System, Information Processing Apparatus, Program, and Information Processing Method

Abstract

Provided is a system that optimizes the operation policy of a facility including facilities with power generation while considering the external environment of the facility and power consumption within the facility. **Solution means**: The system is provided with a prediction module that predicts future power generation amounts and power consumption amounts in a facility including facilities with power generation, and an optimization module that estimates an optimal operation policy of the facilities in the facility based on the predicted power generation amounts and power consumption amounts. The prediction module predicts the power generation amount based on the actual data of power generation and power consumption in the facility and the weather data regarding the location where the facility is installed.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to a system, an information processing apparatus, a program, and an information processing method.

Background Art

[0002] A maintenance management system for maintaining a solar power generation panel has been developed. Document 1 discloses a maintenance management device that diagnoses the safety or life of solar power generation facilities, accumulates information on the performance, procurement, or vendors of devices that control solar power generation panels, and provides diagnostic data, facility data, or procurement information to the server of a service provider.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] On the other hand, in the conventional maintenance management system, the solar panel and the maintenance management system are provided as a one-pack, and the power consumption in the facility equipped with the solar panel has not been considered. In recent years, solar panels may be installed on the roofs of houses, and in such a situation, there has been a need to optimize the operation policy for each facility while considering the external environment of the facility and the power consumption within the facility.

[0005] The problem to be solved by the present disclosure is to provide a system that optimizes the operation policy of equipment while considering the external environment of the facility and the power consumption inside the facility in a facility including equipment that generates electricity.

Means for Solving the Problems

[0006] As one aspect of the present disclosure, there is provided a system including a prediction module that predicts future power generation amounts and power consumption amounts in a facility including a facility accompanied by power generation, and an optimization module that estimates an optimal operation policy of the facility in the facility based on the predicted power generation amounts and power consumption amounts. The prediction module predicts the power generation amount based on actual data of power generation and power consumption in the facility and weather data regarding the location where the facility is installed.

Brief Description of the Drawings

[0007]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Figure 21

Figure 22

Figure 23

Figure 24

Figure 25

Figure 26

Embodiments for Carrying Out the Invention

[0008] <1. Embodiment> Hereinafter, the maintenance management system 1 as an embodiment of the present disclosure will be described with reference to the drawings. In the present specification and each figure, the same reference numerals are given to the same elements as those already described, and detailed descriptions will not be repeated. In the present disclosure, as an example of a user of the maintenance management system 1, a maintenance management user who performs maintenance management of facilities such as a solar panel using the maintenance management system 1 can be cited. The maintenance management user may be in a mode of maintaining and managing the company's own facilities, or may be in a mode of accepting the maintenance management of other companies' facilities. Further, as another example of a user of the maintenance management system 1, a system management user who manages the maintenance management system 1 can be cited. The maintenance management user and the system management user may be the same or different. The "user" in the following description may include both the maintenance management user and the system management user.

[0009] (1.1. Overall Configuration of Maintenance Management System 1) Referring to FIG. 1, the functional configuration of the maintenance management system 1 according to the embodiment will be described. The maintenance management system 1 is a system for performing maintenance management of facilities. As an example of the facility, a solar panel P can be cited. The solar panel P is installed in a solar power plant F as an example. Further, the solar panel P is installed on the roof of a house H as another example. The maintenance management system 1 detects a failure occurring in the solar panel P and estimates the cause of the failure. Further, the maintenance management system 1 grasps the actual power generation amount of the solar panel P and predicts the future power generation amount. When the solar panel P is installed in the house H, the maintenance management system 1 manages the power consumption of household electrical appliances (hereinafter referred to as home appliances) installed in the house H.

[0010] The conservative management system 1 includes an IoT device 100, a data management server 200, a data analysis server 300, and a terminal device 400. The IoT device 100, the data management server 200, the data analysis server 300, and the terminal device 400 are connected to a network N such as the Internet and are configured to communicate with each other.

[0011] The IoT device 100 is a device attached to a facility and performing data transmission and reception through the network N. The IoT device 100 includes a detection unit and a data holding unit. The detection unit detects information regarding the state of the facility or the state of the environment where the facility is installed. As an example, the detection unit detects power generation by the solar panel P and actual data of power consumption in the facility where the solar panel P is installed. The data holding unit is attached to the facility and holds data regarding the facility. In the present embodiment, the detection unit includes a current sensor 120 and a monitoring camera 130. Also, the data holding unit includes a tag 110. However, the detection unit and the data holding unit are not limited to these examples. Specifically, as the detection unit, a sensor that detects other physical quantities such as temperature, pressure, humidity, water pressure, and voltage may be used.

[0012] The tag 110 is, as an example, an RFID (Radio Frequency Identification) tag, and holds and / or transmits data using wireless communication. The tag 110 receives a radio wave (inductive signal) sent from the terminal device 400 operated by the user and returns the held information to the terminal device 400. The information to be returned includes information for accessing a virtual storage stored in the data management server 200.

[0013] The current sensor 120 is a sensor for measuring the current in a predetermined electric circuit and detects the magnitude and change of the current flowing through the electric circuit. The measured current value is sent to the data management server 200. The current sensor 120 may appropriately use known types such as a Hall effect sensor, a shunt resistance type, and a current transformer.

[0014] The monitoring camera 130 is installed around the facility and acquires imaging data. From the imaging data acquired by the monitoring camera 130, it is possible to detect damage caused by wild animals, theft, natural disasters, etc. The imaging data acquired by the monitoring camera 130 is transmitted to the data management server 200.

[0015] The data management server 200 is an information processing device that performs data management as an example. The hardware configuration of the information processing device will be described later. The data managed by the data management server 200 includes data related to the facility, specifically, data related to the state of the facility detected by the detection unit, and data related to the state of the environment in which the facility is installed detected by the detection unit. Details of the processing by the data management server 200 will be described later.

[0016] The data analysis server 300 estimates the content and cause of the failure that occurred in the solar panel P, as well as makes a future prediction regarding the operation of the facility. Specifically, the data analysis server 300 uses a learned model obtained by performing machine learning based on past failure performance data to estimate the content and cause of the failure that occurred in the solar panel P. In addition, the data analysis server 300 uses a learned model obtained by performing machine learning based on past power generation performance data to predict the power generation amount. Details of the processing by the data analysis server 300 will be described later.

