JP7643296B2 - Transformer load prediction system and transformer load prediction method - Google Patents

Description

The present invention relates to a transformer load prediction system and a transformer load prediction method.

Patent Literature 1 describes a load estimation method for a transformer configured for the purpose of estimating the maximum load with high accuracy using an existing meter. In the method, the contract capacity and electricity usage of a plurality of consumers connected to each transformer are acquired, a usage estimation function showing the relationship between the contract capacity and the estimated usage is derived from the contract capacity and the electricity usage, a load estimate value estimating the maximum load of each ammeter-equipped transformer is calculated, an actual load value which is the maximum load measured in each ammeter-equipped transformer is acquired, a load error showing the error between the load estimate value and the actual load value is calculated, a usage error showing the error between the estimated usage obtained using the usage estimation function and the electricity usage is calculated, an error function showing the relationship between the load error and the usage error is derived from the load error and the usage error, and the load estimate value is corrected using the error function.

Patent Document 2 describes a method for calculating the design capacity of a transformer that utilizes the measured values from automatic meter readings of low-voltage consumers to calculate the capacity of a pole-mounted transformer suited to the consumer load current. In the method, low-voltage consumers are classified by contract type, their power usage is compiled into a database to create a standard pattern of power usage, and the standard pattern is stored as data over the past to determine the increase in power usage from the past to the present. When calculating the capacity of a pole-mounted transformer, the required contract type and number of low-voltage consumers are investigated, a contract type load pattern is extracted from the power usage database, the total power is calculated for each time period, and the capacity of the pole-mounted transformer is calculated by multiplying the increase in power usage from the past to the present.

JP 2011-205859 A JP 2010-20653 A

In Patent Document 1 and Patent Document 2, the required capacity of the transformer is calculated based on the contract type (contract form) and night-time power. However, as the electric power industry becomes more and more divided in the future, it is expected that some electric power companies will not be able to obtain information on the sales contract type of consumers, making it difficult to calculate the capacity of the transformer.

In addition, transformer capacity has traditionally been calculated by determining the maximum power for each contract type from the number of households and current for each contract type, and then simply adding them up. However, this method does not necessarily provide sufficient accuracy in calculating the maximum power, and it is thought that there are many cases in which transformers with larger capacities than necessary are installed on site.

The present invention has been made in consideration of the above background, and has an object to provide a transformer load prediction system and a transformer load prediction method that are capable of accurately determining the maximum power of a transformer.

In order to achieve the above object, one of the present inventions is a transformer load prediction system, comprising:
The system is configured using an information processing device having a processor and a storage device, and generates a load prediction model which is a model in which the number of main switches by capacity installed in each consumer receiving power supply from a transformer is used as an explanatory variable and the maximum power of the transformer is used as an objective variable, and the objective variable is obtained by inputting the number of main switches by capacity installed in each consumer receiving power supply from a prediction target transformer, which is a transformer for which maximum power is to be predicted, into the load prediction model as an explanatory variable, and then outputs the obtained objective variable.

According to the present invention, the maximum power predicted for the transformer to be predicted can be obtained with high accuracy. Therefore, a transformer with a capacity required to reliably prevent overload can be appropriately selected. In addition, it is possible to prevent a transformer with an unnecessarily large capacity from being installed at the site, and the operation of the transformer (new installation, replacement, etc.) can be performed efficiently.

Another aspect of the present invention is the above-mentioned transformer load prediction system, which is characterized in that, when a main switch with an unknown capacity is present among the main switches installed at a consumer receiving power supply from the transformer to be predicted, the capacity of the main switch with an unknown capacity is determined based on the contents of the power supply contract for the consumer in which the main switch is installed.

According to the present invention, even if a main switch whose capacity is unknown is installed in a consumer receiving power supply from the transformer to be predicted, the maximum power predicted for the transformer to be predicted can be accurately determined.

Another aspect of the present invention is the transformer load prediction system, comprising:
The model is characterized in that it is generated by any one of simple regression analysis, multiple regression analysis, DNN (Deep Neural Network), and Support Vector Machine (SVM).

Another aspect of the present invention is the above-mentioned transformer load prediction system, characterized in that in the multiple regression analysis, at least one of the number of contracts for the supply method (single-phase three-wire 100V), the total contracted power for metered electricity lighting B, and the number of contracts with water heaters is further used as an explanatory variable.

By performing multiple regression analysis using these explanatory variables, the accuracy of prediction of maximum power can be further improved.

Another aspect of the present invention is the above-mentioned transformer load prediction system, which stores the relationship between the number of main switches receiving power from the transformer and the actual measured value of the transformer's maximum power, determines the relationship between the number of main switches and the predicted value of the transformer's maximum power based on the load prediction model, generates an approximation equation showing the relationship between the number of main switches and the difference between the predicted value and the actual measured value, and outputs information showing the difference determined using the generated approximation equation.

According to the present invention, the difference between the predicted value and the actual measured value, which is an estimation index of the margin (surplus), can be accurately calculated. Therefore, the transformer capacity required for the transformer to be predicted can be accurately determined based on the predicted maximum power and the calculated margin.

Other problems and solutions disclosed in this application are described in the section on the preferred embodiment of the invention.
and the drawings.

According to the present invention, the maximum power of a transformer can be determined with high accuracy, and a transformer with an appropriate capacity can be selected as the transformer to be installed at the site.

