the accuracy of the automatic driving of the operation target 10, to prevent the automatic driving from stopping due to a decrease in the accuracy of the automatic driving. The procedure for making a determination will be explained. Note that the information processing apparatus 100 may implement each of Embodiments 1 to 6 independently, or may implement any of the embodiments in combination.

The acquisition unit 131 acquires past history data Da similar to or related to the driving data Db of the operation target 10 from the history data storage unit 121 using the history search key (see (1) in FIG. 4). Subsequently, the learning unit 132 learns the learning model 20 using the acquired history data Da (see (2) in FIG. 4).

The updating unit 133 updates the plurality of learning models 20 (learning model 20a, learning model 20b, and learning model 20c) (see (3) in FIG. 4). Although three learning models 20 are shown in FIG. 4, the number of learning models 20 is not limited to three, and a different number of learning models 20 may be used as necessary.

The inference unit 134 performs multiple inferences based on the multiple trained learning models 20 (see (4) in FIG. 4). Then, the inference unit 134 inputs the driving data Db (explanatory variable) acquired by the acquisition unit 131 (see (5) in FIG. 4), and infers the recommended value RD based on the learning model 20 (see (5) in FIG. 4). (see 6)). Note that in this embodiment, when inferring a plurality of recommended values RD using a plurality of learning models 20, the inference unit 134 may calculate by majority vote, average, etc. of the plurality of models.

From here, the determination of driving accuracy based on various determination criteria by the determination unit 135 will be explained. The determination unit 135 determines whether a predetermined change has occurred in the driving data Db (explanatory variable) using an abnormality detection algorithm (for example, one using PCA (Principal Component Analysis), etc.) (see FIG. 4 (see (7)). Note that the process (7) in FIG. 4 will be described in the following items as Embodiment 1.

The determination unit 135 determines whether the recommended value RD inferred by the inference unit 134 is within a predetermined range (for example, maximum and minimum threshold values, etc.) (see (8) in FIG. 4). Note that the process (8) in FIG. 4 will be described in the following items as Embodiment 2.

The determination unit 135 calculates the rate of change in a predetermined period for the recommended value RD inferred by the inference unit 134, and determines whether the above-mentioned rate of change is within a predetermined range (for example, a differential change amount threshold, etc.). (See (9) in FIG. 4). Note that the process (9) in FIG. 4 will be explained in the following items as the third embodiment.

The determination unit 135 compares the recommended values RD of the plurality of learning models 20 based on the variance, and determines whether the aforementioned variance exceeds a predetermined threshold (for example, the variance threshold of an ensemble model, etc.). (See (10) in Figure 4). Note that the process (10) in FIG. 4 will be explained in the following items as Embodiment 4.

The determination unit 135 determines whether the predicted variance of the posterior distribution of the recommended value RD exceeds a predetermined threshold (for example, a predicted variance threshold, etc.) when the recommended value RD inferred by the inference unit 134 has a posterior distribution. (See (11) in FIG. 4). Note that the process (11) in FIG. 4 will be described in the following items as Embodiment 5.

The determination unit 135 determines whether the evaluation index (see (12) in FIG. 4) of the operation target 10 exceeds a predetermined threshold (for example, evaluation index threshold, etc.) (see (13) in FIG. 4). ). Note that the process (13) in FIG. 4 will be described in the following items as Embodiment 6.

If the determination unit 135 determines that the driving accuracy has decreased based on various determination criteria, it further determines whether to stop automatic driving (if the determination unit 135 determines that the driving accuracy has decreased based on (14) in FIG. 4), reference). Note that the determination unit 135 may determine whether or not a predetermined condition is satisfied using one or more of the aforementioned determination criteria.

The stopping unit 136 stops automatic operation based on the determination by the determining unit 135 described above (see (15) in FIG. 4). Subsequently, the display unit 137 displays to the user U that the automatic driving has been stopped based on the stopping of the automatic driving by the stopping unit 136 (see (16) in FIG. 4). Then, the user U manually operates the operation target 10 whose automatic operation has been stopped (see (17) in FIG. 4).

On the other hand, when the determination unit 135 determines that the driving accuracy has not decreased based on the determination criteria described above (see “No decrease in driving accuracy” in (14) of FIG. 4), the automatic driving control unit 138 continues automatic operation (see (18) in Figure 4).