[0017] The terminal device 400 is an information processing device equipped with input means and output means and operable by a user and an administrator. The terminal device 400 may be a general-purpose or dedicated personal computer, a smart device, etc. Note that smart devices include tablet terminals, smartphones, smart glasses, smart watches, etc. When the terminal device 400 is a tablet terminal or a smartphone, the terminal device 400 has a display that functions as input means and output means. The terminal device 400 accepts input of data by a touch operation on the user's display.

[0018] The conservative management system 1 may further acquire data from the population satellite S. The data from the population satellite S may be, for example, data related to weather, or data related to satellite photos of the solar panel P. These data may also be stored in the data management server 200 according to the specifications.

[0019] (1.2. Data management server 200) (1.2.1. Functional configuration of the data management server 200) Referring to FIG. 2, the functional configuration of the data management server 200 will be described. The data management server 200 includes a control unit 210 and a storage unit 220. The control unit 210 controls the processing of the data management server 200 as a processor.

[0020] The storage unit 220 stores programs and data necessary for the processing of the control unit 210, data obtained by the processing of the control unit 210, and the like. As an example, the storage unit 220 stores a virtual storage 230.

[0021] The virtual storage 230 stores an IoT data management table 230a. The IoT data management table 230a includes, as an example, an item "management number", an item "T1 management number", an item "T1 address", an item "T2 management number", an item "T2 address", an item "T3 management number", and an item "T3 address".

[0022] The item "management number" is data for uniquely identifying a record in the IoT data management table 230a and is the primary key item of the IoT data management table 230a. The item "T1 management number" is a foreign key item for referring to the primary key item "management number" in the IoT data table T1 stored in the data center S1. The item "T1 address" is address information for accessing the IoT data table T1 stored in the data center S1.

[0023] The item "T2 Management Number" is a foreign key item for referring to the primary key item "Management Number" in the IoT data table T2 stored in the data center S2. The item "T2 Address" is address information for accessing the IoT data table T2 stored in the data center S2.

[0024] The item "T3 Management Number" is a foreign key item for referring to the primary key item "Management Number" in the IoT data table T3 stored in the information processing device within the factory PL. The item "T3 Address" is address information for accessing the IoT data table T3 stored in the information processing device within the factory PL.

[0025] In this way, in the maintenance management system 1, the information detected by the current sensor 120 and the monitoring camera 130 as detection units, as well as the information held by the tag 110 as a data holding unit, are centrally managed in the virtual storage. That is, the actual data is stored in the data center S1, the data center S2, or the information processing device within the factory PL, and the data management server 200 constitutes the virtual storage 230 by storing the address information and key information for accessing those external data. With such a specification, even when the information detected by the current sensor 120 and the monitoring camera 130 as detection units, as well as the information held by the tag 110 as a data holding unit, are stored in various data sensors and information processing devices, they are centrally managed in the virtual storage 230 stored in the data management server 200, so that the user can easily obtain the data necessary for the maintenance management of existing facilities.

[0026] (1.2.2. Screen Example in Data Management) Hereinafter, with reference to FIGS. 3 to 7, an example of a screen displayed on the terminal device 400 for data management will be described. FIG. 3A is a diagram showing a management screen 410 displayed on the display of the terminal device 400. As shown in FIG. 3A, the management screen 410 has a menu list 411 and an organization page 412. The menu list 411 displays a list of screens that can be transitioned from the management screen 410. The organization page 412 is the home screen of the management screen 410. User information 413 registered in the organization is displayed on the organization page 412. When the user selects the facility list icon 411a in the menu list 411, a facility list screen 420 is displayed.

[0027] FIG. 3B is a diagram showing a facility list screen 420 displayed on the display of the terminal device 400. As shown in FIG. 3B, the facility list screen 420 has a facility list 421, a facility details button 422, and a usage edit button 423. The facility list 421 displays a list of facilities with tag 110 attached. The facility details button 422 is a button for displaying detailed information of the facility. When the user presses the facility details button 422, a facility details screen 430 is displayed. The usage edit button 423 is a button for editing the usage of tag 110 (what information is associated with tag 110). When the user presses the usage edit button 423, a usage management screen 450 is displayed.

[0028] FIG. 4A is a diagram showing a facility details screen 430 displayed on the display of the terminal device 400. As shown in FIG. 4A, the facility details screen 430 has facility details information 431 and a facility details edit button 432. The facility details information 431 is information regarding the details of the facility. The facility details edit button 432 is a button for editing the details of the facility. When the user presses the facility details edit button 432, a facility details edit screen 440 is displayed.

[0029] FIG. 4B is a diagram showing an equipment details editing screen 440 displayed on the display of the terminal device 400. As shown in FIG. 4B, the equipment details editing screen 440 has a display setting list 441. The display setting list 441 is a list regarding the setting of whether to display on the management screen as information of the equipment. The display setting list 441 includes a public setting checkbox 442. When the user checks the public setting checkbox 442, the information of the equipment is published on the network N, and users with authorized permissions can access and view it from the outside. On the other hand, when the user unchecks the public setting checkbox 442, the user cannot access and view it from the outside. Thus, in the maintenance management system 1, for each piece of equipment to which the tag 110 is attached, it is possible to set whether to publish the information of the equipment on the network N.

[0030] FIG. 5A is a diagram showing an application management screen 450 displayed on the display of the terminal device 400. As shown in FIG. 5A, the application management screen 450 has an application list 451. The application list 451 is a list regarding the applications of the tag 110. In the maintenance management system 1, multiple applications of what information to associate with the tag 110 can be set. The application list 451 includes a details button 452. When the user presses the details button 452, an application details screen 460 is displayed.

[0031] FIG. 5B is a diagram showing an application details screen 460 displayed on the display of the terminal device 400. As shown in FIG. 5B, the application details screen 460 has an application item list 461. The application item list 461 is a list of data items set for each application. The application item list 461 has an edit button 462. When the user presses the edit button 462, a details editing screen 470 is displayed.

