JP2020028163A - Distribution system load switching system and method - Google Patents

Abstract

To provide a distribution system load switching system and method that incorporates AI.SOLUTION: A distribution system load switching system comprises: a switch operation procedure automatic formation device that forms an operation procedure for a distribution system; a model procedure formation device that obtains a model procedure by modifying an automatic formation procedure formed by the switch operation procedure automatic formation device by an expert; and a machine learning device that performs machine learning by using the automatic formation procedure and the model procedure, and obtains findings for creation of the automatic formation procedure by the switch operation procedure automatic formation device.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a distribution system load switching system for properly maintaining a distribution system load by switching a switch in a distribution system, and more particularly to a distribution system load switching system and method that integrates AI using machine learning.

  In the distribution system, a distribution system change plan was created in accordance with restoration measures after an accident occurred in the distribution system, measures to eliminate overloaded areas in the distribution system, planned construction measures accompanying expansion of the distribution system and system changes, etc. It is necessary to switch the system by switching on and off the switches according to the plan. The distribution system load switching system is a system that determines the distribution system change plan described above, and includes a switch operation procedure automatic creation device that automatically creates an operation procedure such as a switch on / off operation for such work. ing.

  The automatic creation operation procedure by the switchgear operation procedure automatic creation device is modified automatically by the system operator in consideration of the past operation results and the system status such as the time required to move to the local operation site, and is not always automatically created. The procedures are not always performed according to the procedures, and corrections and reviews of the operation procedures by human systems have occurred. At present, there are many cases in which automatically created work procedures are modified with original know-how and constraints based on the experience of unspecified field operation experts.

  Japanese Patent Application Laid-Open Publication No. HEI 10-163873 discloses a method and an apparatus for performing load accommodation by changing a distribution system.

JP 2005-117787 A

  A procedure in which a human system is added to the automatic creation procedure is called a model procedure. In general, knowledge of a skilled person is required to obtain a model procedure, but it is difficult to obtain knowledge by a skilled person, and it is also difficult to appropriately express the knowledge. It is necessary to be able to create a work procedure in consideration of constraints.

  In recent years, attention has been paid to applying AI technology as a method of reflecting the knowledge of the system operator in the automatic creation procedure. It is expected that the operation procedure will be automatically learned by applying the AI technology so that the correction by the human system is not required.

  Accordingly, it is an object of the present invention to provide a distribution system load switching system and method that integrates AI.

  In view of the above, in the present invention, "a switch operation procedure automatic creation device for creating an operation procedure in a power distribution system, and an automatic creation procedure created by the switch operation procedure automatic creation device are changed by a skilled person to form a model procedure. The model procedure creation device to obtain and the machine learning device that performs machine learning using the automatic creation procedure and the model procedure, and obtains the knowledge when the switch operation procedure automatic creation device creates the automatic creation procedure. Characteristic distribution system load switching system ".

  Further, in the present invention, "a power distribution system characterized by performing machine learning using an exemplary procedure in which a generation procedure in an automatically generated power distribution system is changed by a skilled person and obtaining knowledge when an operation procedure is automatically generated. Load switching method ".

  Also, in the present invention, "the creation procedure in the distribution system is changed by a skilled person to obtain an example procedure, and a plurality of example procedures are used to machine-learn the rules from the plan pattern and constraints of the expert, and to create the example in the distribution system. Distribution system load switching method characterized by obtaining knowledge when creating procedures. "

  ADVANTAGE OF THE INVENTION According to this invention, the distribution system load switching system which integrated AI can be obtained. As a result, in the current work, from the automatic creation procedure, the model procedure was based on the know-how and constraints of the skilled person, but from the viewpoint of automatic machine learning, it was created by applying a constraint programming method specialized in data mining automation. By reflecting the work procedure pattern and the constraint conditions based on the accumulated model values, the automatic creation procedure can be created without the expert's judgment.