From here, an example of a method for notifying the user U about the stop of automatic driving described in (16) of FIG. 4 will be explained using FIGS. 5 to 8. First, a system screen displayed on a terminal device operated by user U will be described using FIG. 5. Note that the system screen described in this section is just an example, and the display format, display content, screen arrangement, configuration, combination, etc. are not limited and may be changed as necessary.

Screen SA in FIG. 5 represents visualized information about fluctuations in data related to driving in automatic driving. On the screen SA, data that changes sequentially due to automatic driving is continuously displayed in a chronological direction. On the other hand, a list of data related to driving in automatic driving is displayed on screen SB. Note that the items displayed on the screen SA may be automatically selected by the information processing apparatus 100, or may be selected by the user U himself. If the user U makes the selection himself/herself, the display on the screen SA may be switched based on the items displayed on the screen SB.

In the display SC1 of FIG. 5, information regarding the operating status of automatic operation is displayed. For example, in the case of FIG. 5, when the automatic operation is in operation, the display unit 137 displays the text "Automatic operation in operation" on the display SC1. Then, when the stop unit 136 stops the automatic operation, the display unit 137 displays the text "Automatic operation is stopped" on the display SC2 in FIG. 6. Note that the text content described above is just an example, and the display unit 137 may output other text, images, audio, etc. that can be perceived by the user's five senses as necessary in order to display the driving situation to the user U. Can be used.

In addition, although we have explained the automatic stop of automatic driving by the stop unit 136 up to this point, when the user U receives information about a decrease in the accuracy of automatic driving displayed on the display unit 137, the user U can use his own operation to stop the automatic driving. may be stopped. For example, as shown in display SC3 in FIG. 7 and display SC4 in FIG. You can also switch between stopping and restarting.

[4. Embodiment 1: Judgment based on abnormality detection of explanatory variables]
From here, the above-described driving accuracy determination method will be described in the order of "overview", "configuration of information processing device 100", and "processing procedure" for Embodiments 1 to 6, respectively. Note that the information processing apparatus 100 may implement each of Embodiments 1 to 6 independently, or may implement a plurality of embodiments in combination. For example, the information processing device 100 may determine whether to stop automatic driving based on a single determination result for each embodiment or multiple determination results in multiple embodiments.

First, as Embodiment 1, "determination based on abnormality detection of explanatory variables" will be described. The inference unit 134 calculates the recommended value RD by performing inference based on the learning model 20 that receives the driving data Db (explanatory variables) collected by the operation target 10 as input. Then, the automatic driving control unit 138 performs automatic driving on the operation target 10 using the inferred recommended value RD. However, the driving situation of the operation target 10 may change over time, and the accuracy of automatic driving may deteriorate accordingly. Note that the above-mentioned content is common to Embodiment 1 to Embodiment 6, so the subsequent description will be omitted.

As described above, when the accuracy of automatic driving of the operation target 10 decreases, a change (abnormality) may occur in the driving data Db (explanatory variable) that is input to the learning model 20. Therefore, the determination unit 135 of the first embodiment evaluates the change in the driving data Db (explanatory variable) using an abnormality detection algorithm (for example, one using PCA or the like), and determines whether to stop automatic driving.

[4-1. Configuration of information processing device of Embodiment 1]
The device configuration of the information processing device 100 in Embodiment 1 is the same as in the above-described embodiments. Therefore, in this section, only the additional functions of the determination unit 135 will be explained as differences, and other detailed explanations will be omitted.

(Determination unit 135)
The determining unit 135 in the first embodiment uses an abnormality detection algorithm to determine a change that occurs in the driving data Db (explanatory variable). Specifically, the determination unit 135 sets a predetermined condition to the driving data Db (explanatory variable) based on the result of abnormality detection regarding the driving data Db (explanatory variable) representing the driving situation of the operation target 10. If a change occurs, determine whether to stop automatic operation. Note that the determination unit 135 may perform determination based on an abnormality detection algorithm using unsupervised machine learning (for example, using PCA or the like).

For example, the determination unit 135 performs KDE (kernel density estimation) on a space reduced by PCA (principal component analysis), and if data appears in a region with low density (for example, if the likelihood is low), ) may be determined to be an abnormal occurrence. Further, as another example, the determination unit 135 may determine that an abnormality has occurred using a distance index known as Mahalanobis distance. Furthermore, as another example, the determination unit 135 may determine the occurrence of an abnormality using the T^2 statistic and the Q statistic calculated on the space contracted by PCA.