[0032] FIG. 5C is a diagram showing a detailed editing screen 470 displayed on the display of the terminal device 400. As shown in FIG. 5C, the detailed editing screen 470 has item detailed information 471. The item detailed information 471 is information regarding details of data items set for each use. The item detailed information 471 includes a guest non-display check box 472. When the user checks the guest non-display check box 472, the item becomes non-displayed for guest users with limited access rights (for example, persons in charge of construction companies that temporarily use the maintenance management system 1). On the other hand, when the user unchecks the guest non-display check box 472, the item is displayed for guest users. Thus, in the maintenance management system 1, it is possible to set whether to display for guest users for each data item associated with the tag 110.

[0033] FIG. 6A is a diagram showing an administrative screen 410 displayed on the display of the terminal device 400. As shown in FIG. 6A, the administrative screen 410 has an organization information pull-down list 414. The organization information pull-down list 414 is a list to be selected when performing management, switching, addition, etc. of organization information. The organization information pull-down list 414 includes an organization management icon 414a. When the user presses the organization management icon 414a, an organization management screen 480 is displayed.

[0034] FIG. 6B is a diagram showing an organization management screen 480 displayed on the display of the terminal device 400. As shown in FIG. 6B, the organization management screen 480 has a user list 481. The user list 481 is a list of users belonging to the organization. The user list 481 has a detail button 482. When the user presses the detail button 482, a user information detail screen 490 is displayed.

[0035] FIG. 7A is a diagram showing a user information details screen 490 displayed on the display of the terminal device 400. As shown in FIG. 7A, the user information details screen 490 includes a user information details list 491 and a user information edit button 492. The user information details list 491 is a list of detailed information about the user. The user information edit button 492 is a button for editing user information. When the user presses the user information edit button 492, a user information edit screen 495 is displayed.

[0036] FIG. 7B is a diagram showing a user information edit screen 495 displayed on the display of the terminal device 400. As shown in FIG. 7B, the user information details list 491 in the user information edit screen 495 includes an inspection registration permission checkbox 496. When the user checks the inspection registration permission checkbox 496, regardless of the user's access authority (i.e., even if it is a guest user as shown in FIG. 7B), it is possible to register and update information after performing equipment inspection. On the other hand, when the user unchecks the inspection registration permission checkbox 496, the user cannot register and update information after performing equipment inspection. Thus, in the maintenance management system 1, for each user, the authority to register and update information related to equipment can be granted.

[0037] (1.2.3. Data access at tag 110) With reference to FIGS. 8 and 9, the processing flow of data access using tag 110 will be described. As shown in FIG. 8, in step S210, the terminal device 400 transmits a guiding signal to tag 110. In step S110, tag 110 receives the guiding signal transmitted from the terminal device 400.

[0038] In step S115, based on the received guidance signal, tag 110 transmits access information (for example, the address of virtual storage 230 stored in data management server 200) for accessing data management server 200 to terminal device 400. In step S215, terminal device 400 receives the access information transmitted from tag 110.

[0039] In step S220, based on the access information received from tag 110, terminal device 400 transmits an access request to data management server 200. In step S310, data management server 200 receives the access request from terminal device 400.

[0040] In step S315, data management server 200 transmits a password to terminal device 400. From the perspective of enhancing security level, it is preferable that the password is a one-time password valid for only a certain period. In step S225, terminal device 400 receives the password from data management server 200.

[0041] In step S230, terminal device 400 performs an encryption process. As an example of the encryption process, terminal device 400 generates a hash value from the user ID it holds and the password received from data management server 200. Terminal device 400 further encrypts the hash value based on the position information obtained from GPS (Global Positioning System) or LTE (Long Term Evolution). As the encryption algorithm, for example, a known post-quantum computer encryption algorithm such as lattice-based encryption or hash-based encryption may be adopted. As an example, the fluctuation value (decimal value) of the GPS value of terminal device 400 may be used for lattice-based encryption. Note that the position information is not limited to the above example, and for example, a MAC address, altitude, etc. may be used, or something that changes slightly over time like radio wave intensity or temperature but does not change significantly may be adopted.

[0042] In step S235, the terminal device 400 transmits the generated ciphertext to the data management server 200. In step S320, the data management server 200 receives the ciphertext from the terminal device 400.

[0043] As shown in FIG. 9, in step S330, the data management server 200 performs decryption processing on the received ciphertext. As the decryption processing, for example, a hash value generated by a user ID and a password is confirmed, decryption of the ciphertext is performed from the geographical information of the registered tag 110, and authentication of the decrypted information may be performed using the location information obtained from the access IP (Internet Protocol Address) address of the terminal device 400. By adopting such a processing flow of encryption processing, decryption processing, and authentication using the terminal device 400, hacking (illegal access) from a region different from the facility to which the tag 110 is attached can be prevented.

[0044] In step S335, the data management server 200 transmits the data stored in the virtual storage 230 to the terminal device 400. In step S240, the terminal device 400 receives the data transmitted from the data management server 200. As a result, the user operating the terminal device 400 can acquire the data in the virtual storage 230 stored in the data management server 200.

[0045] (1.3. Data Analysis Server 300) (1.3.1. Functional Configuration of Data Analysis Server 300) Referring to FIG. 10, the functional configuration of the data analysis server 300 will be described. The data analysis server 300 includes a control unit 310 and a storage unit 350. The control unit 310 controls the processing of the data analysis server 300 as a processor. The control unit 310 includes a failure detection module 320, a prediction module 330, and an optimization module 340.

[0046] The failure detection module 320 detects a failure that has occurred in the facility and estimates the cause of the failure based on data and the like detected by the detection unit. Details of the processing by the failure detection module 320 will be described later.

[0047] The prediction module 330 predicts the future power generation amount and power consumption amount in a facility including facilities accompanied by power generation. As an example, the prediction module 330 predicts the power generation amount based on the actual data of power generation and power consumption in the facility and the meteorological data regarding the location where the facility is installed. As an example, when the facility is the solar panel P, the prediction module 330 predicts the future power generation amount by the solar panel P, the power consumption amount in the facility where the solar panel P is installed, and the like. Details of the processing by the prediction module 330 will be described later.

[0048] The optimization module 340 estimates the optimal operation policy of the facility based on the power generation amount and power consumption amount predicted by the prediction module 330. As an example, the optimization module 340 proposes an optimal operation policy regarding the facility in the future facility based on data and the like detected by the detection unit. Details of the processing by the optimization module 340 will be described later.