1 is a diagram showing a schematic configuration of a transformer load prediction system according to an embodiment of the present invention; 2 is a diagram illustrating the relationship between a transformer, a smart meter, a main switch, and a load. FIG. 1 is a graph illustrating a method for determining the capacity of a transformer. FIG. 2 is a diagram showing main functions of a transformer load prediction device. 2 is an example of a hardware configuration of a transformer load prediction device. 1 shows examples of load forecasting models (models a to d). FIG. 13 is a diagram illustrating an example of a simple regression analysis when the capacity of the main switch is 30 A. FIG. 13 is a diagram illustrating an example of a simple regression analysis when the capacity of the main switch is 60 A. FIG. 13 is a diagram illustrating an example of a simple regression analysis when the capacity of the main switch is 100 A. FIG. 13 is a diagram illustrating an example of a simple regression analysis when the capacity of the main switch is 200 A. 13 is an example of the results of a simple regression analysis. 13 is an example of a result of a multiple regression analysis. 13 is a graph showing the results of an analysis of the influence of explanatory variables on a response variable (maximum power). FIG. 2 is a diagram illustrating an example of consumer information. FIG. 11 is a diagram illustrating an example of measurement value information. FIG. 11 is a system flow diagram illustrating a model learning process. 13 is an example of a learning data setting screen. This is an example of regional division. FIG. 11 is a system flow diagram illustrating a load prediction process. 13 is an example of a result display screen. 1 is a graph showing the relationship between the number of contracts and maximum power consumption based on actual measurements. 1 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using a conventional method. 13 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using model a. 13 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using model b. 13 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using model c. 13 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using model d. This is a table showing the coefficient of determination ^R2 for easy comparison. 13 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using model b. 13 is a table showing the coefficients and intercepts of each explanatory variable in model b. 13 is a graph showing the results of predicting the relationship between the number of contracts and maximum demand using model d. 13 is a table showing the coefficients and intercepts of each explanatory variable of model d. 13 is a graph plotting the difference between the predicted value and the actual value for each number of contracts for model b, and showing the upper limit of the interval in which 95% of the predicted values fall. 13 is a graph plotting the difference between the predicted value and the actual value for each number of contracts for model d, and showing the upper limit of the interval in which 95% of the predicted values fall. 13 is a graph plotting the difference between the predicted value and the actual value for each number of contracts for model b, and shows the upper limit of the interval in which 99.7% of the predicted values fall. 13 is a graph plotting the difference between the predicted value and the actual value for each number of contracts for model d, and showing the upper limit of the interval in which 99.7% of the predicted values fall. 14B is a graph showing the results of approximating the upper limit value of the interval in which 95% of the predicted value falls, using an approximation formula, based on FIG. 14A. 15B is a table showing the parameters of each approximation formula in FIG. 15A, the ratio of the number of times that the predicted value exceeds the actual measured value, and the average value of the above differences. 14C is a graph showing the results of approximating the upper limit value of the interval in which 95% of the predicted value falls, using an approximation formula, based on FIG. 14B. 15C , the ratio of the number of times the predicted value exceeds the actual measured value, and the average value of the above differences. 14D is a graph showing the results of approximating the upper limit value of the interval in which 99.7% of the predicted values fall, using an approximation formula, based on FIG. 14C. 15E , a table showing the parameters of each approximation formula, the ratio of the number of times that the predicted value exceeds the actual measured value, and the average value of the above differences. 14D, a graph showing the results of approximating the upper limit value of the interval in which 99.7% of the predicted value falls, using an approximation formula. 15G , a table showing the parameters of each approximation formula, the ratio of the number of times that the predicted value exceeds the actual measured value, and the average value of the above differences.

Hereinafter, an embodiment will be described with reference to the drawings. In the following description, the same or similar components will be denoted by the same reference numerals, and the description thereof will be omitted.

1 shows a schematic configuration of an information processing system (hereinafter referred to as "transformer load prediction system 1") described as one embodiment. As shown in the figure, the transformer load prediction system 1 includes a power information management device 4 and a transformer load prediction device (hereinafter referred to as "load prediction device 100") installed in a system center 3 of an electric utility such as an electric power company, a large number of smart meters SM installed in consumers 2 that receive power supply via a transformer,
The system center 3 is, for example, a control center of a smart grid or a central command center of a power company. The consumers 2 are, for example, general households, public facilities,
These include office buildings, various public facilities, factories, and infrastructure facilities.

The power information management device 4 is configured using an information processing device (computer) and
The communication network 5 is, for example, a wide area network (WAN).
power supply), concentrator, 920Hz band communication network, PLC (Power Line Communication
) and other communication infrastructure.

The smart meter SM transmits measurement values of power supplied from the power transmission and distribution system to the consumers 2 installed in each facility as needed (e.g., every 30 minutes) to the power information management device 4 via the communication network 5. The power information management device 4 provides information for monitoring the presence or absence of an abnormality in the power supply at the consumers 2 based on various measurement values (voltage value, current value, active power, reactive power, power factor, temperature, etc.) transmitted from the smart meter SM.

The load prediction device 100 is configured using an information processing device, and communicates with the load prediction device 100 via a communication means such as a LAN (Local Area Network) or a communication network 5 provided in a system center 3.
The load prediction device 100 is communicatively connected to a power information management device 4. The load prediction device 100 predicts the maximum power (maximum load) of the transformer supplying power to each of the consumers 2, using the measurement values acquired from the smart meter SM and information on the consumers 2 provided by the power information management device 4.
In addition, the load prediction device 100 estimates the margin (
The load prediction device 100 calculates the margin (maximum power) and presents the calculated margin together with the maximum power (maximum load) to the user. Based on the maximum power and margin presented by the load prediction device 100, the user can efficiently select and determine a transformer to be installed at the site that has an appropriate capacity that prevents overload and is not excessive.

2A is a diagram illustrating the relationship between a transformer 7 (such as a pole transformer or an underground transformer) that supplies power to a consumer 2 and devices installed in the consumer 2. As shown in the figure, the transformer 7
The transformer 7 transforms the power sent from the distribution system 8 and supplies it to one or more consumers 2 belonging to its control (the bank of the transformer 7). A smart meter SM is connected to the transformer 7 via a service line and a distribution line (single-phase two-wire system, single-phase three-wire system, three-phase three-wire system, etc.), and a main switch MS (breaker) is provided between the smart meter SM and the load L. As the appropriate capacity of the transformer 7, an appropriate capacity that prevents overload and is not excessive is selected. For example, from among multiple types of transformers 7 with different capacities (5 kW, 10 kW, 20 kW, 30 kW, 50 kW, etc.), the one with the smallest capacity and a value equal to or greater than the predicted maximum power (maximum load) plus a margin is selected. The capacity of the main switch MS (30 A, 40 A, 50 A, 60 A, 75 A, etc.) is selected.
, 100 A, 125 A, 150 A, etc.) is determined by a power supply contract between the consumer 2 and a power supplier such as a power company or an expected maximum load.