An example of anomaly detection using an anomaly detection algorithm based on PCA will now be described. As shown in FIG. 9, the determination unit 135 calculates, by reduction, a region (manifold) in which the majority of normal data exists in the learning target data, and determines that an abnormality occurs if a predetermined distance is away from the region (manifold). It may be determined that this has occurred (for example, see values a and b in FIG. 9). Note that the Mahalanobis distance d shown in FIG. 9 is expressed by the following formula (1). Note that in formula (1), Σ is a "variance-covariance matrix," x is a "variable," and μ is an "average of x."

Specifically, in FIG. 10, the T^2 statistic indicates the distance within the contracted dimension (within the dimension in which normal data mainly exists), and the determination unit 135 calculates the distance based on the above-mentioned distance. An abnormal value is determined based on the value (for example, see value a in FIG. 10). On the other hand, the Q statistic indicates the distance of a dimension that is omitted from the reduction (a value that jumps out in a dimension where normal data does not mainly exist), and the determination unit 135 determines the abnormal value based on the above-mentioned distance. A determination is made (for example, see value b in FIG. 10).

On the other hand, the determination unit 135 may determine that an abnormality has occurred when a predetermined distance is away from the main area of data with high probability density, based on likelihood. Furthermore, the determination unit 135 is not limited to abnormality detection based on PCA, and may determine the occurrence of an abnormality using other methods. For example, as a method using a generative model, the determination unit 135 may determine the occurrence of an abnormality using an autoencoder, which is a reduction method using a neural network, a GAN (Generative Adversarial Network), or the like.

In the above-mentioned method using an autoencoder, the determination unit 135 determines the reconstruction error in "input → reduction → input reconstruction" by saying, "If the reproducibility after learning is low, the majority of normal The degree of abnormality may be determined based on the premise that "learning is being performed using abnormal samples that are separated from the data by a predetermined distance." On the other hand, in the method using GAN, the determination unit 135 uses a learned discriminator (a model that determines whether the GAN product and true data are correct or incorrect) and determines whether an abnormality has occurred if the data is not determined to be true. It can be determined that the

As an example, when using plant process data, the sensor values collected by the plant are processed as multidimensional data with dimensions of "number of sensors x window width" in a time window, and this is used as input to calculate the difference between the normal state and The determining unit 135 may determine whether the predetermined distance is away using the method described above.

[4-2. Processing procedure of Embodiment 1]
Next, the procedure of the information processing method of the information processing apparatus 100 in the first embodiment will be described using FIG. 11. First, the operation target 10 collects driving data Db (explanatory variable) (step S101). Next, the determination unit 135 determines a state change occurring in the driving data Db (explanatory variable) using an abnormality detection algorithm (for example, one using PCA or the like) (step S102).

The determination unit 135 determines that the change occurring in the driving data Db (explanatory variable) is not included in the predetermined range (No in step S103). In that case, the determination unit 135 determines to stop automatic operation (step S104). Then, the stopping unit 136 stops automatic operation based on the determination by the determining unit 135 (step S105). Subsequently, the display unit 137 displays to the user U that the automatic operation has stopped (step S106), and the process ends.

On the other hand, the determination unit 135 determines that the state change occurring in the driving data Db (explanatory variable) is included in a predetermined range (Yes in step S103). In that case, the process returns and processing continues.

[5. Embodiment 2: Judgment based on recommended values]
Next, "determination based on recommended values" will be described as a second embodiment. When the accuracy of automatic operation of the operation target 10 decreases, the recommended value RD may change. Therefore, the determination unit 135 of the second embodiment determines whether or not the automatic driving should be stopped based on whether the recommended value RD is within a predetermined range (for example, maximum or minimum threshold value, etc.).

[5-1. Configuration of information processing device of embodiment 2]
The device configuration of the information processing device 100 in the second embodiment is the same as that in the above-described embodiment. Therefore, in this section, only the additional functions of the determination unit 135 will be explained as a difference, and detailed explanations other than that will be omitted.

(Determination unit 135)
The determination unit 135 in the second embodiment determines, as a predetermined condition, that the recommended value RD inferred based on the learning model 20 using the historical data Da regarding the driving of the operation target 10 is not included in the range of maximum and minimum threshold values. Determine whether to stop automatic driving. Note that the determination unit 135 may set an arbitrary range for the maximum and minimum threshold values as necessary. In addition, the determination unit 135 may perform other determination methods, such as a conventional method of learning a model using the amount of operation (true value) as learning data to perform anomaly detection, or a method of performing conventional operation using a discriminator or an anomaly detection algorithm. A classification method for identifying whether the amount (true value) or the recommended value RD, etc. may be used.

