JP7152461B2 - Prediction model generation device and method - Google Patents

Description

The present invention relates to a predictive model generating apparatus and method, and is suitable for application to a predictive model generating apparatus that generates a predictive model for predicting the occurrence of equipment failure below a low-voltage service line in a distribution system, for example.

2. Description of the Related Art In recent years, installation of smart meters, which are watt-hour meters with two-way communication functions, to consumers has been promoted. By using smart meters, it is possible to obtain detailed data such as the amount of electricity used by consumers every 30 minutes (hereafter referred to as the 30-minute value) and reverse power flow from in-house power generation, which could not be obtained with conventional metering systems. , information on various events detected by the smart meter, such as voltage drops, power outages, and power restorations (hereafter referred to as event information).

For such smart meters, Patent Document 1 discloses a technique related to the collection of regular meter readings, which is the basic function of the smart meters. Also, in recent years, various technologies related to smart meters, such as technologies related to other communication methods, have been proposed.

Furthermore, in recent years, with the spread of smart meters, there is a movement to consider the use of meter reading information such as 30-minute values and monthly power consumption sent from smart meters to center equipment, and event information for various events. There is also For example, Patent Document 2 describes in detail the causes of accidental power outages and distribution equipment failures by utilizing 30-minute values and event information from smart meters installed in power pipes, and then identifying the causes. A data management system has been proposed for each of these systems, which is capable of instructing how to deal with these accidental power outages and failures of power distribution facilities.

Further, Patent Literature 3 discloses a technique for estimating the position where a failure such as disconnection occurs based on an event notification of a smart meter, in relation to a phase group estimation processing program for estimating the failure position of distribution equipment. .

JP 2005-352532 A JP 2018-207690 A WO2018/179283

By the way, generally, it is difficult to predict equipment failure below the low-voltage service line, such as blowing of a fuse provided in a low-voltage service line that connects a transformer installed on a utility pole and a smart meter, or disconnection. As a result, the maintenance and management of power distribution systems are faced with problems such as an increase in costs due to emergency outsourcing work, a deterioration in service quality due to defects in low-voltage service lines, and a sense of busyness due to power outages that occur day and night.

In addition, the techniques disclosed in Patent Documents 2 and 3 require a program dedicated to the equipment in order to identify and estimate the failure of the power equipment, so it is not easy to expand to other equipment.

The present invention has been made in consideration of the above points, and solves the above problems at once, reduces the cost required for maintenance and management of the distribution system, improves the quality of the distribution service, and makes it possible to cope with power outages. The object of the present invention is to propose a predictive model generating apparatus and method that can eliminate the sense of anxiety.

In order to solve this problem, in the present invention, in a prediction model generation device for generating a prediction model for predicting the occurrence of equipment failure below the low voltage service line, each specific value detected by the smart meter installed in each consumer leveling each of the specific events detected by the smart meter within a predetermined first period into a plurality of levels based on the frequency of occurrence of each of the specific events recognized based on the event information of the event; an event leveling unit that counts the number of occurrences of the specific event of the level for each specific event; A prediction model generation unit for generating the model is provided.

Further, in the present invention, a prediction model generation method executed by a prediction model generation device for generating a prediction model for predicting the occurrence of equipment failure below the low-voltage service line, wherein the smart meter installed in each customer each specific event detected by the smart meter within a predetermined first period is divided into a plurality of levels based on the frequency of occurrence of each specific event recognized based on the event information of each specific event detected by a first step of respectively leveling and counting, for each of the specific events, the number of occurrences of the specific event of each of the levels; A second step of generating is provided.

According to the predictive model generating apparatus and method of the present invention, it is possible to accurately predict the occurrence of equipment failure below the low-voltage service line based on the generated predictive model. By taking measures in advance, it is possible to reduce the increase in the cost of maintenance and management work for the distribution system due to emergency outsourcing work and the frequency of power outage response.

According to the present invention, it is possible to realize a predictive model generating apparatus and method capable of improving the quality of power distribution service while reducing the cost required for maintenance and management of the power distribution system, and eliminating the busy feeling of dealing with power outages.

1 is a block diagram showing a physical configuration of a predictive model generation device according to this embodiment; FIG. 4 is a chart showing the configuration of an event information management table; FIG. 10 is a chart showing the configuration of a 30-minute value management table; FIG. 1 is a block diagram showing a logical configuration of a predictive model generation device according to this embodiment; FIG. (A) and (B) are diagrams and charts for explaining leveling of specific events by an event leveling unit. FIG. 4 is a diagram for explaining leveling of specific events by an event leveling unit; FIG. 4 is a chart for explaining event leveling information; FIG. FIG. 10 is a chart for explaining learning data; FIG. It is a figure which shows the structural example of a prediction model. FIG. 4 is a diagram for explaining a learning data acquisition period, a verification data acquisition period, and an actual data acquisition period; 7 is a graph for explaining additional periods of the verification data acquisition period; (A) to (C) are Venn diagrams for explaining hit rate and coverage rate. FIG. 11 is a chart showing a configuration example of a first event count management table; FIG. 4 is a chart showing a configuration example of an event leveling table; 4 is a chart showing a configuration example of a power consumption management table; 4 is a chart showing a configuration example of an equipment information management table; 4 is a chart showing a configuration example of an elapsed time management table;

One embodiment of the present invention will be described in detail below with reference to the drawings.