The figure which shows the example of a schematic structure of the distribution system load switching system which concerns on this invention. The figure which shows the simple example of the distribution system which should perform an overload measure, for example. The figure which shows the example of the index value in the automatic preparation procedure O1 and the model procedure O2. In FIG. The figure which shows the example of calculation of the index value of the division number of a distribution line which is 1 index value. In FIG. The figure which shows the calculation example of the index value of the number of forward and backward of the switch in the distribution line which is the index value of a few. In FIG. The figure which shows the index value calculation example of the number of forwards / reverses of the automatic voltage regulator SVR in the distribution line which is an index value of 4,5. In FIG. The figure which shows the index value calculation example of the switch ON / OFF state classification in the distribution line which is the index value of 6 and 7. The figure which shows the business flow in the current business. The figure which shows the usage method which developed the machine learning function.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram showing a schematic configuration example of a distribution system load switching system according to the present invention. The distribution system load switching system shown in FIG. 1 roughly represents functions realized here as devices, and can be expressed as being constituted by three devices.

  The first device is a switch operation procedure automatic creation device 10 for automatically creating a work procedure such as an on / off operation of a switch, and the second device has a human-based hand added to the automatic creation procedure. An example procedure creating apparatus 20 for creating an example procedure, which is a procedure. The third apparatus performs machine learning using an automatic creation procedure from the switch operation procedure automatic creating apparatus 10 and an example procedure from the example procedure creating apparatus 20. This is the machine learning device 30 that performs.

  When automatically creating a work procedure from an initial state, the switcher work procedure automatic creation device 10 selects a work to be performed (selection of a target system, selection of a subject), and specifies necessary parameters. Automatically create Here, the parameters include various devices such as a circuit breaker, a switch, an SVR, and a photovoltaic power generation in a selected distribution system, information on their attributes and capacities, and information on the priority of these devices, load on each section of the distribution system. The amount is given. Also, by giving a subject name setting (selection of a target system, selection of a subject name) as the input I1, the already created work procedure can be displayed again and edited.

  In each created work procedure, operation information such as which switch is to be turned on and off and in what order are listed, and the operation is performed in accordance with the registered procedure and reflected in the distribution system diagram. be able to. As a result, the procedure created by the switch operation procedure automatic creation device 10 and obtained as the output O1 is the automatic creation procedure. In the automatic creation procedure O1, for example, a switch operation symbol to be operated, an opening / closing operation direction, a switch to be operated next, etc. are sequentially described as a switch operation procedure for overload countermeasures.

  The model procedure creating apparatus 20 is composed of a computer having a monitor and an input means. The expert procedure creating apparatus 20 is created by the switch operating procedure automatic creating apparatus 10 and displayed on a monitor. This is a device for appropriately correcting the operation procedure in consideration of know-how and constraints (input I2). A modified version of the automatic creation procedure O1 modified by a skilled person is the model procedure O2. If the expert does not make any modifications, the automatic creation procedure O1 is set as the model procedure O2.

  The exemplary procedure creation device 20 may be configured separately from the switcher operation procedure automatic creation device 10, but may be realized in the editing function of the switcher operation procedure automatic creation device 10. Good.

  FIG. 2 shows a simple example of a power distribution system in which an overload measure is to be executed. Here, a system in which distribution lines FD are arranged from substation buses BUS1 and BUS2 via distribution circuit breakers FCB1 and FCB2 is shown, and a tree-like distribution system is constructed by a switch DC (DC1 to DC7 in the illustrated example). Is formed.

  In FIG. 2, DC5 to DC7 are distant control switches, which are controlled to open and close in accordance with, for example, an open / close command signal by remote control from a command station provided on the bus side. As shown in FIG. 2, the exemplary distribution system is a system including two types of switches (on / off states) of a distribution breaker FCB, an automatic switch, and an interconnection switch of a distribution substation. DC1 to DC4 are automatic switches, and DC5 to DC7 are interconnected switches, and a section is formed between switches. Further, with respect to the distribution breaker FCB and the switch DC in FIG. 2, those shown in black are in the ON state, and those shown in white are, for example, the distribution breaker FCB2 in the OFF state.