[5-2. Processing procedure of Embodiment 2]
Next, the procedure of the information processing method of the information processing apparatus 100 in the second embodiment will be described using FIG. 12. First, the operation target 10 collects driving data Db (step S201). Next, the acquisition unit 131 uses the history search key to acquire history data Da similar to the driving data Db at the time of operation of the operation target 10 from the history data storage unit 121 (step S202).

The learning unit 132 performs learning of the learning model 20 using the similar history data Da acquired by the acquisition unit 131 (step S203). Subsequently, the updating unit 133 updates the learning model 20 (step S204). Then, the inference unit 134 infers the recommended value RD based on the updated learning model 20 inputting the driving data Db (explanatory variable) (step S205).

The determining unit 135 determines that the inferred recommended value RD is not included in a predetermined range (for example, maximum/minimum threshold value, etc.) (No in step S206). In that case, the determination unit 135 determines to stop automatic operation (step S207). Then, the stopping unit 136 stops automatic operation based on the determination by the determining unit 135 (step S208). Subsequently, the display unit 137 displays to the user U that the automatic operation has stopped (step S209), and the process ends.

On the other hand, the determination unit 135 determines that the inferred recommended value RD is included in a predetermined range (for example, maximum and minimum threshold values, etc.) (Yes in step S206). In that case, the process returns and processing continues.

[6. Embodiment 3: Judgment based on rate of change in recommended value]
Next, as Embodiment 3, "determination based on rate of change in recommended value" will be described. When the accuracy of automatic operation of the operation target 10 decreases, the recommended value RD may also change. Therefore, the determination unit 135 of the third embodiment determines whether or not automatic driving should be stopped based on whether the rate of change of the recommended value RD in a predetermined period is within a predetermined range (for example, a differential change amount threshold, etc.). do.

[6-1. Configuration of information processing device of embodiment 3]
The device configuration of the information processing device 100 in Embodiment 3 is the same as in the above-described embodiments. Therefore, in this section, only the additional functions of the determination unit 135 will be explained as a difference, and detailed explanations other than that will be omitted.

(Determination unit 135)
The determination unit 135 in the third embodiment determines, as a predetermined condition, that the rate of change in the recommended value RD inferred based on the learning model 20 using the historical data Da regarding the driving of the operation target 10 is included in the range of the difference change amount threshold. determines whether to stop automatic operation if the Note that the determination unit 135 may set an arbitrary range for the difference change amount threshold value as necessary. In addition, the determination unit 135 may perform other determination methods, such as a conventional method of learning a model using the amount of operation (true value) as learning data to perform anomaly detection, or a method of performing conventional operation using a discriminator or an anomaly detection algorithm. A classification method for identifying whether the amount (true value) or the recommended value RD, etc. may be used.

[6-2. Processing procedure of Embodiment 3]
Next, the procedure of the information processing method of the information processing apparatus 100 in the third embodiment will be described using FIG. 13. First, the operation target 10 collects driving data Db (step S301). Next, the acquisition unit 131 uses the history search key to acquire history data Da similar to the driving data Db at the time of operation of the operation target 10 from the history data storage unit 121 (step S302).

The learning unit 132 performs learning of the learning model 20 using the similar history data Da acquired by the acquisition unit 131 (step S303). Subsequently, the updating unit 133 updates the learning model 20 (step S304). Then, the inference unit 134 infers the recommended value RD based on the updated learning model 20 inputting the driving data Db (explanatory variable) (step S305).

The determination unit 135 determines that the rate of change of the inferred recommended value RD in a predetermined period is not included in a predetermined range (for example, a difference change amount threshold, etc.) (No in step S306). In that case, the determination unit 135 determines to stop automatic operation (step S307). Then, the stopping unit 136 stops automatic operation based on the determination by the determining unit 135 (step S308). Subsequently, the display unit 137 displays to the user U that the automatic operation has stopped (step S309), and the process ends.

On the other hand, the determination unit 135 determines that the rate of change of the inferred recommended value RD in a predetermined period is included in a predetermined range (for example, a difference change amount threshold, etc.) (Yes in step S306). In that case, the process returns and processing continues.