(1) Configuration of Predictive Model Generating Device According to Present Embodiment (1-1) Physical Configuration of Predictive Model Generating Device In FIG. 1, 1 indicates the predictive model generating device according to the present embodiment. This prediction model generation device 1 is configured with a restriction cancellation discount system 2 , a meter data management system (MDMS) 3 and various other systems 4 .

The prediction model generation device 1 has a prediction model generation function for generating a prediction model for predicting the occurrence of equipment failure (hereinafter referred to as a power failure) below the low voltage service line in the distribution system, and a CPU (Central Processing Unit) 10 , a memory 11A, a storage device 11B and a display device 12.

The CPU 10 is a processor that controls the operation of the prediction model generation device 1 as a whole. The memory 11A is composed of, for example, a volatile semiconductor memory, and is used as a work memory for the CPU 10. FIG. The memory 11A stores an analysis tool 13, which is application software for realizing such a predictive model generating function, which will be described later.

The storage device 11B is used to store programs and necessary information for a long period of time, and is composed of a large-capacity nonvolatile storage device such as a hard disk device, SSD (Solid State Drive), or flash memory. In the storage device 11B, in addition to various programs, a first event frequency management table 14, an event leveling table 15, a power consumption management table 16, an equipment information management table 17, an elapsed time management table 18, and a second event The frequency management table 19, the prediction model 20 generated by the prediction model generation function, and the like are stored.

The display device 12 is composed of, for example, a liquid crystal display or an organic EL (Electro Luminescence) display, and is used to display necessary information.

On the other hand, the meter data management system 3 is a computer device having a function of managing event information and 30-minute values transmitted from each smart meter (not shown) within its jurisdiction and configuration information of each smart meter. .

In practice, the meter data management system 3 registers and manages event information appropriately transmitted from smart meters installed in each customer in an event information management table 21 as shown in FIG. In FIG. 2 , each row of information in the event information management table 21 represents event information 40 .

Further, the meter data management system 3 registers and manages the 30-minute value of the corresponding consumer, which is transmitted from each smart meter in a 30-minute cycle, for example, in the 30-minute value management table 22 shown in FIG. . In FIG. 3, the values in each column of each row of the 30-minute value management table 22 represent the 30-minute value 42 of the corresponding time period of one customer. Furthermore, the meter data management system 3 registers and manages the configuration information of each smart meter such as a single-phase three-wire type and a three-phase three-wire type in a configuration information table (not shown).

These pieces of information managed by the meter data management system 3 (event information 40, consumer 30-minute values 42, and smart meter configuration information) are stored in semiconductor memory such as USB (Universal Serial Bus) memory, It is manually provided to the prediction model generation device 1 via an optical disc such as a CD (Compact Disc) or a DVD (Digital Versatile Disc), or a portable storage device such as a portable hard disk device.

The restriction cancellation discount system 2 is composed of one or a plurality of computer devices that manage records of various power outages that have occurred within its jurisdiction. The various systems 4 are composed of one or a plurality of computer devices that manage equipment information such as installation date of each smart meter to the customer, inspection results and facility design information, completion information, and repair results.

Information on various power outage records managed by the restriction cancellation discount system 2 and the above-mentioned equipment information managed by the various systems 4 are also manually sent to the prediction model generation device via a semiconductor memory, an optical disk, or a portable storage device. 1.

(1-2) Logical Configuration of Predictive Model Generating Device FIG. 4 shows the logical configuration of the predictive model generating device 1 . As is clear from FIG. 4, the predictive model generation device 1 includes a first event number management unit 30, an event leveling unit 31, a power consumption management unit 32, an equipment information management unit 33, an elapsed time management unit 34, A second event frequency management unit 35, a prediction model generation unit 36, a model evaluation unit 37, a failure occurrence prediction unit 38, a first event frequency management table 14, an event leveling table 15, a power consumption management table 16, and equipment information It comprises a management table 17, an elapsed time management table 18, and a second event count management table 19. FIG.

First event frequency management unit 30, event leveling unit 31, power consumption management unit 32, facility information management unit 33, elapsed time management unit 34, second event frequency management unit 35, prediction model generation unit 36, model evaluation The unit 37 and the failure prediction unit 38 are functional units that are implemented by the CPU 10 (FIG. 1) executing the analysis tool 13 (FIG. 1) stored in the memory 11A (FIG. 11).