[0049] The storage unit 350 stores programs and data necessary for the processing of the control unit 310, data obtained by the processing of the control unit 310, and the like. As an example, the storage unit 350 includes IoT past data 355, failure data 360, learning data 365, failure mode correct data 370, meteorological data 375, a failure detection model 380, a prediction model 390, and an optimization model 395.

[0050] The IoT past data 355 is historical data obtained from the current sensor 120 and the monitoring camera 130 as the detection unit. The failure data 360 is the historical data of the IoT past data 355 when a failure occurred in the facility in the past. The learning data 365 is data detected in a general failure that does not depend on individual facilities or the installed area. Specifically, it may include surges caused by lightning strikes, output waveform data of a power conditioner when a breaker is tripped, and the like. The correct failure mode data 370 is the correct data regarding the correspondence relationship (failure mode) between the cause of the failure and the event of the failure.

[0051] The weather data 375 is data regarding the weather of the environment where the facility is installed. The weather data 375 may be, for example, data obtained from the Japan Meteorological Agency, a meteorological business support center, or a private operator. Also, in the obtained data, when there is no weather data corresponding to the installation location of the facility, the weather data may be generated in consideration of the influence of the wind direction calculated from the surrounding weather information and / or the influence of mountains around the installation location of the facility.

[0052] The failure detection model 380 is a trained model that is used when machine learning is executed by the failure detection module 320 to estimate failure detection. The prediction model 390 is a trained model that is used when machine learning is executed by the prediction module 330 to estimate failure detection. The optimization model 395 is a data model that is utilized when the processing by the optimization module 340 is performed.

[0053] (1.3.2. Details of the failure detection model 380) The failure detection model 380 includes an anomaly detection model 381, a waveform classification model 382, a failure classification model 383, a failure mode model 384, an explanatory variable model 385, and a failure detection model 386.

[0054] The anomaly detection model 381 is a trained model that has learned whether the data acquired by the detection unit is abnormal. The waveform classification model 382 is a trained model that has learned the correspondence between the waveform of the data acquired by the detection unit and the device that output the data. The fault classification model 383 is a trained model that has learned the correspondence between the waveform of the data acquired by the detection unit and the classification of the fault (i.e., the fault event).

[0055] The fault mode model 384 is a trained model that has learned the correspondence (fault mode) between the cause of the fault and the fault event. The explanatory variable model 385 is a trained model that has learned the correspondence between the data acquired by the detection unit at the time of fault occurrence and the explanatory variables with a high correlation. The fault detection model 386 is a trained model that has learned the correspondence between the data acquired by the detection unit at the time of fault occurrence and the cause of the fault.

[0056] (1.3.3. Processing of the Fault Detection Module 320) Referring to FIG. 12, the processing flow of the fault detection module 320 will be described. In step S410, pre-training processing is performed. In the pre-training processing, · Anomaly detection learning · Waveform classification learning · Fault classification learning · Fault mode learning · Explanatory variable learning · Fault detection learning such machine learning processing is performed. The following will be described in order. In the following learning processing, known machine learning algorithms can be used. Specific algorithms include, for example, linear regression, logistic regression, support vector machine, k-nearest neighbor method, decision tree, random forest, neural network, naive Bayes, etc. These algorithms can be appropriately selected according to the content of learning, data structure, calculation resources, etc. Also, the pre-training processing may be specified to be executed at a predetermined interval (for example, once a month).

[0057] As shown in FIG. 13, in anomaly detection learning, using the IoT past data 355 as input data, data with a high occurrence frequency is regarded as normal data, and data with a low occurrence frequency is regarded as abnormal data. Then, the classification (class division) between normal data and abnormal data is learned to generate an anomaly detection model 381. Note that, for the input data, after performing the range detection process described later, the determination of the occurrence frequency may be performed.

[0058] As shown in FIG. 14, in waveform classification learning, using the IoT past data 355 and the learning data 365 as input data, the correspondence between the waveform of the data included in the IoT past data 355 and the device that output the data is learned to generate a waveform classification model 382.

[0059] As shown in FIG. 15, in failure classification learning, using the IoT past data 355, the learning data 365, and the failure mode correct answer data 370 as input data, for the failure data included in the failure mode correct answer data 370, the correspondence between the failure event and the waveform of the failure data is learned to generate a failure classification model 383.

[0060] As shown in FIG. 16, in failure mode learning, first, normalization processing is performed on the failure data 360. Since the failure data 360 includes data related to failures registered by the person in charge of the construction company during the inspection work, the written expression of the failure content may differ depending on the person in charge. In the normalization process, using an LLM (Large Language Model / large-scale language model) or the like, the text of the failure content included in the failure data 360 is digitized. For those with a high similarity to the content of the existing failure list, corrections are made to match the content of the failure list. For those with a low similarity to the content of the existing failure list, a new failure content is named and added to the failure list. In this way, by normalizing the description of the failure content, the noise of the input data in the subsequent learning process can be reduced. The normalized data is stored in the failure mode correct answer data 370.

[0061] Next, using the failure mode correct data 370 as input data, learn the failure mode, which is the correspondence between the cause of the failure and the event of the failure. As an example, in the case of a failure of a solar panel, learn the correspondence between the cause of the failure and the possible events that may occur, such as lightning strike + silicon burnout, falling rock + glass breakage, wild animal damage + cable disconnection, etc., and generate a failure mode model 384.

[0062] As shown in FIG. 17, in explanatory variable learning, first, use the IoT past data 355 and the meteorological data 375 as input variable data, and the failure mode correct data 370 as target variable data to perform input variable extraction. In input variable extraction, only the data with a high correlation value with the failure mode correct data 370 is extracted as the input variable 396 from various data included in the IoT past data 355 and the meteorological data 375.

[0063] In calculating the correlation value in input variable extraction, for example, a correlation matrix may be calculated and a DNN (Deep Neural Network) model may be used. Specifically, for example, calculate the correlation coefficient between the input variable data and the target variable data and create a correlation matrix. Here, by normalizing the value of the correlation coefficient (for example, in the range of 0 to 1), the influence degrees of the explanatory variables can be compared on a unified scale. Furthermore, the explanatory variables with a correlation coefficient below a predetermined threshold may be treated as a fixed value 0 in the DNN input layer, and the corresponding nodes may be deleted to reduce the subsequent learning process. In this way, by reducing the dimensionality of the explanatory variables, learning can be focused on important variables, and the learning speed of the DNN model can be improved and the calculation cost can be reduced. Then, perform performance evaluation of the DNN model and adjust the threshold and parameters as necessary. As a result, the input variable 396 is extracted.