[7. Embodiment 4: Judgment based on ensemble variance threshold]
Next, as Embodiment 4, "determination based on ensemble variance threshold" will be described. When the accuracy of automatic driving of the operation target 10 decreases, the variance of the plurality of recommended values RD inferred based on the plurality of updated learning models 20 may increase. Therefore, the determination unit 135 of the fourth embodiment determines whether the variance of the plurality of recommended values RD inferred based on the plurality of learning models 20 exceeds a predetermined threshold (for example, the variance threshold of an ensemble model, etc.). Based on whether or not the automatic operation is stopped, it is determined whether the automatic operation is to be stopped or not.

[7-1. Configuration of information processing device of embodiment 4]
The device configuration of the information processing device 100 in the fourth embodiment is the same as that in the above-described embodiments. Therefore, in this section, only the additional functions of the determination unit 135 will be explained as a difference, and detailed explanations other than that will be omitted.

(Determination unit 135)
The determination unit 135 in the fourth embodiment determines, as a predetermined condition, that the variance of the plurality of recommended values RD inferred based on the plurality of learning models 20 using the historical data Da regarding the driving of the operation target 10 is included in a predetermined range. If not, determine whether to stop automatic operation. Note that the determination unit 135 may set any range as necessary for the above-mentioned predetermined range.

[7-2. Processing procedure of Embodiment 4]
Next, the procedure of the information processing method of the information processing apparatus 100 in the fourth embodiment will be described using FIG. 14. Note that this item will be explained on the assumption that there are a plurality of learning models 20 and a plurality of recommended values RD inferred based on the learning models 20. Further, in embodiments other than the fourth embodiment (Embodiments 1 to 3 and Embodiments 5 and 6 described later), a plurality of learning models 20 and recommended values RD may exist.

First, the operation target 10 collects driving data Db (step S401). Next, the acquisition unit 131 uses the history search key to acquire history data Da similar to the driving data Db at the time of operation of the operation target 10 from the history data storage unit 121 (step S402).

The learning unit 132 performs learning of the learning model 20 using the similar history data Da acquired by the acquisition unit 131 (step S403). Subsequently, the updating unit 133 updates the plurality of learning models 20 (step S404). Then, the inference unit 134 infers the recommended value RD based on the updated learning model 20 inputting the driving data Db (explanatory variable) (step S405).

If the variance of the learning model 20 is not included in the predetermined range, in other words, the variance of the plurality of recommended values RD inferred based on the plurality of learning models 20, the determination unit 135 determines that the above-mentioned variance is within a predetermined threshold (for example, variance threshold of the ensemble model, etc.) is exceeded (No in step S406). In that case, the determination unit 135 determines to stop automatic operation (step S407). Then, the stopping unit 136 stops automatic operation based on the determination by the determining unit 135 (step S408). Subsequently, the display unit 137 displays to the user U that the automatic operation has stopped (step S409), and the process ends.

On the other hand, the determination unit 135 determines that the variance of the learning model 20 is within a predetermined range, in other words, the variance of the plurality of recommended values RD inferred based on the plurality of learning models 20 is determined to be within a predetermined threshold (e.g. , ensemble model variance threshold, etc.) (Yes in step S406). In that case, the process returns and processing continues.

[8. Embodiment 5: Judgment based on predicted variance of recommended values]
Next, as a fifth embodiment, "determination based on predicted variance of recommended values" will be described. If the recommended value RD inferred by the inference unit 134 has a posterior distribution when the accuracy of automatic driving of the operation target 10 decreases, the predicted variance of the recommended value RD may become large. Therefore, the determination unit 135 of the fifth embodiment determines whether to stop automatic driving based on whether the predicted variance of the posterior distribution of the recommended value RD described above exceeds a predetermined threshold (for example, a predicted variance threshold, etc.). .

[8-1. Configuration of information processing device of embodiment 5]
The device configuration of the information processing device 100 in the fifth embodiment is the same as that in the above-described embodiments. Therefore, in this section, only the additional functions of the determination unit 135 will be explained as a difference, and detailed explanations other than that will be omitted.

(Determination unit 135)
When the recommended value RD inferred based on the learning model 20 using the historical data Da regarding the driving of the operation target 10 has a posterior distribution, the determination unit 135 in the fifth embodiment determines, as a predetermined condition, the predicted variance of the recommended value RD. If the value is not within a predetermined range, a determination is made to stop automatic operation. Note that the determination unit 135 may set any range as necessary for the above-mentioned predetermined range.

[8-2. Processing procedure of Embodiment 5]
Next, the procedure of the information processing method of the information processing apparatus 100 in the fifth embodiment will be described using FIG. 15. Note that in the processing procedure of the fifth embodiment, the explanation will be given on the premise that the inferred recommended value RD has a posterior distribution.