FIG. 2B is a diagram for explaining a method for determining the capacity of a transformer. Here, as an example, a method for determining the capacity of the transformer 7A shown in FIG. 2A will be explained. The horizontal axis of the graph shown in the figure is time (hours), and the vertical axis is power. Solid lines P, Q, and R are the power consumption (measured values of the smart meter SM) of three loads L (P, Q, R) that are supplied with power from the transformer 7A, respectively, and the dashed line G is the total value of the power consumption of the three loads L (P, Q, R). In this example, the total value around 12:00 during the period shown in the graph is the maximum power (7.8 kW). As the capacity of the transformer 7A, it is necessary to select a transformer with a capacity of 7.8 kW or more, and for example, a transformer with a capacity of 10 kW is selected.

3 is a block diagram showing main functions of the load prediction device 100. As shown in the figure, the load prediction device 100 includes a storage unit 110, a consumer information management unit 120, a measurement value acquisition unit 130, and a load prediction unit 140.
30, learning data generation unit 135, model learning unit 140, prediction target transformer information acquisition unit 145
, a load prediction unit 150 , a margin calculation unit 155 , and a result presentation unit 160 .

Of the above functions, the memory unit 110 stores each piece of information (data) such as customer information 111, measurement value information 112, learning data 113, load prediction model 114, prediction target transformer information 115, load prediction result 116, and margin calculation result 117.

The consumer information 111 includes information about each consumer 2 (hereinafter referred to as "consumer information"). The consumer information 111 is obtained from a user via a user interface, as well as from various systems (such as a power distribution control system and a customer management system) operated by an electric utility such as a power company. The consumer information 111 includes the latest consumer information and past consumer information about each consumer 2. Specific examples of consumer information will be described later.

The measurement value information 112 includes power values measured by the smart meter SM for each consumer 2 at each predetermined time (time series data of measurement values (data in which measurement values are associated with time information) acquired by the power information management device 4 from the smart meter SM of each consumer 2 via the communication network 5, hereinafter referred to as "measurement value information"). The measurement value information 112 includes the most recently acquired information and information acquired in the past. In the measurement value information 112, each piece of measurement value information is managed in association with the identifier of the smart meter SM from which it was acquired (meter ID, described later). The entity of the measurement value information 112 may be, for example, a file managed by a file system, a table in a database, a storage area in a network storage (for example, a URL (Uniform Resource Locator)),
The data is managed in a storage area specified by a data Locator (data source locator), etc.

The learning data 113 includes learning data (teacher data) used for learning a load prediction model 114, which will be described later. The learning data 113 is generated from the measurement value information 112 or information obtained by interpolating missing information of the measurement value information 112. The learning data 113 will be described in detail later.

The load forecast model 114 is a model for predicting loads of the loads that are subordinate to the loads that are supplied with power (present in the bank).
Information on a transformer (hereinafter referred to as a "transformer to be predicted") that is a target of prediction of maximum power, which is the maximum value of power consumption in the consumer 2 (hereinafter referred to as "transformer to be predicted") (hereinafter referred to as "transformer information to be predicted").
is a machine learning model that outputs the maximum power as a response variable when the load prediction model 114 is input as an explanatory variable. The substance of the load prediction model 114 is, for example, a polynomial, a determinant, a mathematical expression, a vector, or the like that includes adjustable parameters. The load prediction model 114 will be described in detail later.

The prediction target transformer information 115 includes prediction target transformer information that is input as an explanatory variable to the load prediction model 114. The prediction target transformer information is, for example, automatically generated by the load prediction device 100 based on the customer information 111 and the measurement value information 112, or is input by a user via a user interface.

The load prediction result 116 includes a response variable (hereinafter, referred to as a “load prediction result”) that is output by the load prediction model 114 by inputting the explanatory variables to the load prediction model 114 .

The margin calculation result 117 includes the margin calculated by the margin calculation unit 155.

3, the consumer information management unit 120 manages (acquires, edits, registers, and deletes consumer information) the consumer information 111. The consumer information management unit 120 provides, for example, a user interface for managing the consumer information.

The measurement value acquiring unit 130 acquires measurement value information sent from the power information management device 4 at a predetermined time interval (such as every 30 minutes), and manages the acquired measurement value information as measurement value information 112 .

The learning data generation unit 135 generates learning data 113 (teacher data) based on the consumer information 111 and the measurement value information 112 .

The prediction target transformer information acquisition unit 145 acquires prediction target transformer information.

The load prediction unit 150 obtains the future maximum power of the transformer to be predicted by inputting the information of the transformer to be predicted as an explanatory variable into the load prediction model 114. The maximum power obtained by the load prediction unit 150 is stored in the storage unit 110 as the load prediction result 116.

The margin calculation unit 155 calculates the margin based on the predicted value of the maximum power calculated by the load prediction unit 150 and the actual measured value (measured value) of the maximum power.
10 stores the result as the margin calculation result 117. The method of calculating the margin will be described later in detail.

The result presentation unit 160 presents the load prediction result 116 and the margin calculation result 117 to the user via a user interface.

4 shows an example of a hardware configuration of an information processing device used to realize the load prediction device 100. The illustrated information processing device 10 includes a processor 11, a main memory device 12, an auxiliary memory device 13, and a processor 14.
3, an input device 14, an output device 15, and a communication device 16. Specific examples of the information processing device 10 include personal computers, office computers, various server devices,
A general-purpose machine, etc. The information processing device 10 may be realized, in whole or in part, by using virtual information processing resources provided by using a virtualization technology, a process space separation technology, etc., such as a virtual server provided by a cloud system. The load prediction device 100 may be realized by using a plurality of information processing devices 10 connected to each other so as to be able to communicate with each other.