The first event frequency management unit 30 selects the date and time of equipment failure (blackout) that occurred in one of the low-voltage service lines from the information on the performance of various power failures provided as described above, and the predetermined period related to the low-voltage service line that occurred. (At least the period from the current time to "a/aa" in FIG. 10) minutes of power failure record information 45 and event information 40 for the same period from the provided event information 40 are extracted, and these extracted information , the number of occurrences of specific events (hereinafter referred to as specific events) detected by each smart meter within that period is counted for each smart meter and each event type. The first event frequency management unit 30 stores and manages the count result in the first event frequency management table 14, which will be described later with reference to FIG.

Note that the event information 40 acquired by the first event frequency management unit 30 is used to generate the prediction model 20 based on the event information 40 as will be described later. It shall be a sign of equipment failure below the low voltage service line. Therefore, it is essential that the facility failure (power failure) has already occurred at this time. Therefore, when acquiring the event information 40, the first event count management unit 30 acquires event information of an event that occurred during a predetermined period (hereinafter referred to as a learning data acquisition period) some time before the current time. (See FIG. 10). The length of this learning data acquisition period is the period that includes most of the specific events (at least a predetermined percentage) that are signs of facility failure below the low-voltage service line that occurred at the end of the learning data acquisition period, and is empirically determined. length is defined.

In the following description, it is assumed that the low-voltage service line is a single-phase three-wire system. Therefore, the event of each smart meter for which the first event frequency management unit 30 should acquire the event information 40 is the "1 side voltage "1-side power failure" in which the power supply from the 1-side wire is stopped, and the voltage of the 3-side wire of the low-voltage lead-in line connected to the 3-side terminal of the smart meter There are three types of "3 side voltage drop" in which the power supply is stopped.

However, in the present embodiment, an event for which event information 40 should be acquired is also “cover opening/closing” in which the cover of the smart meter is opened/closed by a worker or the like for inspection work. Even in this case, the first event count management unit 30 selects only the three types of events, ie, "voltage drop on side 1", "blackout on side 1", and "voltage drop on side 3", excluding "cover opening/closing", as described above. The number of occurrences of each event is counted as a "specific event".

Based on the event information 40 acquired by the first event frequency management unit 30, the event leveling unit 31 assigns each specific event detected by the smart meter within the learning data acquisition period from "1" to It has a function to level to 5 levels up to "5".

Specifically, as shown in FIGS. 5A, 5B, and 6, the event leveling unit 31 generates a specific event for which the smart meter is to be leveled (hereinafter, this is appropriately referred to as a leveling target specific event). is detected, and the number of times the smart meter detected a specific event of the same type within the previous predetermined time (“N hours” in FIG. 6, for example, 3 hours) is 0 to 1 levels the leveling target specific event to "Level 1 (Lv.1)".

If the number of times the smart meter detects the specific event of the same type within the predetermined time immediately before the smart meter detects the specific event to be leveled is 2 to 4 times, the event leveling unit 31 determines "level 2 (Lv.2)", 5-8 times "Level 3 (Lv.3)", 9-13 times "Level 4 (Lv.4)", 14 times or more Levels each leveling target specific event to "Level 5 (Lv.5)".

Furthermore, the event leveling unit 31 similarly levels "all specific events" and "'1 side power outage' or '1 side voltage drop'".

Specifically, for the “all specific events”, the event leveling unit 31 determines that the smart meter detects any specific events (leveling target specific events) within a predetermined time immediately before the smart meter detects any specific events. "Level 1 (Lv.1)" if the number of times the event was detected is 0 to 1, "Level 2 (Lv.2)" if it is 2 to 4, and "Level 2 (Lv.2)" if it is 5 to 8. "Level 3 (Lv.3)", "Level 4 (Lv.4)" in the case of 9 to 13 times, "Level 5 (Lv.5)" in the case of 14 times or more, the leveling target specific event level each.

In addition, the event leveling unit 31 detects a specific event (leveling target specific event) of "1 side power failure" or "1 side voltage drop" for "'1 side power failure' or '1 side voltage drop'" If the number of times the smart meter has detected a specific event of "1 side power outage" or "1 side voltage drop" within the predetermined time immediately before the time of the event is 0 to 1 times, "Level 1 (Lv.1) ”, “Level 2 (Lv.2)” for 2 to 4 times, “Level 3 (Lv.3)” for 5 to 8 times, “Level 4 (Lv.3)” for 9 to 13 times Level 4)”, and if it is 14 times or more, level the leveling target specific event to “Level 5 (Lv.5)” respectively.