First, the operation target 10 collects driving data Db (step S501). Next, the acquisition unit 131 uses the history search key to acquire history data Da similar to the driving data Db at the time of operation of the operation target 10 from the history data storage unit 121 (step S502).

The learning unit 132 performs learning of the learning model 20 using the similar history data Da acquired by the acquisition unit 131 (step S503). Subsequently, the updating unit 133 updates the learning model 20 (step S504). Then, the inference unit 134 infers the recommended value RD based on the updated learning model 20 inputting the driving data Db (explanatory variable) (step S505).

The determination unit 135 determines that the predicted variance of the posterior distribution of the recommended value RD is not included in a predetermined range, in other words, the predicted variance of the posterior distribution of the recommended value RD exceeds a predetermined threshold (for example, a predicted variance threshold). (No in step S506). In that case, the determination unit 135 determines to stop automatic operation (step S507). Then, the stopping unit 136 stops automatic operation based on the determination by the determining unit 135 (step S508). Subsequently, the display unit 137 displays to the user U that the automatic operation has stopped (step S509), and the process ends.

On the other hand, the determination unit 135 determines that the predicted variance of the posterior distribution of the recommended value RD is included in a predetermined range, in other words, the predicted variance of the posterior distribution of the recommended value RD does not exceed a predetermined threshold (for example, a predicted variance threshold). Determination is made (Yes in step S506). In that case, the process returns and processing continues.

[9. Embodiment 6: Judgment based on evaluation index of operation target]
Next, as Embodiment 6, "determination based on evaluation index of operation target" will be described. When the accuracy of automatic operation of the operation object 10 decreases, the evaluation index of the operation object 10 (for example, the production amount, stability, etc. of the operation object 10) may change. Therefore, the determination unit 135 of the sixth embodiment determines whether to stop automatic driving based on whether or not the evaluation index of the operation target 10 exceeds a predetermined threshold (for example, an evaluation index threshold).

[9-1. Configuration of information processing device of embodiment 6]
The device configuration of the information processing device 100 in Embodiment 6 is the same as in the above-described embodiments. Therefore, in this section, only the additional functions of the determination unit 135 will be explained as differences, and other detailed explanations will be omitted.

(Determination unit 135)
The determination unit 135 in the sixth embodiment determines to stop automatic driving when the evaluation index of the operation target 10 is not included in a predetermined range as a predetermined condition. Note that the determination unit 135 may set any range as necessary for the above-mentioned predetermined range.

[9-2. Processing procedure of Embodiment 6]
Next, the procedure of the information processing method of the information processing apparatus 100 in the sixth embodiment will be described using FIG. 16. First, the operation target 10 collects driving data Db (step S601). Next, the acquisition unit 131 uses the history search key to acquire history data Da similar to the driving data Db at the time of operation of the operation target 10 from the history data storage unit 121 (step S602).

The learning unit 132 performs learning of the learning model 20 using the similar history data Da acquired by the acquisition unit 131 (step S603). Subsequently, the updating unit 133 updates the learning model 20 (step S604). Then, the inference unit 134 infers the recommended value RD based on the updated learning model 20 inputting the driving data Db (explanatory variable) (step S605). Thereafter, the automatic driving control unit 138 performs automatic driving using the inferred recommended value RD (step S606).

The determination unit 135 determines that the evaluation index of the operation target 10 obtained as a result of automatic driving is not included in a predetermined range, in other words, the evaluation index of the operation target 10 exceeds a predetermined threshold (for example, an evaluation index threshold). (No in step S607). In that case, the determination unit 135 determines to stop automatic operation (step S608). Then, the stopping unit 136 stops automatic operation based on the determination by the determining unit 135 (step S609). Subsequently, the display unit 137 displays to the user U that the automatic operation has stopped (step S610), and the process ends.

On the other hand, the determination unit 135 determines that the evaluation index of the operation target 10 is included in a predetermined range, in other words, the evaluation index of the operation target 10 does not exceed a predetermined threshold (for example, an evaluation index threshold) (step S607 (Yes). In that case, the process returns and processing continues.

[10. effect〕
In the prior art, the user U manually operates the operation target 10 using the recommended value RD inferred based on the learning model 20 inputting the driving data Db. Here, presentation of the recommended value RD according to the prior art and manual operation of the operation target 10 by the user U will be explained using FIG. 17.