  In the following description, the automatic creation procedure O1 prepared by the switch operation procedure automatic creation device 10 and the exemplary procedure O2 prepared by the exemplary procedure creation device 20 for the distribution system of FIG. Content

  The machine learning device 30 includes input units 31 and 32, an index value storage database DB, a machine learning unit 33, and an output unit 34.

  The input units 31 and 32 calculate an index value for each of the automatic creation procedure O1 and the model procedure O2. FIG. 3 shows examples of index values in the automatic creation procedure O1 and the model procedure O2. Here, the index value is a numerical value so that humans and computers can easily recognize the index when the configuration, operation, state, etc. of the distribution system are used as the index, and quantifies the characteristics of the work procedure It is.

  For example, FIG. 3 shows an example of the index value. The index value of 1 is the number of divisions of the distribution line. The index value of No. 2 is the number of forward switches. The index value of No. 3 is the number of reverse feeds of the switch. The index value of No. 4 is the number of forwards when the automatic voltage regulator SVR is installed on the distribution line. The index value of No. 5 is the number of reverse feeds when installing the automatic voltage regulator SVR on the distribution line. The index value of No. 6 is the number of ON / OFF switch types. The index value of No. 7 is the number of switchgear types. The index value of 8 is the load of each interchange section. Other examples of index values are as illustrated in FIG. 3, but other index values can be grasped as index values. Note that these indices can be used for various operations in a switch, a circuit breaker, an automatic voltage regulator SVR, a load, a photovoltaic power generation, etc., which are components of the distribution system, or an operation state of the distribution system (for example, in each interchange section). Load, etc.).

  Next, specific calculation examples of representative index values will be described with reference to FIGS. 4 to 7 exemplify the power distribution system shown in FIG. 2, in which the switches DC1, DC3, and DC5 are turned off, and the switches DC2, DC4, DC6, DC7, and the circuit breaker FCB1 are turned on. A description will be given with reference to the distribution system described above. Regarding the distribution breaker FCB and the switch DC in these figures, those shown in black are in the ON state, and those shown in white are, for example, the distribution breaker FCB2 in the OFF state.

  FIG. It is a figure which shows the index value calculation example of the division | segmentation number of a distribution line which is 1 index value. The number of divisions indicates how many distribution lines are divided and accommodated in the event of a system power failure. This index indicates whether priority is given to collective accommodation of the entire system or to fine divisional accommodation. In FIG. 3, since one distribution line FD is divided into two parts, distribution line A and distribution line B, the number of divisions is two.

  FIG. It is a figure which shows the index value calculation example of the forward / backward number of switches of the distribution line which is 2 or 3 index values. The forward transmission refers to a state in which the distribution breaker FCB of the power supply substation exists on the primary side of the switch DC, and the reverse transmission refers to a state in which the distribution breaker FCB exists on the secondary side of the switch DC. This index indicates the number of switches in which a normally-open switch is accommodated by forward / reverse feed for one distribution line FD. In FIG. 5, the number of forward feeds is one switch DC4, and the number of reverse feeds is one switch DC2.

  FIG. 6 shows NO. It is a figure which shows the index value calculation example of the number of forwards / reverses of the automatic voltage regulator SVR in the distribution line which is an index value of 4,5. This index indicates the number of automatic voltage regulators SVR that are forward / reverse on the interchange route. FIG. 6 shows that the automatic voltage regulator SVR is installed at an intermediate position between the switch DC2 and the switch DC4. In this case, the number of forward feeds of the automatic voltage regulator SVR is 0 and the number of reverse feeds is 1. .

  FIG. 7 shows NO. It is a figure which shows the index value calculation example of a switch ON / OFF state classification in a distribution line which is an index value of 6 and 7. This index indicates whether the ON / OFF switch and the ON / OFF switch are remote control or manual between the interchange groups. In FIG. 7, the on / off switch is remote control 4, manual 0, and the off switch is far control 1, manual 0. The more the entry, the less the cut. The relationship is that if the number of cuts increases, the entry decreases.