The reason why "'1 side power failure' or '1 side voltage drop'" is targeted for leveling is that the smart meter detects the power failure and voltage drop that occurred on the 1 side terminal as "1 side power failure" and "1 side voltage drop", respectively. ” can be detected and output as different events, whereas the power failure and voltage drop that occurred at the 3-side terminal can only be detected as one type of event, “3-side voltage drop”. This is to align the output units of the side terminals.

Then, the event leveling unit 31 converts the acquired event information 40 as shown in FIG. and for each of "'1-side blackout' or '1-side voltage drop'", the corresponding specific event ("1-side blackout", "1-side voltage drop", "3-side voltage drop", "all specific events" or The event leveling information 41 as shown in FIG. 7 in which only the leveling results of "'1 side power outage' or '1 side voltage drop'" are collected is processed.

As is clear from FIG. 7, the event leveling information 41 includes "1-side power failure", "1-side voltage drop", "3-side voltage drop", "all specific events", and ""1-side power failure" or " 1-side voltage drop”” is information in which the number of occurrences of the corresponding specific event within a predetermined time period immediately before the occurrence and the level given to the specific event by the leveling described above are associated with each other.

Also, based on the event leveling information 41 obtained by such processing, the event leveling unit 31 detects “1 side power outage”, “1 side voltage drop”, and “3 side voltage drop” detected by the smart meter for each smart meter. ”, “all specific events”, and ““side 1 power outage” or “side 1 voltage drop”” are counted, and the count results are used for “1 side power outage” and “1 side voltage drop", "3 side voltage drop", "all specific events", and "'side 1 power outage' or 'side 1 voltage drop'" are stored and managed in event leveling tables 15 generated respectively.

The power usage amount management unit 32 extracts the 30-minute value 42 of each consumer during the learning data acquisition period from the provided 30-minute value 42 (Fig. 4) of each consumer, and Based on the 30-minute value 42 of , the maximum value of power usage for one day in each season of spring, summer, autumn and winter (hereinafter referred to as the maximum daily power usage) and the 30-minute value 42 (hereinafter referred to as the 30-minute maximum value), respectively, and stores and manages the calculation results and extraction results in the power consumption management table 16 described later with reference to FIG. . In the case of the present embodiment, the power usage amount management unit 32 sets “spring” to May, “summer” to August, “autumn” to November, and “winter” to February. Calculate and extract the daily power consumption maximum value and the 30-minute maximum value, respectively.

The facility information management unit 33 has a function of extracting the installation date of each smart meter from the facility information provided as described above, and storing and managing the extraction result in the facility information management table 17 described later with reference to FIG. have In addition, the elapsed time management unit 34 calculates the number of days elapsed from the installation date of each smart meter based on the facility information of each smart meter stored in the facility information management table 17, and the calculation results will be described later with reference to FIG. It has a function of registering and managing the elapsed time management table 18 .

Based on the event information 40 provided as described above, the second event frequency management unit 35 detects “1 side power failure”, “1 side voltage drop”, and “3 events” detected by each smart meter within a predetermined period. Count the number of occurrences of "1 side voltage drop", "all specific events", and "'1 side power outage' or '1 side voltage drop'" for each smart meter and event type, and count the count results as a second event to be described later. It has a function of storing and managing in the frequency management table 19 .

The events for which the number of occurrences is counted by the second event frequency management unit 35 are the above-mentioned "1 side power outage", "1 side voltage drop", "3 side voltage drop", "all specific events" and ""1 side voltage drop". Power failure” or “Voltage drop on side 1”, only events related to cases where equipment failure below the low-voltage service line has not yet occurred. Therefore, the second event frequency management unit 35 can be divided into "3-side voltage drop", "1-side power outage", " The number of occurrences of "1 side voltage drop", "all specific events" and "'1 side power outage' or '1 side voltage drop'" is counted.

The prediction model generation unit 36 uses various information stored in the first event frequency management table 14, the event leveling table 15, the power consumption management table 16, the equipment information management table 17, and the elapsed time management table 18 as learning data. machine-learning the correlation between the frequency of occurrence of specific events for each type of smart meter and the occurrence of equipment failure below the low-voltage service line, and a prediction model 20 for predicting the occurrence of such equipment failure for each type of smart meter. , respectively.

In practice, as shown in FIG. 8, the prediction model generation unit 36 includes information 50A representing the number of occurrences of each specific event for each smart meter stored in the first event number management table 14 (FIG. 4), and event Information 50B representing the number of occurrences of events of each level for each smart meter stored in the leveling table 15 (FIG. 4), and the consumer corresponding to each smart meter stored in the power consumption management table 16 (FIG. 4) Information 50C representing the daily maximum power consumption and 30-minute maximum value for each season, and information 50D representing the installation date of each smart meter stored in the equipment information management table 17 (Fig. 4) to the customer. , and information 50E representing the elapsed time from the installation date to the consumer for each smart meter stored in the elapsed time management table 18 (FIG. 4), and learning data are collected for each smart meter using the meter ID as a key. Generate 50.