In the figure, the processor 11 is, for example, a CPU (Central Processing Unit),
PU (Micro Processing Unit), GPU (Graphics Processing Unit), FPGA (Fie
ld Programmable Gate Array), ASIC (Application Specific Integrated Circuit)
), AI (Artificial Intelligence) chips, etc.

The main memory device 12 is a device for storing programs and data, and is, for example, a ROM (Read Only Memory).
Only Memory, RAM (Random Access Memory), Non-Volatile Memory (NVRAM (Non
Volatile RAM), etc.

The auxiliary storage device 13 may be, for example, a solid state drive (SSD), a hard disk drive, an optical storage device (such as a compact disc (CD) or a digital versatile disc (DVD)),
The auxiliary storage device 13 may be a storage system, a read/write device for a recording medium such as an IC card, an SD card, or an optical recording medium, or a storage area of a cloud server. Programs and data can be read into the auxiliary storage device 13 via a recording medium reader or a communication device 16.
The programs and data stored in the main memory 3 are read into the main memory 12 as required.

The input device 14 is an interface that accepts input from the outside, and is, for example, a keyboard, a mouse, a touch panel, a card reader, a pen-input tablet, a voice input device, or the like.

The output device 15 is an interface that outputs various information such as the process progress and the process results.
The output device 15 is, for example, a display device (LCD (Liquid Crystal Display)) that visualizes the various information described above.
a device that converts the above-mentioned various information into voice (a voice output device (a speaker, etc.)), and a device that converts the above-mentioned various information into text (a printer, etc.). Note that, for example, the information processing device 10 may be configured to input and output information to and from other devices via the communication device 16.

The input device 14 and the output device 15 constitute a user interface that receives information from the user and presents information.

The communication device 16 communicates with other devices via a communication infrastructure such as a communication network 5 (
A device that realizes wired or wireless communication, such as a NIC (Network Interface Card).
rd), a wireless communication module, a USB module, etc.

The information processing device 10 includes, for example, an operating system, a file system, a database (DB),
MS (DataBase Management System) (relational database, NoSQL, etc.),
A key-value store (KVS) or the like may also be introduced.

The functions of the load prediction device 100 are implemented by the processor 1 of each of the information processing devices 10.
1 reads out and executes a program stored in each main memory 12, or the hardware (FPGA, ASIC, AI, etc.) constituting the load prediction device 100
The load prediction device 100 stores the above-mentioned various types of information (data) as, for example, a table in a database or a file managed by a file system.

=Load forecast model=
In this embodiment, taking into consideration the case where the capacity of the main switch MS provided in the consumer premises 2 cannot be obtained, the load prediction model is expressed by the following equation.
[Equation 1]
Load forecast model = Maximum power (customer with unknown main switch capacity)
+ Maximum power (customers with known main switch capacity)
...Equation 1

Here, the maximum power in the first term of the above formula can use a conventional formula for calculating maximum power (hereinafter referred to as the "conventional formula"), or can be found by simple regression analysis and multiple regression analysis by substituting the main switch capacity with the actual (understood) contract details (contract type, etc.). On the other hand, the second term in the above formula cannot use the conventional formula because the contract details (contract type, etc.) are unknown, and must be found by either simple regression analysis or multiple regression analysis.

Therefore, in this embodiment, a study was conducted on each of the four models (models a to d) shown in FIG. 5, which are combinations of the first and second terms of the above formula.

<Conventional>
An example of the conventional formula used for models a and c is shown in the following formula.
[Number 2]
P1 = a + b + c + d ・・・ Formula 2
In the above formula, P1 is the expected peak load, a is the flat-rate equivalent (Σ(kW)), and b is the metered lighting equivalent A
(Σ(current value x number of houses x voltage)), c is the TOU equivalent (Σ(current value x 1.5 x number of houses x voltage)), d is
This is the equivalent of metered electricity B (Σ(kW×0.7)). Σ means the sum for each consumer (contract).
It is.

When applying the above formula, for example, the current value per household of consumer 2 using metered electricity A is assumed based on past measurements and the number of households using metered electricity A in the same bank (in increments of one household). In this example, for time-of-use lighting contracts, the assumed current value is multiplied by 1.5, and for metered electricity B, the contracted power (KVA) is multiplied by 0.7.

<Simple regression analysis>
Regarding the first term of model a or the simple regression analysis of model b, verification was performed by determining the slope and intercept for each capacity of the main switch MS, using the maximum power for each number of units for each capacity of the main switch MS as an explanatory variable.

6A to 6D show an example of a simple regression analysis.
13 is a graph showing an example of the results of a simple regression analysis of the relationship between the number of units (number of contracts) and the maximum power for each number of units when the capacity of S is 30 A, 60 A, 100 A, and 200 A.

FIG. 6E shows the capacity of the main switch MS of 30A, 40A, 50A, 60A, and 75A.
The slope and intercept were obtained by performing a simple regression analysis for the cases of A, 100A, 125A, 150A, and 200A. The explanatory variables used to obtain the slope and intercept in Table 2 are shown in FIG.
A different one was used than that used in Figures 5A to 5D.

An example of the first term of model a (or model b) constructed based on FIG. 6E is shown in the following equation.
[Equation 3]
The first term of model a (or model b)
=((1.299×n+2.120)×n ₃₀ +(1.279×n+2.029)×n ₄₀ +...
+(4.928×n+5.501)×n ₁₅₀ +(6.544×n+7.199)×n ₂₀₀ )/n
...Equation 3
In the above formula, n is the total number of contracts in the bank of the transformer being forecasted, and n _x is the number of main switchgears by capacity (x is capacity [A]).

<Multiple regression analysis>
For the second term of model c or the multiple regression analysis of model d, the explanatory variables were
(registered trademark) to obtain the coefficients (including constant parts) for the number of units by main switch capacity, thereby conducting verification.