  The other indices will not be described in detail, but the indices are indexed and grasped in the same way.

  Returning to FIG. 3, numerical examples of index values in the automatic creation procedure O1 and the model procedure O2 are NO. 1 to NO8 are illustrated. NO. The number of divisions of the distribution line of No. 1 is set to “1” in the automatic creation procedure O1, but is set to “3” in the model procedure O2. Regarding the number of ON / OFF switch types of 6, the number of ON / OFF switches of the remote control switch is "4" in the automatic creation procedure O1, but the ON / OFF switch number of the remote control switch is "2" in the exemplary procedure O2. . Regarding the number of switchgear types of 7, the number of open switches is 2 for the remote control switch in the automatic creation procedure O1, but the number of open switches is 4 for the remote control switch in the exemplary procedure O2. . In the automatic creation procedure O1, the section 1 is set to “100 kW” and the section 2 is set to “200 kW” for the load of each of the eight interchange sections, but in the exemplary procedure O2, the section 1 is set to “300 kW” and the section 2 is set to “50 kW”.

  According to FIG. 3, the index values of NO1, 6, 7, and 8 are corrected by the expert using the model procedure creating apparatus 20 with respect to the automatic creating procedure O1 prepared by the switch operating procedure automatic creating apparatus 10. It can be said that this is a correction item of the exemplary procedure O2.

  Further, according to FIG. 3, the index values of NO2 to NO5 are the same in the automatic creation procedure O1 and the model procedure O2, and can be said to be items that have not been modified by a skilled person. In the case of this power distribution system, it can be said that the expert does not object to the decision items of the switch operation procedure automatic creation device 10 and agrees with each other.

  Further, according to FIG. 3, the designer of the distribution system load switching system of FIG. 1 to NO. 14, but at least in this distribution system, the NO. It means that events corresponding to cases 9 to 14 were not extracted.

  Returning to FIG. From the input units 31 and 32 in FIG. 1, among the index values in FIG. 3, NO. Data is input for 1 to NO8.

  The index values from the input units 31 and 32 are sequentially input to the index value storage database DB in FIG. 1, and the index value of the automatic creation procedure O1 and the index value of the exemplary procedure O2 are stored in a comparison structure as shown in FIG. Will be done. In addition, the index value at the time of another subject setting (selection of target system, selection of subject) I1 is also sequentially input and stored in the index value storage database DB. Therefore, data is formed inside the index value storage database DB including the information of subject setting.

  The machine learning unit 33 of FIG. 1 performs a statistical analysis of the index values for the index values stored in the index value storage database DB. The analysis process is preferably performed in a state where the index values accumulated in the index value storage database DB have reached a considerable number, and may be executed, for example, at a frequency of about once / month.

  In the processing in the machine learning unit 33, first, the index value of each subject is analyzed, and the priority of the index value is analyzed. Next, the priority order of the index values is analyzed for all of the accumulated subjects. Here, the former analysis of the index value of each subject is based on No. 2 which is an item modified by a skilled person in this subject. For each of the index values 1, 6, 7, and 8, a priority is determined among these index values, and the priority order is determined by the index value alone. On the other hand, in the latter determination of the priority order of the index values for all the accumulated subjects, the priority of the index values is determined while observing the correlation between the index values.

  The latter analysis processing for all the stored subject names will be specifically described with reference to the index values stored in the index value storage database DB of FIG. 3 as an example. In the case of the predetermined subject setting I1 in FIG. The index values of 1, 6, 7, and 8 are different between the automatic creation procedure O1 and the model procedure O2. For example, in the one-time / month process of the machine learning unit 33, many cases obtained by setting a plurality of subjects are accumulated, and the difference is analyzed in the entire accumulated cases.