Based on the learning data 50 thus generated, the predictive model generation unit 36 generates a decision tree model as shown in FIG. 9 as the predictive model 20 using an existing technique.

The model evaluation section 37 has a function of evaluating the accuracy of the prediction model 20 generated by the prediction model generation section 36 . In practice, the model evaluation unit 37 uses the data stored in the first event frequency management table 14, the event leveling table 15, the power consumption management table 16, the equipment information management table 17, and the elapsed time management table 18, respectively. , various data related to equipment failure (power outage) below the low-voltage service line for which countermeasures have already been completed are used as verification data 51, and the verification data 51 is input to the prediction model 20. FIG.

As shown in FIG. 10, the learning data 50 used by the prediction model generation unit 36 to generate the prediction model 20 is long-term data in order to obtain more samples until equipment failure occurs in the low-voltage service line and below. On the other hand, if the verification data 51 used by the model evaluation unit 37 to verify the accuracy of the prediction model 20 is data for a period sufficient for such verification (hereinafter referred to as a verification period) good. However, in addition to the data of the verification period, various data of a predetermined period (hereinafter referred to as an additional period) are added and used as the verification data 51 . This is for the purpose of using the predictive model 20 to predict facility defects below the low-voltage service line that occurred before and after the start of the verification period.

In this case, as shown in FIG. 11, the horizontal axis represents the number of past days before the occurrence of equipment failure below the low voltage service line, and the occurrence of specific events detected by the corresponding smart meters until such failure occurred. When I created a graph of several actual examples with the cumulative number of times on the vertical axis, 80% of the specific events detected until the equipment failure below the low voltage service line finally occurred. We were able to confirm that it occurred up to 60 days ago. Therefore, in this embodiment, the additional period is set to 60 days, and the verification period is set to 60 days from the time when the last specific event occurs or when there is a track record of dealing with facility failures below the low-voltage service line. The period up to the previous day shall be selected as the verification period.

Then, the model evaluation unit 37 compares the prediction result of the equipment failure below the low-voltage service line obtained by inputting the above-described verification data into the prediction model 20 with the actual equipment failure below the low-voltage service line, so as to evaluate the prediction model. Evaluate accuracy. In this case, such an evaluation is performed by calculating the accuracy rate and coverage rate of the equipment failure occurrence prediction by the prediction model 20 .

Here, as shown in FIG. 12A, "A" is a set of events in which a low-voltage service line equipment failure (power outage) actually occurred, and a set of events predicted by the prediction model that a low-voltage service line equipment failure will occur. is "C", the set "B", which is the overlapping part of the set "A" and the set "C", is the set of events for which the prediction model was able to correctly predict the equipment failure of the low voltage service line.

Thus, such a hit rate can be calculated by dividing the number of events belonging to the set "B" by the number of events belonging to the set "C" (B/C), and such a coverage rate can be calculated by dividing the number of events belonging to the set "B" by the number of events belonging to the set "C". It can be calculated by dividing the number of events belonging to set "A" by the number of events belonging to set "A" (B/A). The relationship among set "A", set "B" and set "C" when the hit rate is 100% is shown in Fig. 12(B). , set “B” and set “C” are shown in FIG. 12(C).

Then, the model evaluation unit 37 displays on the display device 12 (FIG. 1) the accuracy rate and coverage rate of the occurrence of equipment failure below the low-voltage service line using the prediction model 20 calculated in this manner.

As a result, the user can confirm the prediction accuracy of the prediction model 20 created at that time based on this display result. Further, when the prediction accuracy is low, the user repeats regenerating the prediction model 20 by providing the prediction model generation unit 36 with learning data 50 for a longer period of time than the previous time. Thereby, the accuracy of the prediction model 20 can be gradually improved. Then, when the accuracy of the prediction model 20 reaches a certain level or higher, the user instructs the prediction model generation device 1 to predict the occurrence of equipment failure below the low-voltage service line using the prediction model 20. .

As a result, the defect occurrence prediction unit 38, which has received such an instruction, uses each data stored in the event leveling table 15, the power consumption management table 16, the equipment information management table 17, and the elapsed time management table 18 at that time, Among the data stored in the second event frequency management table 19, data that can be a sign of equipment failure below the low-voltage service line that has not yet been detected is captured as actual data 52, and this actual data 52 is used by the prediction model 20. Predict equipment failure below the low-voltage service line that will be discovered or will occur in the future. Then, the defect occurrence prediction unit 38 notifies this prediction result 53 to the person in charge of facility maintenance of the company or the consignment company, and instructs inspection of the facilities below the predicted low-voltage service line.