Figure 6F shows an example of values output by the above statistical model of multiple regression analysis. In addition to the estimated values (estimates) of the coefficients (including constant parts) of the number of units by main switch capacity, the standard error (stderr)
, t-value, p-value (p>|t|), confidence interval ([0.025 0.975]), and coefficient of determination (R ² ) are shown.

An example of the second term of model c or model d constructed based on FIG. 6F is shown in the following equation.
[Equation 4]
Maximum power [kW] = 0.4042n ₃₀ + 0.9847n ₄₀ + 1.7370n ₅₀ +2.0604n ₆₀ ...+ 6.5222n ₂₀₀
+ 5.1914
...Equation 4
In the above formula, n _x is the number of main switchgear of each capacity (x is the capacity [A]).

<Verification of adding explanatory variables>
In order to improve the accuracy of multiple regression analysis, we verified the addition of explanatory variables (features). Specifically, we used the "XGboost" data analysis library in Python (registered trademark).
" was used to rank the degree of influence of the explanatory variables on the objective variable (maximum power).

The result of the ranking is shown in Fig. 7. In this embodiment, the top three explanatory variables in Fig. 7, i.e., "number of contracts for supply system (single-phase three-wire 100V)", "total contracted power for metered electricity B", and "number of contracts with water heaters" were added and multiple regression analysis was performed to confirm the effect on accuracy.

=Learning data=
The learning data (teaching data) used for learning the load prediction model 114 is the measured value (3
The measured values (0 minute values) were randomly extracted, and the extracted measured values were used as explanatory variables. When data was missing, the average value of contracts with common attributes was used for interpolation. In addition, the power value (30 minute value) sent from the smart meter SM was converted to an instantaneous value (1 minute value) using the following formula.

[Equation 5]
Instantaneous value (winter peak) = 30-minute value x 1.425 + 0.149 (kW) ... Formula 5
[Equation 6]
Instantaneous value (summer peak) = 30-minute value x 1.347 + 0.050 (kW) ... Formula 6

Using the information obtained above, the sum of the instantaneous values for each hour and each transformer was calculated, and learning data was generated using the maximum value in a specified period as the objective variable (maximum power for each transformer in a specified period).

= Consumer Information =
FIG. 8A shows an example of the customer information 111. As shown in the figure, the customer information 11
1 includes a main switch capacity 111a, an instrument ID 111b, a business establishment code 111c, and a business establishment name 11
1d, transformer pole number 111e, lead-in pole number f, contract type code 111g, contract classification 1
11h, contracted power (subordinate B) 111i, supply method code 111j, supply method 111k, presence or absence of electric water heater 111l, electric water heater capacity (heat pump) 111m, number of electric water heaters (heat pump) 111n, hot water heater capacity 111o, number of hot water heaters 111p, presence or absence of renewable energy 111q,
The consumer information 111 is made up of a plurality of records (entries) including information on each item such as latitude 111r, longitude 111s, and location of use 111t. One record of the consumer information 111 corresponds to one consumer (one measuring device (smart meter) or one main switch MS).

Among these, the main switch capacity 111a stores information indicating the capacity of the main switch MS installed in the consumer premises 2. The main switch capacity 111a is acquired, for example, when a new installation, replacement, inspection, etc. is performed. In addition, as the main switch capacity 111a, for example, a capacity estimated from the material of the service line supplying power to the consumer premises 2 may be used.

The meter ID 111b is a power meter (smart meter S) installed in the consumer 2.
The instrument ID, which is an identifier of the instrument (M), is stored.

The business establishment code 111c stores a business establishment code that is an identifier of the business establishment of an electric utility such as an electric power company that supplies power to the consumer 2. The business establishment name 111d stores the name of the business establishment.

The transformer pole utility pole number 111e stores a transformer pole utility pole number that is an identifier of the transformer pole that supplies power to the customer 2. The service pole number f stores the number of the service pole to which the service line to the customer 2 is connected.

The contract type code 111g stores a contract type code, which is information indicating the type of the contract that the consumer 2 has concluded with a power supply business operator such as a power company.
Contract classification, which is information indicating the type of classified contract, is stored. Contract Power (Sub B) 111
In i, the contracted power when the contract classification is "metered lighting B" is stored.

The supply method code 111j stores a supply method code (1: single-phase two-wire system, 2: single-phase three-wire system, etc.) that is a code indicating the method of supplying power to the consumer 2. The supply method 111k stores information indicating the supply method (single-phase two-wire system or single-phase three-wire system, etc.).

The electric water heater presence/absence 111l stores information (presence/absence) indicating whether or not an electric water heater is installed in the consumer 2. The electric water heater capacity (heat pump) 111m stores information indicating the capacity of a heat pump type electric water heater installed in the consumer 2. The number of electric water heaters (heat pump) 111n stores the number of heat pump type electric water heaters installed in the consumer 2.

The water heater capacity 111o stores information indicating the capacity of the water heater installed in the consumer premises 2. The water heater count 111p stores the number of water heaters installed in the consumer premises 2.

The renewable energy presence/absence 111q stores information indicating whether or not the consumer 2 has a generator that uses renewable energy.

The latitude 111r and the longitude 111s respectively store the latitude and longitude of a service pole to which a service line to the customer 2 is connected. The place of use 111t stores information indicating the location of the service pole to which a service line to the customer 2 is connected.

FIG. 8B shows an example of the measurement value information 112. As shown in the figure, the measurement value information 112 includes the following:
It is made up of a plurality of records each having items such as a meter ID 1121, a measurement date and time 1122, and a power value 1123. One record of the measurement value information 112 corresponds to a measurement value at a certain measurement date and time transmitted from a certain smart meter SM.

Among the above items, the meter ID of the smart meter SM is stored in the meter ID 1121. The date and time of measurement 1122 is stored as the date and time when the measurement value is measured. The power value 1123 is stored as the power value (30-minute value) measured by the smart meter SM.