  Now, all the 200 cases that have been accumulated are checked, and the changed procedure is No. Assuming that there are four of 1, 6, 7, and 8, machine learning is performed as “higher priority” when there are many cases that have been changed considering the combination of the four indices as follows.

  In this comprehensive example, 120 of the 200 cases were No. No. 1 and 8 were changed. Nos. 1, 6, and 7 were changed. It is assumed that 1.6.7.8 has been changed.

  From this, No. No. 1 is determined to be the first priority since it has been changed in all cases. No. Nos. 6 and 7 have been changed for 50 out of 200 cases, so the priority is 2nd in the tie, and No. 8 has been changed for 30 out of 200 cases, and is therefore ranked third in priority. The procedure created in this way is calculated and set as an automatic creation procedure (O2 ').

  In the above case, the priority of the index value is determined and reflected in the automatic creation procedure (O2 ′), and the case of outputting the index value is shown. Anything that outputs knowledge such as correlations may be used. The output of the priority or the like means, in short, a knowledge obtained by learning opportunities by focusing on the difference between the index values in the automatic creation procedure O1 and the model procedure O2.

  The output unit 34 in FIG. 1 feeds back the creation procedure performed by the machine learning unit 33 to the switch operation procedure automatic creation device 10. In this case, the feedback may be such that the knowledge obtained by the machine learning unit 33 is directly and automatically reflected on the switchgear work procedure automatic creation device 10, or the switchgear work procedure automatic generation is performed based on the judgment of the operator. The information may be reflected on the creation device 10.

  In the feedback to the switch operation procedure automatic creation device 10, for example, an operator who attaches importance to the creation procedure which is a learning result of the machine learning unit 33 with respect to the index value A determines that “priority” is given, and the switch operation procedure automatic creation is performed. When there is a difference in the determination that the creation device 10 determines “not prioritized”, the priority of the index value A in the switch operation procedure automatic creation device 10 is set higher than the current setting so that the priority is higher than the current setting. Automatically change parameters.

  The above describes an example of creating a work procedure using AI based on the index values in FIG. The calculation examples of the index values in FIGS. 4 to 7 merely show how one index value is divided, and one index value is used when actually creating a work procedure using AI. It is good to pay attention to. If the index values focused on in the automatic creation procedure O1 and the model procedure O2 are different, the AI is used to find differences in other index values that are difficult for humans to find (operator's rules, know-how, etc.) and automatically learn. It is better to let. For example, for 120 out of 200 cases, the expert If Nos. 1 and 8 have been changed, 1 and No. It is considered that there is an operator rule, know-how, and the like showing the correlation of 8, and it is preferable that knowledge is acquired by machine learning and reflected in the switch operation procedure automatic creation device 10.

  FIG. 8A is a diagram showing a business flow in a current business. As shown in FIG. 1, in the current operation, an automatic creation procedure O1 is obtained by giving an input I1 such as selection of a target system, selection of a work subject, etc. to the switchgear work procedure automatic creation device 10 and then automatic creation. This is a flow in which a model procedure O2 is obtained by modifying the procedure O1 in a model procedure creating apparatus 20 by a human-based hand, such as know-how and restrictions of a skilled person.

  On the other hand, in the present invention, as shown in FIG. 1, the contents of the automatic creation procedure O1 and the exemplary procedure O2 are indexed and grasped, machine learning is performed on different index value portions, and knowledge obtained by machine learning is obtained. Is reflected in the switch operation procedure automatic creation device 10 so that the knowledge based on the AI is reflected in the subsequent plans.

  A further use of the machine learning function according to the invention is shown in FIG. 8b. According to the usage system of FIG. 8B, by applying a new constraint programming technique in which machine learning and mathematical calculation techniques are linked, the created model procedure O2 data is stored in the database DB, and work is performed from the past model procedure O2. By automatically learning the pattern of the procedure O1 and generating the work procedure O1 in consideration of the work procedure pattern and the constraint conditions in the machine learning device 30, it is possible to automatically create the work procedure O2 'without the judgment of a skilled person. Is possible.