In FIG. 17, the information processing device 1 according to the prior art has past historical data Da (for example, The learning model 20 is trained using the user's operation history (temperature, pressure, flow rate, raw material input amount, production amount, etc.) (see (1) in FIG. 17). Next, the information processing device 1 displays, on the guidance screen 30, the recommended value RD inferred based on the above-described learning model 20, which takes the driving data Db as input (see (2) in FIG. 17), and displays the recommended value RD for the user U. (See (3) in FIG. 17). The user U then manually operates the operation target 10 based on the display from the information processing device 1 (see (4) in FIG. 17).

As described above, in the conventional technology, the user U manually operates the operation target 10 based on the recommended value RD that is presented, so even when the accuracy of the recommended value RD changes, the user U himself or herself cannot continue or stop operation. It was possible to make a judgment. However, it is difficult to judge the accuracy of the recommended value RD during automatic operation without the intervention of the user U, and even if the accuracy of the recommended value RD changes, it may be difficult to determine whether to switch to manual operation at an appropriate timing. Ta.

Therefore, the information processing device 100 in the present embodiment makes a determination regarding the operation of automatic driving of the operating target 10 when the data regarding the accuracy of automatic driving of the operating target 10 satisfies a predetermined condition, and determines the determination by the determining unit 135. Based on the result, automatic operation of the operation target 10 is stopped. Therefore, according to this embodiment, the information processing device 100 has the following effects.

The information processing device 100 has the effect of making it easier to determine the timing of switching to manual operation when the accuracy of automatic operation of the operation target 10 changes.

Furthermore, the information processing device 100 dynamically determines whether or not automatic driving can be continued based on data related to the driving of the operation target 10, so that automatic driving can be performed without being influenced by user U's experience, proficiency level, skill, etc. This has the effect of making it possible to safely stop automatic driving.

[11. Hardware configuration]
Each component of each device shown in the drawings is functionally conceptual, and does not necessarily need to be physically configured as shown in the drawings. In other words, the specific form of distributing and integrating each device is not limited to what is shown in the diagram, and all or part of the devices can be functionally or physically distributed or integrated in arbitrary units depending on various loads and usage conditions. Can be integrated and configured. Furthermore, all or any part of each processing function performed by each device can be realized by a CPU and a program that is analyzed and executed by the CPU, or can be realized as hardware using wired logic.

Further, among the processes described in this embodiment, all or part of the processes described as being performed automatically can also be performed manually by a known method. In addition, information including processing procedures, control procedures, specific names, and various data and parameters shown in the drawings can be arbitrarily changed unless otherwise specified.

[program]
As one embodiment, the information processing apparatus 100 can be implemented by installing an information processing program that executes the above-described information processing method on a desired computer as package software or online software. For example, by causing the information processing apparatus to execute the above information processing program, the information processing apparatus can function as the information processing apparatus 100. The information processing device referred to here includes a desktop or notebook personal computer. In addition, information processing devices include mobile communication terminals such as smartphones and mobile phones, and slate terminals such as PDAs (Personal Digital Assistants).

FIG. 18 is a diagram illustrating an example of a computer on which the information processing device 100 is implemented. Computer 1000 includes, for example, a memory 1010 and a CPU 1020. The computer 1000 also includes a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. These parts are connected by a bus 1080.

Memory 1010 includes ROM (Read Only Memory) 1011 and RAM 1012. The ROM 1011 stores, for example, a boot program such as BIOS (Basic Input Output System). Hard disk drive interface 1030 is connected to hard disk drive 1090. Disk drive interface 1040 is connected to disk drive 1100. For example, a removable storage medium such as a magnetic disk or an optical disk is inserted into disk drive 1100. Serial port interface 1050 is connected to, for example, mouse 1110 and keyboard 1120. Video adapter 1060 is connected to display 1130, for example.

The hard disk drive 1090 stores, for example, an OS 1091, application programs 1092, program modules 1093, and program data 1094. That is, a program that defines each process of the information processing apparatus 100 is implemented as a program module 1093 in which computer-executable code is written. Program module 1093 is stored in hard disk drive 1090, for example. For example, a program module 1093 for executing processing similar to the functional configuration of the information processing apparatus 100 is stored in the hard disk drive 1090. Note that the hard disk drive 1090 may be replaced by an SSD (Solid State Drive).

Further, the setting data used in the processing of the embodiment described above is stored as program data 1094 in, for example, the memory 1010 or the hard disk drive 1090. Then, the CPU 1020 reads out the program module 1093 and program data 1094 stored in the memory 1010 and the hard disk drive 1090 to the RAM 1012 as necessary, and executes the processing of the embodiment described above.