=About margin=
As described above, in order to avoid overloading of the transformer 7, it is necessary to set an appropriate margin for the predicted value predicted by the load prediction model 114. In this embodiment, for the difference between the predicted value and the actual measurement value, the upper limit value (average value + standard deviation σ × 2) of the interval in which 95% of the measurement value falls and the upper limit value (average value + standard deviation σ × 3) of the interval in which 99.7% of the measurement value falls are calculated, and an approximation formula for each of the calculated upper limit values (relationship between the number of contracts (number of consumers 2) and the above difference for each transformer 7) is generated and used to determine the margin. Specific results of an investigation into the approximation formulas will be described later.

= Model learning =
FIG. 9 shows a process in which a learning data generation unit 135 and a model learning unit 140 of the load prediction device 100 generate learning data and learn a load prediction model 114 using the generated learning data (including learning of coefficients and constants in conventional equations, and learning of coefficients and constants in simple regression analysis and multiple regression analysis).
9 is a flowchart illustrating a process (hereinafter, referred to as a "model learning process S900") performed when

The timing for executing the model learning process S900 is not necessarily limited, but for example, the load prediction device 100 executes the model learning process S900 when new learning data 113 is generated, when an explicit execution instruction is received from a user via a user interface, etc. The model learning process S900 will be described below with reference to the same figure.

First, the learning data generating unit 135 receives from the user the range of measurement values to be used as learning data (
The designation (extraction conditions) of the area division, target period, transformer (bank), etc. is accepted, learning data is generated based on the measurement values corresponding to the accepted range, and the generated learning data is called learning data 1.
The value is stored as 13 (S911).

FIG. 10A shows a screen (hereinafter, referred to as a “learning data setting screen 1000”) that the learning data generation unit 135 presents to the user via a user interface when accepting the above designation.
As shown in the figure, the example learning data setting screen 1000 has a region division specification field 1010, a target period specification field 1020, a transformer (bank) specification field 1030, and a setting button 1050.

In the area category specification field 1010, an area category is set, which is information for specifying the area in which the transformer is installed. The area category specification field 1010 is in the form of a pull-down menu.
The user can specify "ALL" to specify all regions, or can specify, for example, one or more region divisions exemplified in FIG. 10B.

The period of measurement is specified in the target period specification field 1010.
is in the form of a pull-down menu, and the user can select "ALL" to specify the entire period,
One or more periods can be specified.

In the transformer (bank) specification field 1030, information for identifying the transformer (bank) (in this example, the transformer pole number) can be specified. The transformer (bank) specification field 1030 is in the form of a pull-down menu, and the user can select "AL" to specify all transformers (banks), or "B" to specify all transformers (banks).
In addition to "L", one or more transformer (bank) identifiers (such as transformer pole number) can be specified.

When the user operates the setting button 1050, the learning data generation unit 135 acquires (extracts) the measurement value corresponding to the specification content accepted on the screen from the measurement value information 112 and generates learning data. The generated learning data is stored in the storage unit 110 as learning data 113. Note that the specification field shown in the figure is merely an example, and other conditions may be specified.

9 , the model learning unit 140 then learns the load prediction model 114 based on the learning data 113 (S912). The model learning unit 140 may, for example, verify the prediction accuracy of the learned load prediction model 114. In that case, for example, the learning data 113 is classified in advance into data for learning and data for verification, and the load prediction model 114 is learned using the learning data, and the load prediction model 114 is verified using the verification data.

11 is a flowchart for explaining a process (hereinafter referred to as "load prediction process S1100") in which the load prediction device 100 predicts maximum power and calculates margins, and presents the results to a user. The load prediction process S1100 is executed at a predetermined timing (periodically, at a scheduled date and time, when an execution instruction is received from a user, etc.). The load prediction process S1100 will be explained below with reference to the same figure.

First, the prediction target transformer information acquisition unit 145 acquires prediction target transformer information 115, which is an explanatory variable to be input to the load prediction model 114, from the consumer information 111 and the measurement value information 112 (S1111). Note that information acquired from the consumer (new transformer installation information 118 shown in the figure) may be input as an explanatory variable. For example, the load (number of contracts by main switch capacity, etc.) expected when a transformer is newly installed in association with the construction of a housing complex, condominium, apartment, etc. is registered as an explanatory variable in the new transformer installation information 118. Also, for example, when a new supply is made from an existing transformer to a house or store, information on the consumer to whom the new supply is made is registered as an explanatory variable in the new transformer installation information 118. Note that in this case, the required capacity of the transformer is estimated, including the existing contract, and the result of the load prediction is used, for example, to determine whether or not the capacity of the existing transformer needs to be increased.

Next, the load prediction unit 150 inputs the prediction target transformer information 115 to the load prediction model 114 and obtains the maximum power as the output (S1112).

Next, the tolerance calculation unit 155 calculates the tolerance (S1113).

Next, the result presentation unit 160 outputs (displays) a screen (hereinafter referred to as "result presentation screen 1200") showing the maximum power calculated in S1112 and the margin calculated in S1113 on the output device 15 (display device) (S1114).

12 shows an example of a result display screen 1200. The example result display screen 1200 has a display field 1210 for the transformer pole number, a display field 1220 for the maximum power (predicted value), and a display field 1230 for the margin.
230, and a display field 1240 for the recommended transformer capacity.

The transformer pole number display field 1210 displays the transformer pole number of the transformer to be predicted (information for identifying the transformer to be predicted). The maximum power (predicted value) display field 1220 displays the predicted value of the maximum power of the transformer to be predicted, predicted by the load prediction unit 150. The margin display field 1230 displays the margin calculated by the margin calculation unit 155. The recommended transformer capacity display field 1240 automatically selects and displays a transformer capacity that is equal to or greater than the sum of the predicted value of the maximum power displayed in the maximum power (predicted value) display field 1220 and the predicted value of the margin displayed in the margin display field 1230, among the selectable transformer capacities, and is closest to said value. Note that instead of automatic selection, candidate transformer capacities may be presented to allow the user to manually select or determine the transformer capacity.