Claims (13)

  A switch work procedure automatic creation device for creating a work procedure in a power distribution system, a model procedure creation apparatus that obtains a model procedure by changing an automatic creation procedure created by the switch work procedure automatic creation apparatus by a skilled person, A power distribution system comprising: a machine learning device that performs machine learning using a creation procedure and the model procedure, and the switch operation procedure automatic creation device obtains knowledge when creating the automatic creation procedure. Load switching system. The distribution system load switching system according to claim 1, wherein
The machine learning device is different from the automatic creation procedure and the exemplary procedure in an index value storage database that stores the index values quantifying the characteristics of the work procedure in the power distribution system with respect to the automatic creation procedure and the exemplary procedure. A load switching system for a distribution system, comprising: a machine learning unit that performs machine learning by focusing on an index value of a quantified numerical value and obtains the knowledge. The distribution system load switching system according to claim 2, wherein
The machine learning device, wherein, for the index value of the quantified numerical value different between the automatic creation procedure and the model procedure in the same power distribution system, a priority order between the index values is obtained as the knowledge. Load switching system. The distribution system load switching system according to claim 2 or 3, wherein:
The machine learning device, for the index value of the quantified numerical value different between the automatic creation procedure and the model procedure in a plurality of distribution systems, to obtain the priority of the index value as the knowledge based on the frequency of occurrence of the index value. A distribution system load switching system characterized by the following. The power distribution system load switching system according to any one of claims 2 to 4, wherein
The machine learning device is characterized in that, for the index value of the quantified numerical value different between the automatic creation procedure and the model procedure in a plurality of distribution systems, the knowledge is obtained based on the frequency of occurrence of a combination of index values. Distribution system load switching system. The distribution system load switching system according to any one of claims 2 to 5, wherein:
A distribution system load switching system, characterized in that a division number indicating how many distribution lines are divided and accommodated is used as the index value. The distribution system load switching system according to any one of claims 2 to 5, wherein:
As the index value, for one distribution line, the number of forward switches indicating the number of switches in which the always-in remote control switch is accommodated in the forward direction, or the number of switches in which the always-in remote control switch is accommodated in the reverse direction. A power distribution system load switching system, characterized by using a number of reverse transmissions indicating the number of units. The distribution system load switching system according to any one of claims 2 to 5, wherein:
A power distribution system using, as the index value, an SVR forward number indicating the number of SVRs to be forwarded on an interchange route or an SVR reverse number to indicate an SVR number to be backward on one distribution line. Load switching system. The distribution system load switching system according to any one of claims 2 to 5, wherein:
As the index value, in one distribution line, an on-operation type that classifies whether the on-off switch is remote control or manual, or an off-operation type that separates whether the on-off switch is remote control or manual is used. Distribution system load switching system.   A distribution system load switching method, wherein machine learning is performed using an exemplary procedure in which a work procedure in an automatically created distribution system is changed by an expert to obtain knowledge when the work procedure is automatically created. The distribution system load switching method according to claim 10, wherein
An index value obtained by quantifying the characteristics of the work procedure in the distribution system is stored as data for the work procedure and the model procedure, and machine learning is performed by focusing on index values of quantified values different between the work procedure and the model procedure. And a machine learning unit for obtaining the knowledge.   When creating a creation procedure in the distribution system by changing the creation procedure in the distribution system by a skilled person to obtain an example procedure, using a plurality of the above-described example procedures to machine-learn a rule from a plan pattern and constraints of the expert. A load switching method for a distribution system, characterized by obtaining the knowledge of the above. The distribution system load switching method according to claim 12, wherein
An index value quantifying the characteristics of the work procedure in the distribution system is stored as data on the exemplary procedure, the rules are learned from the expert's plan patterns and constraints, and the knowledge when creating the creation procedure in the distribution system is obtained. Distributing system load switching method characterized by obtaining.

Images