[0064] After that, using the extracted input variable 396 and the failure mode correct data 370, perform explanatory variable learning. That is, learn the correlation between the input variable 396 and the failure mode correct data 370, and generate an explanatory variable model 385.

[0065] As shown in FIG. 18, in the failure detection learning, using the IoT past data 355, the failure mode correct data 370, and the weather data 375 as input data, the correspondence between the data waveform of the IoT past data 355 and the failure mode is learned, and the failure detection model 386 is generated. Also, using the failure mode correlation data 397 indicating the correlation values between the failure modes, the multidimensional distance of the IoT past data 355 is calculated. Here, the failure mode correlation data 397 is data regarding the correlation relationship between the input variable 396 and the failure mode correct data 370.

[0066] Returning to FIG. 12, the description continues. In step S420, preprocessing is performed on the data included in the input IoT past data 355. The preprocessing includes · Range detection · Waveform classification inference is performed. Note that the IoT past data 355 may be acquired at a predetermined interval (for example, every few minutes), and the preprocessing (S420) and subsequent inference processing (S430) may be implemented.

[0067] In range detection, for the waveform of the target data, it is determined which range is the waveform delimiter (that is, the waveform period, or the repetition of the period). By performing such processing, it becomes clear which period of data should be the analysis target, and the accuracy of the subsequent inference processing can be improved.

[0068] In waveform classification inference, for the waveform of the target data, it is classified which device in the facility output the waveform. The waveform classification is executed using the waveform classification model 382.

[0069] In step S430, inference processing is performed. In the inference processing, · Abnormality detection inference · Failure classification inference · Failure detection inference · Failure mode inference is performed. Hereinafter, the description will be made with reference to FIG. 19.

[0070] As shown in FIG. 19, the inference process (S430) may include a case where the pre-processed IoT past data 355 is used as input data and a case where the failure data 360 is input. In the case where the pre-processed IoT past data 355 is used as input data, first, an anomaly detection inference is performed.

[0071] In the anomaly detection inference, it is determined whether there is an anomaly in the pre-processed IoT past data 355 using the anomaly detection model 381. If it is determined that there is an anomaly, explanatory variable filtering is performed on the IoT past data 355.

[0072] In the explanatory variable filtering, data corresponding to the input variable 396 (see FIG. 17) having a high correlation with the failure mode correct data 370 is extracted from the pre-processed IoT past data 355 using the explanatory variable model 385. After the explanatory variable filtering is performed on the IoT past data 355 in this way, the failure classification inference and the failure detection inference are executed.

[0073] In the case where the failure data 360 is used as input data, data corresponding to the input variable 396 having a high correlation with the failure mode correct data 370 is extracted from the failure data 360. After the explanatory variable filtering is performed on the IoT past data 355 in this way, the failure classification inference and the failure detection inference are executed.

[0074] In the failure classification inference, the classification of the failure (i.e., the event of the failure) is estimated using the failure classification model 383 for the IoT past data 355 or the failure data 360 on which the explanatory variable filtering has been performed. The classification of the failure is estimated together with the probabilities corresponding to multiple failure events, such as a 70% probability of "cable disconnection" and a 30% probability of "glass breakage".

[0075] In the failure detection inference, the cause of the failure is estimated using the failure detection model 386 for the IoT past data 355 or the failure data 360 on which the explanatory variable filtering has been performed. The cause of the failure is, for example, a 60% probability of "lightning strike", " EarthquakeThe probability of "」 is 40 percent, and so on. Along with the probabilities corresponding to the causes of multiple failures, they are estimated. When the failure events and causes are estimated, failure mode inference is executed.

[0076] In failure mode inference, for the IoT past data 355 or failure data 360 in which multiple failure events and causes are estimated, an estimation of the failure mode, which is a combination of the failure event and cause, is performed. In the estimation of the failure mode, based on the estimated failure classification and failure cause, the failure mode is estimated according to the following procedure. 1. Calculate the probability of all combinations of failure events and causes (failure mode probability) 2. Identify the maximum probability Pm, which is the highest failure mode probability among those calculated 3. Determine whether it is registered in the failure mode correct data 370 4. Identify the registered maximum probability P, which is the highest failure mode probability among those registered 5. Calculate the number of occurrences of events and causes 6. Calculate the reliability

[0077] Hereinafter, an explanation will be given based on a specific example shown in FIG. 20. In the example shown in FIG. 20, by the previous failure classification inference, the probabilities of four failure events A to D are calculated respectively. Also, by the previous failure detection inference, the probabilities of three failure causes a to h are calculated respectively. First, as step 1, for all combinations of events A to D and causes a to h (that is, 12 combinations of Aa, Ab,..., Dh), the failure mode probability is calculated. Next, as step 2, among those calculated, the maximum probability Pm, which is the highest failure mode probability, is identified. In the example shown in FIG. 20, as Pm, the failure mode probability 0.56 of the combination of Aa is identified.

[0078] Next, as step 3, determine whether it is registered as a failure mode in the failure mode correct data 370. In the example shown in FIG. 20, it is determined that "A-i", "C-i", "C-ha", and "D-i" are not registered as failure modes. Next, as step 4, among the registered ones, specify the registered maximum probability P that is the highest failure mode probability. In the example shown in FIG. 20, as P, the failure mode probability 0.42 of the combination of B-i is specified.

[0079] Next, as step 5, calculate the number of occurrences of each event and cause. The number of occurrences here refers to, in the example shown in FIG. 20, the number of each event and cause after subtracting the failure modes not registered in the failure mode correct data 370 from all combinations of events and causes. Specifically, as shown in FIG. 20, the number of occurrences of event A is 2 because the failure mode "A-i" is not registered. The number of occurrences of event B is 3. The number of occurrences of event C is 1 because the failure modes "C-i" and "C-ha" are not registered. The number of occurrences of event D is 2 because the failure mode "D-i" is not registered. Similarly, the number of occurrences is calculated for causes i to ha.