As described above in detail, according to the transformer load prediction system 1 of the present embodiment, the maximum power predicted for the transformer to be predicted can be calculated with high accuracy using the load prediction model 114. In addition, the margin can be calculated with high accuracy by the margin calculation unit 155. Then, based on the predicted maximum power and the calculated margin, the transformer capacity required for the transformer to be predicted can be determined with high accuracy. This makes it possible to accurately and appropriately select a transformer with a capacity required to reliably prevent overload. In addition, it is possible to prevent a transformer with an unnecessarily large capacity from being installed at the site, making it possible to efficiently operate the transformer (new installation, replacement, etc.).

=Verification results=
Next, various verification results performed on the transformer load prediction system 1 described above will be described.

<Comparative verification of models>
For each of the four load prediction models 114 (models a to d) shown in Table 1, four load prediction models 114 were generated by learning using learning data based on the main switch capacity and contract details (contract type, etc.) and the power values acquired from the smart meter SM, and the respective prediction results were compared.

Figures 13A and 13B are both comparative examples, with Figure 13A being a graph showing the relationship between the number of contracts and maximum power consumption based on measured values (actual values), and Figure 13B being a graph predicting the relationship between the number of contracts and maximum power consumption based on a conventional formula. Figure 13B also shows the coefficient of determination ^R2 . The maximum value of the coefficient of determination ^R2 is 1.0. The closer the value of the coefficient of determination ^R2 is to 1.0, the better the accuracy of the model is (the better the predicted value represents the actual value).

13C to 13F are graphs showing the results of predicting the relationship between the number of contracts and maximum power demand using the load prediction model 114. Of these, FIG. 13C shows the prediction results using model a (conventional method + simple regression analysis), FIG. 13D shows the prediction results using model b (simple regression analysis), and FIG. 13E shows the prediction results using model c.
FIG. 13F shows the prediction results using model d (multiple regression analysis), and FIG. 13G shows the prediction results using model d (multiple regression analysis).

Figure 13G shows a table for easily comparing the coefficients of determination ^R2 of Figures 13C to 13F. From the values of the coefficients of determination ^R2 of each model, it can be seen that high accuracy was obtained for model b (simple regression analysis) and model d (multiple regression analysis).

<Verification of Models b and d>
For each of model b (simple regression analysis) and model d (multiple regression analysis), the load prediction model 114 was trained using learning data based on measurement values over a wider period, and each prediction result was verified again.

Fig. 13H is a graph showing the prediction results by model b (simple regression analysis) trained using the above training data, and Fig. 13I is a table showing the coefficients and intercepts of each explanatory variable of model b. In this example, the coefficient of determination ^R2 was 0.715, confirming that high accuracy was obtained.

FIG. 13J is a graph showing the prediction results by model d (multiple regression analysis) trained using the above training data, and FIG. 13K is a table showing the coefficients and intercepts of each explanatory variable in model d. In this example, the coefficient of determination ^R2 is 0.810, which is even more accurate than model b. As shown in the figure, in this case, the three explanatory variables (
A multiple regression analysis was performed by adding explanatory variables (number of contracts for the supply system (single-phase three-wire 100V), total contracted power for metered electricity lighting B, and number of contracts with water heaters), and it is presumed that the addition of these explanatory variables also contributed to the improvement of accuracy. It can also be seen that by simply adding a small number of explanatory variables that have a high degree of influence on the objective variable in this way, the processing load (calculation load) can be reduced and accuracy can be improved efficiently.

<Verification of margin calculation method>
Next, in order to set the margin appropriately, the number of contracts and the difference (
The relationship between the predicted value and the actual measured value was examined.

FIG. 14A shows the relationship between the number of main switches (or the number of subscribers) for model b (simple regression analysis).
This is a graph in which the above difference is plotted for each. In this figure, the upper limit value (average value + standard deviation σ x 2) of the interval in which 95% of the predicted values fall is shown by a plot with a different pattern from the other plots. The percentage of predicted values greater than the actual values was 57%.

FIG. 14B shows the relationship between the number of main switches (or the number of customers) for model d (multiple regression analysis).
This is a graph plotting the above difference for each. In this figure, the upper limit of the range (average value + standard deviation σ x 2) in which 95% of the predicted values fall is shown by plotting with different patterns. The percentage of predicted values greater than the actual values was 59%.

FIG. 14C shows the effect of the number of main switches (or the number of subscribers) on model b (simple regression analysis).
This is a graph plotting the above difference for each. In this figure, the upper limit of the interval (average value + standard deviation σ x 3) in which 99.7% of the predicted values fall is shown by a plot with a different pattern from the other plots. The percentage of predicted values greater than the actual values was 57%.

FIG. 14D shows the relationship between the number of main switches (or the number of customers) for model d (multiple regression analysis).
This is a graph plotting the above difference for each. In this figure, the upper limit of the interval (average value + standard deviation σ x 3) in which 99.7% of the predicted values fall is shown by a plot with a different pattern from the other plots. The percentage of predicted values greater than the actual values was 59%.

Next, based on the graphs shown in FIGS. 14A to 14D, the upper limit of the interval in which 95% of the predicted value falls was approximated using four types of approximation formulas (linear approximation, exponential approximation, power approximation, and logarithmic approximation).

15A is a graph showing the results of approximating the upper limit of the interval in which 95% of the predicted value falls based on FIG. 14A using an approximation formula. FIG. 15B shows the parameters of each approximation formula,
The percentage of times the predicted values exceed the actual values, and the average of the differences are shown in table format.

As shown in Fig. 15B, the percentage of predicted values exceeding the actual values was 95% for the conventional method and 94% for the average section of each approximation formula, which was the same for both. On the other hand, the average difference was 5.58 kW when the approximation formula was not used (hereinafter referred to as the "conventional method"), while it was about 3.7 kW when the approximation formula was used, which shows that the accuracy is improved by using the approximation formula.

Fig. 15C is a graph showing the results of approximating the upper limit of the interval in which 95% of the predicted values fall using an approximation formula based on Fig. 14B, and Fig. 15D shows the above results in a table format, including the parameters of each approximation formula, the percentage of the number of times the predicted values exceed the actual values, and the average value of the above differences.