As the actual data 52, as shown in FIG. 10, data from the period just before the current time (hereinafter referred to as the analysis period) to the extent that equipment failure below the low-voltage service line can be predicted can be used. A period obtained by adding a predetermined additional period to the analysis period is set as an actual data acquisition period, and various data in the actual data acquisition period are used as the actual data 52 . This is to allow the predictive model 20 to predict facility failures in the low-voltage service line that occurred before and after the analysis period started.

On the other hand, the first event frequency management table 14 holds the number of occurrences of each specific event detected by each smart meter within the most recent predetermined period counted by the first event frequency management unit 30 (FIG. 4). This table is used for management. The first event count management table 14, as shown in FIG. 13, comprises a management number column 14A, an instrument ID column 14B, a 1-side voltage drop column 14C, a 1-side power outage column 14D, and a 3-side voltage drop column 14E. be done. In the first event frequency management table 14, one row corresponds to one smart meter.

The management number column 14A stores a management number (a serial number in this embodiment) assigned to the information in that row in the first event count management table 14, and the instrument ID column 14B stores the information. An identifier (meter ID) unique to the smart meter assigned to the smart meter corresponding to is stored.

The 1-side voltage drop column 14C stores the number of occurrences of "1-side voltage drop" detected by the corresponding smart meter within a predetermined period counted by the first event frequency management unit 30. The power failure column 14D stores the number of occurrences of the “1-side power failure” detected by the smart meter within the predetermined period counted by the first event frequency management unit 30 . Furthermore, in the 3-side voltage drop column 14E, occurrence of "3-side voltage drop (including 3-side power failure)" detected within a predetermined period of time counted by the first event frequency management unit 30 by the corresponding smart meter The number of times is stored.

Therefore, in the case of the example of FIG. 13, the smart meter with the meter ID "instrument A" has "1 side voltage drop" "11" times, "1 side power failure" "1" time, and " 3 side voltage drop” is detected “0” times.

The event leveling table 15 includes "1 side power outage", "1 side voltage drop", "3 side voltage drop", "all specific events" and ""1 side power outage" leveled by the event leveling unit 31 (Fig. 4). or ``Voltage drop on side 1'''', which is used to hold and manage the total number of occurrences for each level of the event. voltage drop", "all specific events" and "'one side blackout' or 'one side voltage drop'", respectively. The event leveling table 15, as shown in FIG. 14, comprises a serial number column 15A, an instrument ID column 15B, and a plurality of level columns 15C. In the event leveling table 15, one row corresponds to one smart meter.

The reference number column 15A stores the management number (serial number in this embodiment) assigned to the information in that row in the event leveling table 15, and the instrument ID column 15B stores the smart number corresponding to the information. A meter ID of the smart meter assigned to the meter is stored.

Each level column 15C is provided corresponding to each of the five levels 1 to 5, and each level column 15C stores the number of occurrences of the corresponding level counted by the event leveling section 31. be done.

Therefore, in the example of FIG. 14, for the smart meter with the meter ID "Instrument A", the number of events leveled to level 1 ("Lv. 2"), the number of times the event was detected was "4", level 3 ("Lv.3"), level 4 ("Lv.4"), and level 5 ("Lv.5"). It is shown that the number of times of detection of the event detected was "0".

The power consumption management table 16 includes the maximum value of the power consumption for one day for each season of each consumer calculated and extracted by the power consumption management unit 32 (hereinafter referred to as the daily maximum value), This table is used to manage the maximum 30-minute value (hereinafter referred to as the 30-minute maximum value). As shown in FIG. 15, the power consumption management table 16 includes a reference number column 16A, a meter ID column 16B, a daily maximum value column 16C for each season, and a 30-minute maximum value column 16D. In the power consumption management table 16, one row corresponds to one smart meter installed in one customer.

The reference number column 16A stores a management number (a serial number in this embodiment) assigned to the information in that row in the power consumption management table 16, and the meter ID column 16B stores the information corresponding to the information. The meter ID of the smart meter assigned to the smart meter is stored.

In addition, the daily maximum value column 16C for each season stores the power usage amount (daily maximum value) of the day when the power usage amount was maximum in the corresponding season, and the 30-minute value maximum value column 16D for each season is stored. stores the maximum 30-minute value (maximum 30-minute value) in the corresponding season.

As described above, in the present embodiment, "winter" is February, "spring" is May, "summer" is August, and "autumn" is November. The maximum value column 16C and the maximum 30-minute value column 16D store the daily maximum value and the 30-minute maximum value in February, respectively. 16D stores the daily maximum value and the 30-minute maximum value in May, respectively. The daily maximum value field 16C and the 30-minute maximum value field 16D corresponding to "summer" store the daily maximum value and the 30-minute maximum value in August, respectively. The column 16C and 30-minute maximum value column 16D store the daily maximum value and the 30-minute maximum value in November, respectively.