[0080] Next, as step 6, calculate the reliability R. The reliability is an index indicating the reliability considering the possibility of unknown failures for the estimated failure mode. As an example, the reliability R is calculated by the following formula. R = P - Pd + Pc Pd (reliability reduction term): maximum probability Pm - registered maximum probability P Pc (reliability increase term): number of occurrences of the event corresponding to the registered maximum probability P / number of occurrences of all events

[0081] In this way, the larger the maximum probability Pm is compared to the registered maximum probability P, the higher the possibility of unknown failures, and the reliability decreases. Also, the larger the number of occurrences of the event corresponding to the registered maximum probability P, the higher the possibility of that event, and the reliability decreases. By calculating such a reliability R, it is possible to accurately estimate the combination of the event and cause of the failure in consideration of the occurrence of unknown failures.

[0082] In the example shown in FIG. 20, the reliability R is calculated as the following value. P = 0.42 Pm = 0.56 Pd = 0.14 Pc = 3 / 8 = 0.375 R = 0.42 - 0.14 + 0.375 = 0.655

[0083] Note that the specific calculation formulas for the reliability decrease term Pd and the reliability increase term Pc are not limited to the above example. For example, the reliability R may be calculated using the above calculation formula after multiplying at least one of the reliability decrease term Pd or the reliability increase term Pc by a predetermined coefficient. Also, as the reliability increase term Pc, the occurrence frequency of the cause may be considered in addition to (or instead of) the occurrence frequency of the event.

[0084] The failure detection module 320 causes the terminal device to display a plurality of failure modes (i.e., combinations of failure events and causes) with a high failure mode probability. In this case, it may be a specification to also display the reliability R together with the probability of the failure mode. Here, regarding the reliability R, the calculated value of the reliability R may be compared with a predetermined reference value and displayed with labels such as "high" reliability, "medium" reliability, and "low" reliability.

[0085] (1.3.4. Processing of Prediction Module 330) With reference to FIGS. 21 to 25, the processing flows of the prediction module 330 and the optimization module 340 will be described. As shown in FIG. 21, in step S510, pre-learning processing is performed. In the pre-learning processing, · Explanatory variable learning · Prediction learning such machine learning processing is performed. The following will explain in order.

[0086] As shown in FIG. 22, in the explanatory variable learning, the IoT past data 355 and the meteorological data 375 are used as input variable data, and the IoT past data 355 is used as target variable data to perform input variable extraction. The IoT past data 355 in the input variable data includes the actual power consumption of the facility. The IoT past data 355 in the target variable data includes the actual power generation amount.

[0087] After that, using the extracted input variables 396 and the IoT past data 355, explanatory variable learning is performed. That is, the correlation between the input variables 396 and the IoT past data 355 is learned to generate an explanatory variable model 385.

[0088] As shown in FIG. 23, in the prediction learning, using the IoT past data 355 and the meteorological data 375 as input data, machine learning is performed to generate a prediction model 390. The IoT past data 355 may include the power generation amount of the solar panel P, the power consumption, the facility information, the inspection information, the repair information, etc. For the machine learning, the algorithm used in the above-described failure detection model 380 may be adopted.

[0089] In addition, in the learning of the prediction model 390, the above-described explanatory variable model 385 may be used. Specifically, instead of the failure mode correct data 370, using the data related to the power generation amount of the solar panel P included in the IoT past data 355, input variable extraction is performed with the data having a high correlation with the power generation amount as input variables from the IoT past data and the meteorological data. After that, prediction learning may be executed with the extracted input variables and the prediction variables including the data related to the power generation amount of the solar panel P included in the IoT past data 355.

[0090] Next, as step S520, predictive inference is performed. In the example shown in FIG. 24, as an example of predictive inference, after performing explanatory variable filtering using IoT past data 355 and weather data 375 as input data, the power generation amount of the solar panel P is estimated using the prediction model 390. Note that the predictive inference may be performed at a predetermined frequency (for example, once a day). Further, correction considering the aging deterioration of the solar panel P may be performed on the estimated power generation amount. Further, the parameters of the prediction model 390 that improve the prediction accuracy may be appropriately adjusted by comparing the estimated predicted amount with the actual value of the power generation amount updated daily.

[0091] (1.3.5. Processing of the optimization module 340) With reference to FIG. 25, the optimization process by the optimization module 340 will be described. The optimization module 340 estimates an optimal operation policy regarding the power of the facilities in the facility including the solar panel P. As an example, in the example shown in FIG. 21, in the house H where the solar panel P is installed, an optimal operation policy based on the charging amount in the rechargeable battery B and the power consumption amount in the home appliance E is estimated.

[0092] (Optimization during normal times) First, the optimization regarding the power during normal times will be described. In this example, the rechargeable battery B can be charged by the solar panel P and by an outlet. In this case, · Input power amount from the solar panel P: P1(t) · Input power amount from the outlet: P2(t) · Output power amount consumed by home appliances: P3(t) · Self-discharge amount: P4 · Charging amount: P5(t) · Waste power amount: P6(t) · Power consumption amount: P7(t) · Battery remaining amount: Pbm(t) An optimization problem of reducing the electricity bill (that is, reducing the input power amount P2) occurs while adjusting these eight parameters. Note that it is assumed that the self-discharge is constant regardless of the time t.

[0093] The optimization module 340 calculates an approximate solution by quantum annealing for the above optimization problem. Specifically, first, Set the Hamiltonian Hmin of the objective function JPEG0007692144000002.jpg12147. Here, Hcom is the Hamiltonian of the constraint condition. Also, for the weighting coefficients λ1 and λ2, the condition λ1 < λ2 is imposed so that the input power P2(t) from the outlet is less than the waste power P6(t). Also, t = 0 and n are arbitrary times. As an example, t = 0 may be 24 hours before the present, and t = n may be 24 hours after the present.

[0094] The constraint conditions may be, for example, as follows. JPEG0007692144000003.jpg4147 Here, He: Condition regarding energy balance Hbat: Condition regarding update of battery remaining amount Hre: Condition regarding non-negativity of battery remaining amount Hch: Condition regarding prohibition of simultaneous charging and discharging Had: Condition regarding priority of home appliances consuming power That is.

[0095] Specifically, since the input power to the rechargeable battery B and the input power from the rechargeable battery B are equal, P1(t) + P2(t) + Pbm(t) = P3(t) + P4(t) + P5(t) holds. Therefore, JPEG0007692144000004.jpg12147 is obtained.