Note that the program module 1093 and the program data 1094 are not limited to being stored in the hard disk drive 1090, but may be stored in a removable storage medium, for example, and read by the CPU 1020 via the disk drive 1100 or the like. Alternatively, program module 1093 and program data 1094 may be stored in another computer connected via a network (LAN, WAN (Wide Area Network), etc.). Program module 1093 and program data 1094 may then be read by CPU 1020 from another computer via network interface 1070.

[12. others〕
Although this embodiment has been described above, this embodiment is not limited by the description and drawings that form part of the disclosure. That is, all other embodiments, examples, operational techniques, etc. made by those skilled in the art based on this embodiment are included in the scope of this embodiment.

1 Information processing device 10 Operation target 10a Operation target 10b Operation target 20 Learning model 30 Guidance screen 100 Information processing device 110 Communication unit 120 Storage unit 121 History data storage unit 122 Model storage unit 123 Recommended value storage unit 130 Control unit 131 Acquisition unit 132 Learning unit 133 Update unit 134 Inference unit 135 Judgment unit 136 Stop unit 137 Display unit 138 Automatic operation control unit U User Da History data Db Driving data RD Recommended value SA Screen SB Screen SC1 Display SC2 Display SC3 Display SC4 Display 1000 Computer 1010 Memory 1011 ROM
1012 RAM
1020 CPU
1030 Hard disk drive interface 1040 Disk drive interface 1050 Serial port interface 1060 Video adapter 1070 Network interface 1080 Bus 1090 Hard disk drive 1091 OS
1092 Application program 1093 Program module 1094 Program data 1100 Disk drive 1110 Mouse 1120 Keyboard

Claims (8)

a determination unit that determines the operation of the automatic driving of the operation target when data regarding the accuracy of the automatic driving of the operation target satisfies a predetermined condition;
a stop unit that stops automatic operation of the operation target based on a determination result of the determination unit;
has
As the predetermined condition, the determination unit determines whether the automatic driving is performed when the variance of a plurality of recommended values inferred based on a plurality of learning models using historical data regarding driving of the operation target is not within a predetermined range. determine whether to stop,
An information processing device characterized by: The determination unit determines whether to stop the automatic driving when a predetermined change occurs in the explanatory variable, as the predetermined condition, based on the result of abnormality detection regarding the explanatory variable representing the driving situation of the operation target. do,
The information processing device according to claim 1, characterized in that: The determination unit determines, as the predetermined condition, that the automatic driving is stopped when a recommended value inferred based on a learning model using historical data regarding driving of the operation target is not included in a range of maximum and minimum threshold values. make a judgment,
The information processing device according to claim 1, characterized in that: As the predetermined condition, the determination unit may perform the automatic operation when a rate of change in the recommended value inferred based on a learning model using historical data regarding driving of the operation target is not included in a range of a difference change amount threshold. Judging whether to stop driving,
The information processing device according to claim 1, characterized in that: When the recommended value inferred based on a learning model using historical data regarding driving of the operation target has a posterior distribution, the determining unit determines, as the predetermined condition, that a predicted variance of the recommended value is within a predetermined range. determining whether to stop the automatic operation if the automatic operation does not occur;
The information processing device according to claim 1, characterized in that: The determination unit determines to stop the automatic operation when the evaluation index of the operation target is not included in a predetermined range as the predetermined condition.
The information processing device according to claim 1, characterized in that: An information processing method executed by an information processing device, the method comprising:
a determination step of determining the operation of the automatic driving of the operating target when data regarding the accuracy of the automatic driving of the operating target satisfies a predetermined condition;
a stopping step of stopping automatic operation of the operation target based on the determination result of the determining step ;
including;
In the determination step, if the predetermined condition is that the automatic driving is performed when the variance of a plurality of recommended values inferred based on a plurality of learning models using historical data regarding driving of the operation target is not included in a predetermined range. determine whether to stop,
An information processing method characterized by: a determination step of determining the operation of the automatic driving of the operating target when data regarding the accuracy of the automatic driving of the operating target satisfies a predetermined condition;
a stopping step of stopping automatic operation of the operation target based on the determination result of the determining step ;
make the computer run
The determining step includes determining, as the predetermined condition, that the automatic driving is performed when the variance of the plurality of recommended values inferred based on the plurality of learning models using historical data regarding the driving of the operation target is not within a predetermined range. determine whether to stop,
An information processing program characterized by:

Images