Therefore, in the case of the example of FIG. 15, for a consumer installed with a smart meter with a meter ID of "meter A", the daily maximum value in "winter" is "550", the 30-minute maximum value is "40", and " The daily maximum value for spring is ``500'', the 30-minute maximum value is ``30'', the daily maximum value for ``Summer'' is ``600'', the 30-minute maximum value is ``50'', and the daily maximum value for ``Autumn'' is `` 450”, and the 30 minute maximum was shown to be “30”.

The facility information management table 17 is a table used to hold and manage the installation date of each smart meter acquired by the facility information management unit 33 (FIG. 4). It is configured with a meter ID column 17B and a meter installation date column 17C. In the facility information management table 17, one row corresponds to one smart meter.

The reference number column 17A stores a management number (serial number in this embodiment) assigned to the information in that row in the power consumption management table 16, and the meter ID column 17B stores the information corresponding to the information. The meter ID of the smart meter assigned to the smart meter is stored. The meter installation date column 17C stores the date when the corresponding smart meter was installed at the corresponding consumer.

Therefore, in the example of FIG. 16, the smart meter with the meter ID "meter A" is installed on the consumer side corresponding to "2018/3/16".

In addition, the elapsed time management table 18 is calculated by the elapsed time management unit 34 (FIG. 4) based on the installation date of each smart meter stored in the equipment information management table 17 at the customer's end. This is a table used to hold and manage the number of days that have passed since installation at the customer's end, and as shown in FIG. be done. In the elapsed time management table 18, one row corresponds to one smart meter.

The reference number column 18A stores the management number (serial number in this embodiment) assigned to the information in that row in the elapsed time management table 18, and the instrument ID column 18B stores the information corresponding to the information. A meter ID of the smart meter assigned to the smart meter is stored. The elapsed time column 18C stores the elapsed time (in the present embodiment, the number of days) since the corresponding smart meter was installed at the customer's end.

Therefore, in the example of FIG. 17, the number of days since the smart meter with the meter ID "meter A" was installed at the corresponding consumer's destination is "808" days.

The second event count management table 19 has the same configuration as the first event count management table 14 except that the actual data 52 of the actual data acquisition period described above with reference to FIG. 10 is stored. detailed description is omitted.

(2) Operation and Effects of the Predictive Model Generating Device of the Present Embodiment In the predictive model generating device 1 of the present embodiment having the above configuration, "1 side power outage", "1 side voltage drop", and "3 side voltage". Specific events corresponding to each event such as drop", "all specific events", and "'1 side power failure' or '1 side voltage drop'" are respectively leveled, and this leveling result, the number of occurrences of each specific event, and each Learning data 50 is generated based on the seasonal maximum daily power consumption value and the maximum 30-minute value of the consumer and the elapsed time since each smart meter was installed on the consumer side, and the generated learning A predictive model 20 is generated based on data 50 .

In this case, using the leveling results of specific events to predict facility failures below the low-voltage service line is based on the empirical rule that the frequency of failures increases as equipment and equipment near the end of their lives. This is based on the assumption that as the number of occurrences of events increases, the occurrence of equipment failure below the low-voltage service line will approach, and by using the leveling results of specific events in this way, a more accurate prediction model 20 is generated. It is considered possible.

In addition, in this prediction model generation device 1, by using the seasonal maximum daily power consumption value and the maximum 30-minute value of each consumer as learning data, A prediction model 20 that takes into account the quantity of electricity is generated, and furthermore, by using equipment information such as the elapsed time since each smart meter was installed on the consumer side as learning data 50, the prediction model 20 that also takes into account deterioration over time. can be generated.

Therefore, according to the prediction model generating device 1, it is possible to generate a highly accurate prediction model 20, and based on this prediction model 20, it is possible to accurately predict the occurrence of equipment failure below the low voltage service line, By taking advance measures such as parts replacement before the occurrence of such equipment failure, it is possible to reduce the increase in maintenance and management costs of the distribution system due to emergency outsourcing work and the frequency of power outage response. Therefore, according to the predictive model generation device 1, it is possible to improve the quality of the distribution service while reducing the cost required for the maintenance and management of the distribution system, and to eliminate the busy feeling of dealing with power failures.

(3) Other Embodiments In the above-described embodiment, the case where the prediction model generation device is configured by one computer device was described, but the present invention is not limited to this. It may also be composed of a plurality of computer devices interconnected through a network. In this case, the first event frequency management unit 30, the event leveling unit 31, the power consumption management unit 32, the facility information management unit 33, the elapsed time management unit 34, the second event frequency management unit 35, the prediction model generation The unit 36, the model evaluation unit 37, and the defect occurrence prediction unit 38 may be arranged in a distributed manner in these plural computer devices.