[0096] Also, focusing on the remaining amount of the rechargeable battery B, Pbm(t + 1) = Pbm(t) + P5(t) - P7(t) - P4 holds. Therefore, JPEG0007692144000005.jpg12147 is obtained.

[0097] Also, since the remaining amount of the rechargeable battery B does not become negative, Pbm(t) ≥ 0 holds. Therefore, It becomes 12147 for JPEG0007692144000006.jpg.

[0098] Also, since charging the rechargeable battery B and discharging from the rechargeable battery B cannot be performed simultaneously, P5(t) × P7(t) = 0 holds. Therefore, It becomes 12147 for JPEG0007692144000007.jpg.

[0099] Also, considering the priority of power consumption in each household appliance, It becomes 15147 for JPEG0007692144000008.jpg. Here, δ(i) is an adjustment coefficient assigned to each of the plurality (1 to m) of household appliances. The higher the priority of the household appliance, the larger the adjustment coefficient δ(i). Note that during normal times, the adjustment coefficient may be constant. Also, for the weighting coefficient λ3, a condition of λ3 < λ2 is imposed so that the output power amount P3(t) consumed by the household appliance is smaller than the waste power amount P6(t).

[0100] In this way, the optimization module 340 can set the Hamiltonian Hcom of the constraint conditions, calculate an approximate solution of the objective function Hmin by quantum annealing, and estimate Pn(t) (where n = 1 to 7). In this way, in the present embodiment, in a facility including facilities with power generation, it is possible to estimate the optimal operation policy for each facility while considering the external environment of the facility and the power consumption inside.

[0101] Each set Hamiltonian is stored in the optimization model 395. For the execution of quantum annealing, existing quantum computing services such as D-Wave Systems may be used. Also, the optimization module 340 may be implemented on a gate-type quantum computer that is not a quantum annealing, or an approximate solution of the above optimization problem may be calculated by a simulated annealing method without using a quantum computer. Further, in the above optimization, an approximate solution is calculated to reduce the input power amount P2 from the outlet, but a negative value may be allowed as the input power amount P2. By doing so, the technical idea of the present disclosure can be applied even in the case of selling the generated electricity to the outside.

[0102] (Optimization during disasters) Assume a case where power supply stops 6 hours later due to a planned power outage or the like as a disaster. In this case, the constraint condition Hcom is It may be JPEG0007692144000009.jpg21147. Here, with to = 6, P2(t) = 0 when t ≥ to. By setting the Hamiltonian Hcom of such a constraint condition, it is possible to estimate the operation policy of the facilities in the facility regarding the optimal power for each facility when the power supply stops during a disaster or the like.

[0103] Note that in the event of a disaster, in Equation (7), the adjustment coefficient δ of the home appliances is adjusted according to the priority. Specifically, for example, the adjustment coefficient δ(i) of home appliances with high priority during a disaster, such as a refrigerator or a communication device, is set to δ(i)>0, and the adjustment coefficient δ(i) of other home appliances is set to 0. By doing so, considering the priority of home appliances, the power consumption in the facility can be set, and the operation policy of the facility in the future can be determined. Note that in the adjustment of the adjustment coefficient δ, explanatory variable learning may be used. That is, by learning the correlation between the actual values of a plurality of output power amounts P3 and the actual value of the input power amount P2, the home appliances that contribute greatly to the input from the outlet are specified, and the priority of the home appliances may be increased (that is, the adjustment coefficient δ is increased).

[0104] (1.4. Hardware Configuration of Information Processing Apparatus) Referring to FIG. 26, the hardware configuration of an information processing apparatus used as a data management server 200, a data analysis server 300, a terminal device 400, etc. will be described. As an example, the information processing apparatus is realized by a computer 90 shown in FIG. 26. The computer 90 may include a CPU 91, a ROM 92, a RAM 93, a storage 94, an input interface 95, an output interface 96, and a communication interface 97.

[0105] The CPU 91 functions as a processor that executes processing. Specifically, the CPU 91 uses the RAM 93 as a work memory and executes a program stored in at least one of the ROM 92 or the storage 94. During the execution of the program, the CPU 91 controls each component via the system bus 98 and executes various processes. As an example, the CPU 91 functions as the control unit 210 or the control unit 310.

[0106] The ROM 92 stores a program for controlling the operation of the computer 90. The ROM 92 stores programs necessary for causing the computer 90 to realize the above-described various processes. The RAM 93 functions as a storage area where the programs stored in the ROM 92 are expanded.

[0107] The storage 94 stores data necessary for program execution and data obtained by program execution. The storage 94 includes one or more selected from a Hard Disk Drive (HDD) and a Solid State Drive (SSD). As an example, the storage 94 functions as the storage unit 220.

[0108] The input interface (I / F) 95 can connect the computer 90 and an input device 95a. The input interface 95 is, for example, a serial bus interface such as USB. The CPU 91 can read various data from the input device 95a via the input interface 95.

[0109] The output interface (I / F) 96 can connect the computer 90 and the output device 96a. The output interface 96 is, for example, a video output interface such as a Digital Visual Interface (DVI) or a High-Definition Multimedia Interface (HDMI (registered trademark)). The CPU 91 can transmit data to the output device 96a via the output interface 96 and output the data to the output device 96a.

[0110] The input device 95a is an example of an input means and includes one or more selected from a mouse, a keyboard, a microphone (voice input), and a touch pad. The output device 96a is an example of an output means and includes one or more selected from a display, a projector, a printer, and a speaker. A device having both functions of the input device 95a and the output device 96a, such as a touch panel, may be used.

[0111] The communication interface (I / F) 97 can connect the computer 90 and an external server 97a outside the computer 90. The communication interface 97 is, for example, a network card such as a LAN card. The CPU 91 can read various data from the external server 97a via the communication interface 97.

[0112] Note that each process executed by the data management server 200, the data analysis server 300, or the terminal device 400 may be realized by one computer 90 or may be realized by the cooperation of a plurality of computers 90.

[0113] The processing of the various data described above can be recorded as a program executable by a computer on a magnetic disk (such as a flexible disk and a hard disk), an optical disk (such as a CD-ROM, CD-R, CD-RW, DVD-ROM, DVD±R, DVD±RW), a semiconductor memory, or other non-transitory computer-readable storage media.