In the above-described embodiment, as described above with reference to FIG. 8, the information 50A representing the number of occurrences of each specific event for each smart meter stored in the first event number management table 14 (FIG. 4) and the event Information 50B representing the number of occurrences of events of each level for each smart meter stored in the leveling table 15 (FIG. 4), and the consumer corresponding to each smart meter stored in the power consumption management table 16 (FIG. 4) Information 50C representing the daily maximum power consumption and 30-minute maximum value for each season, and information 50D representing the installation date of each smart meter stored in the equipment information management table 17 (Fig. 4) to the customer. , and information 50E representing the elapsed time from the installation date to each customer, stored in the elapsed time management table 18 (FIG. 4), are grouped for each smart meter using the meter ID as a key to generate learning data 50. However, the present invention is not limited to this, and part of these information may be omitted, or other information may be added as learning data 50 in addition to or in place of these information. may

Furthermore, in the above-described embodiment, the case where various information held by the control cancellation discount system 2, the meter data management system 3, and the various systems 4 is manually provided to the prediction model generation device 1 has been described. The invention is not limited to this, for example, the prediction model generation device 1 is connected to the control suspension discount system 2, the meter data management system 3, and various systems 4 via the network, and the prediction model generation device 1 is connected to the control suspension discount system 2 , the meter data management system 3 and various systems 4 via a network.

INDUSTRIAL APPLICABILITY The present invention can be applied to a predictive model generation device that generates a predictive model for predicting the occurrence of equipment failure below the low voltage service line.

1 Prediction model generation device 10 CPU 13 Analysis tool 14 First event count management table 15 Event leveling table 16 Power consumption management table 17 Facilities Information management table 18 Elapsed time management table 19 Second event frequency management table 20 Prediction model 30 First event frequency management unit 31 Event leveling unit 32 Power consumption management unit 33 Equipment information management unit 34 Elapsed time management unit 35 Second event frequency management unit 36 Prediction model generation unit 37 Model evaluation unit 38 Defect occurrence prediction unit 40 Event information 42 30-minute value 43 Configuration information 44 Equipment information 50 Learning data 51 Verification data 52 Actual data 53 …… Prediction result.

Claims (10)

In a prediction model generation device that generates a prediction model for predicting the occurrence of equipment failure below the low voltage service line,
detected by the smart meter within a predetermined first period based on the frequency of occurrence of each specific event recognized based on the event information of each specific event detected by the smart meter installed in each consumer an event leveling unit that levels each of the specific events into a plurality of levels, and counts the number of occurrences of the specific event of each of the levels for each of the specific events;
and a prediction model generation unit that generates the prediction model based on the number of occurrences of each of the specific events for each level counted by the event leveling unit. The first period is
2. The predictive model generation device according to claim 1, wherein the period includes a predetermined ratio or more of the specific event that is a sign of equipment failure below the service line that occurred at the end of the first period. The prediction model generation unit is
3. The predictive model is generated based on the number of occurrences of each specific event for each level and the elapsed time from the installation date of each smart meter to the customer. 3. The prediction model generation device according to 2. Further comprising a power usage amount management unit that calculates the maximum value of the 30-minute value for each season and the maximum value of the power usage amount for each day, based on the 30-minute value of each consumer measured by the smart meter. ,
The prediction model generation unit is
2. The prediction model is generated based on the number of occurrences of each specific event for each level, the maximum value of the 30-minute values for each season, and the maximum value of power consumption for each day. The predictive model generation device according to any one of claims 1 to 3. The specific event is
The predictive model generation device according to any one of claims 1 to 4, wherein the event is an event that can be a sign of the equipment failure. A prediction model generation method executed by a prediction model generation device that generates a prediction model for predicting the occurrence of equipment failure below a low-voltage service line,
detected by the smart meter within a predetermined first period based on the frequency of occurrence of each specific event recognized based on the event information of each specific event detected by the smart meter installed in each customer a first step of leveling each of the specific events into a plurality of levels, and counting the number of occurrences of the specific event of each of the levels for each of the specific events;
and a second step of generating the prediction model based on the counted number of occurrences of each of the specific events for each level. The first period is
7. The predictive model generation method according to claim 6, wherein the period includes a predetermined ratio or more of the specific event that is a sign of equipment failure below the service line that occurred at the end of the first period. In the second step,
6. The predictive model is generated based on the number of occurrences of each specific event for each level and the elapsed time from the installation date of each smart meter to the customer. 7. The predictive model generation method according to 7. In the first step,
Based on the 30-minute value of each consumer measured by the smart meter, calculating the maximum value of the 30-minute value for each season and the maximum value of the daily power consumption,
In the second step,
7. The prediction model is generated based on the number of occurrences of each of the specific events for each level, the maximum value of the 30-minute values for each season, and the maximum value of power consumption for each day. The prediction model generation method according to any one of claims 1 to 8. The specific event is
The predictive model generation method according to any one of claims 6 to 9, wherein the event is an event that can be a sign of the equipment failure.

Images