[0114] For example, the information recorded on the recording medium can be read by a computer (or an embedded system). In the recording medium, the recording format (storage format) is arbitrary. For example, the computer reads a program from the recording medium and causes the processor to execute the instructions described in the program based on this program. In the computer, the acquisition (or reading) of the program may be performed through a network.

[0115] <2. Other Embodiments> As described above, the maintenance management system 1 according to this embodiment has been described. However, the application of the technical idea of the present disclosure is not limited to the above embodiment. For example, satellite photos obtained from the geostationary satellite S may be used for fault detection. Specifically, for example, images from a predetermined time ago (for example, 30 minutes ago) and the current image are acquired from the satellite image, and by performing differential analysis and moving object analysis, changes and moving objects in the image are extracted. Further, in combination with time and GPS information, a database of similar images and fault cases taken in the past is referred to, and an abnormality is detected by comparing with the current situation. For example, a person who intrudes for the purpose of theft at night shows a different time and movement pattern from a repair worker, so such specific features can be detected.

[0116] Although several embodiments of the present disclosure have been illustrated above, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, changes, etc. can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalent scope thereof. In addition, the above-described embodiments can be implemented in combination with each other.

Explanation of Reference Numerals

[0117] 1: Conservative management system, 90: Computer, 91: CPU, 92: ROM, 93: RAM, 94: Storage, 95: Input interface, 95a: Input device, 96: Output interface, 96a: Output device, 97: Communication interface, 97a: External server, 98: System bus, 100: IoT device, 110: Tag, 120: Current sensor, 130: Surveillance camera, 200: Data management server, 210: Control unit, 220: Storage unit, 230: Virtual storage, 230a: IoT data management table, 300: Data analysis server, 310: Control unit, 320: Fault detection module, 330: Prediction module, 340: Optimization module, 350: Storage unit, 355: IoT past data, 360: Fault data, 365: Learning data, 370: Fault mode correct data, 375: Weather data, 380: Fault detection model, 381: Anomaly detection model, 382: Waveform classification model, 383: Fault classification model, 384: Fault mode model, 385: Explanatory variable model, 386: Fault detection model, 390: Prediction model, 395: Optimization model, 396: Input variable, 397: Fault mode correlation data, 400: Terminal device, 410: Management screen, 411: Menu list, 411a: Equipment list icon, 412: Organization page, 413: User information, 414: Organization information pull-down list, 414a: Organization management icon, 420: Equipment list screen, 421: Equipment list, 422: Equipment details button, 423: Usage edit button, 430: Equipment details screen, 431: Equipment details information, 432: Equipment details edit button, 440: Equipment details edit screen, 441: Display setting list, 442: Public setting checkbox, 450: Usage management screen, 451: Usage list, 452: Details button, 460: Usage details screen, 461: Usage item list, 462: Edit button, 470: Details edit screen, 471: Item details information, 472: Guest non-display checkbox, 480: Organization management screen, 481: User list, 482: Details button, 490: User information details screen, 491: User information details list, 492: User information edit button, 495: User information edit screen, 496: Inspection registration permission checkbox

Claims

1. A prediction module that predicts future power generation and power consumption in a facility including equipment that generates power; an optimization module that estimates an optimal operation policy for equipment in the facility based on the predicted power generation and power consumption; A battery for charging the generated electricity; The prediction module predicts the amount of power generation based on actual data of power generation and power consumption in the facility and meteorological data related to a location where the equipment is installed; The optimization module calculates an approximate solution to a minimization problem that minimizes an objective function including the amount of power generated, the amount of power input from the outlet, the amount of power consumed, the amount of charge in the battery, and the amount of waste power as parameters when estimating the optimal operation policy.

2. The system comprises: A detection unit that detects data related to a state of the equipment or a state of the environment in which the equipment is installed, The system according to claim 1 , wherein the detection unit detects performance data of the power generation and consumption.

3. The system according to claim 2 , wherein the prediction module performs machine learning using the data detected by the detection unit and the weather data as input variables and the amount of power generation as a response variable, to generate a trained model for predicting the amount of power generation.

4. The system of claim 3, wherein the prediction module extracts parameters highly correlated with the amount of power generation from the data detected by the detection unit and the weather data, and predicts the amount of power generation by inputting the extracted parameters into the trained model.

5. The system of claim 1 , wherein the optimization module uses a quantum computer to calculate the approximate solution.

6. The system according to claim 1 , wherein the optimization module extracts a parameter having a high correlation with the input power amount from among a plurality of parameters related to the power consumption amount, and calculates the approximate solution.

7. A processor and a memory, The processor, A prediction module that predicts future power generation and power consumption in a facility including equipment that generates power; an optimization module that estimates an optimal operation policy for equipment in the facility based on the predicted power generation amount and power consumption amount; The prediction module predicts the amount of power generation based on actual data of power generation and power consumption in the facility and meteorological data related to a location where the equipment is installed; The optimization module is an information processing device that calculates an approximate solution to a minimization problem that minimizes an objective function that includes the amount of power generated, the amount of input power from the outlet, the amount of power consumed, the amount of battery charge, and the amount of waste power when estimating the optimal operation policy.

8. A program for causing an information processing device having a processor and a memory to function, The processor, A step of predicting future amounts of power generation and power consumption in a facility including equipment that generates power; and estimating an optimal operation policy for equipment in the facility based on the predicted power generation and power consumption. In the step of predicting, the amount of power generation is predicted based on actual data of power generation and power consumption in the facility and meteorological data related to a location where the equipment is installed; In the estimation step, a program calculates an approximate solution to a minimization problem that minimizes an objective function including the amount of power generated, the amount of power input from the outlet, the amount of power consumed, the amount of battery charge, and the amount of waste power as parameters when estimating the optimal operation policy.

9. An information processing method for causing an information processing device having a processor and a memory to function, comprising: The processor, A step of predicting future amounts of power generation and power consumption in a facility including equipment that generates power; and estimating an optimal operation policy for equipment in the facility based on the predicted power generation and power consumption. In the step of predicting, the amount of power generation is predicted based on actual data of power generation and power consumption in the facility and meteorological data related to a location where the equipment is installed; In the estimation step, an information processing method is provided in which an approximate solution to a minimization problem that minimizes an objective function including the amount of power generated, the amount of power input from the outlet, the amount of power consumed, the amount of battery charge, and the amount of waste power is calculated when estimating the optimal operation policy.

Images