As shown in Fig. 15D, the percentage of predicted values exceeding the actual values was 95% for the conventional method and 94% for the average section of each approximation formula, which was the same for both. On the other hand, the average difference was 5.58 kW for the conventional method, while it was about 3.0 kW for all cases using the approximation formula, which shows that the accuracy is improved by using the approximation formula.

Fig. 15E is a graph showing the results of approximating the upper limit of the interval in which 99.7% of the predicted values fall based on Fig. 14C using an approximation formula, while Fig. 15F shows the above results in a table format, including the parameters of each approximation formula, the percentage of predicted values that exceed the actual values, and the average value of the above differences.

As shown in Fig. 15F, the percentage of predicted values exceeding the actual values was 95% for the conventional method, while the average of the sections for each approximation formula was 98%, meaning that the approximation formula outperformed the conventional method. On the other hand, the average difference was 5.58 kW for the conventional method, while it was about 6.0 kW for all cases using the approximation formula, meaning that the approximation formula was less accurate.

Fig. 15G is a graph showing the results of approximating the upper limit of the interval in which 99.7% of the predicted values fall based on Fig. 14D using an approximation formula, and Fig. 15H shows the above results in a table format, including the parameters of each approximation formula, the percentage of the number of times the predicted values exceed the actual values, and the average value of the above differences.

As shown in FIG. 15H, the average difference is similar between the conventional method and each approximation formula, but the percentage of predicted values exceeding the actual measured value is 95% for the conventional method and 9% for the interval average of each approximation formula.
8%, and it can be seen that the use of the approximation formula can cover more cases.

From the above, it has been confirmed that by generating an approximation equation showing the relationship between the number of main switches and the difference between the predicted value and the actual measured value, it is possible to accurately determine the above-mentioned difference, which can basically be used as a guideline for tolerance, except for the case of Figure 15F.

Although the embodiment of the present invention has been described above in detail, the above description is for the purpose of facilitating understanding of the present invention and does not limit the present invention. The present invention may be modified or improved without departing from the spirit thereof, and the present invention naturally includes equivalents thereof. For example, the above embodiment has been described in detail to explain the present invention in an easily understandable manner, and is not necessarily limited to those having all of the configurations described. In addition, it is possible to add, delete, or replace a part of the configuration of the above embodiment with another configuration.

For example, in the above, the load prediction model 114 is described as the model shown in FIG. 5 as an example, but the load prediction model 114 may be realized by other types of models, such as a Deep Neural Network (DNN) or a Support Vector Machine (SVM).

REFERENCE SIGNS LIST 1 Transformer load forecasting system, 2 Consumer, 3 System center, 4 Power information management device, 5 Communication network, 7 Transformer, 8 Power distribution system, SM Smart meter, MS Main switch, L Load, 100 Load forecasting device, 110 Memory unit, 111 Consumer information, 11
2 Measurement value information, 113 Learning data, 114 Load forecast model, 115 Information on transformer to be forecast, 116 Load forecast result, 117 Margin calculation result, 118 Information on new transformer installation, 12
0 Customer information management unit, 130 Measurement value acquisition unit, 135 Learning data generation unit, 140 Model learning unit, 145 Prediction target transformer information acquisition unit, 150 Load prediction unit, 155 Tolerance calculation unit, 160 Result presentation unit, S900 Model learning process, 1000 Learning data setting screen, S1100 Load prediction process, 1200 Result presentation screen

Claims (8)

The information processing device includes a processor and a storage device.
A load prediction model is generated, which is a model in which the number of main switches by capacity provided in each consumer receiving power supply from the transformer is used as an explanatory variable and the maximum power of the transformer is used as a target variable;
The objective variable is obtained by inputting the number of main switches by capacity provided in each consumer receiving power supply from a transformer to be predicted, which is a transformer to be predicted maximum power, into the load prediction model as an explanatory variable, and the objective variable thus obtained is output.
Transformer load forecasting system. 2. The transformer load prediction system according to claim 1,
If a main switch with an unknown capacity is provided in a consumer receiving power supply from the transformer to be predicted, the capacity of the main switch with an unknown capacity is identified based on the contents of a power supply contract of the consumer in which the main switch is provided.
Transformer load forecasting system. 3. A transformer load prediction system according to claim 1,
The load prediction model is a simple regression analysis, a multiple regression analysis, a DNN (Deep Neural Network),
and a support vector machine (SVM). 4. A transformer load prediction system according to claim 3, comprising:
In the multiple regression analysis, at least one of the number of contracts for the supply system (single-phase three-wire 100V), the total contracted power of metered electricity B, and the number of contracts with water heaters is further used as an explanatory variable.
Transformer load forecasting system. 3. A transformer load prediction system according to claim 1,
storing a relationship between the number of main switches receiving power from the transformer and an actual measured value of the maximum power of the transformer;
A relationship between the number of main switches and a predicted value of maximum power of the transformer is obtained based on the load prediction model.
generating an approximation equation showing a relationship between the number of main switches and a difference between the predicted value and the actual measured value;
outputting information indicating the difference calculated using the generated approximation formula;
Transformer load forecasting system. An information processing device having a processor and a storage device,
A step of generating a load prediction model, which is a model in which the number of main switches by capacity provided in each consumer receiving power supply from the transformer is an explanatory variable and the maximum power of the transformer is an objective variable;
a step of inputting the number of main switches by capacity provided in each consumer receiving power supply from a transformer to be predicted, which is a transformer to be predicted for maximum power, as an explanatory variable into the load prediction model to obtain the objective variable, and outputting the obtained objective variable;
A transformer load forecasting method. 7. A transformer load prediction method according to claim 6, comprising the steps of:
The transformer load prediction method further includes, when a main switch with an unknown capacity is present among the main switches installed at a consumer receiving power supply from the transformer to be predicted, a step of specifying the capacity of the main switch with an unknown capacity based on the contents of the power supply contract of the consumer in which the main switch is installed. 7. A transformer load prediction method according to claim 6, comprising the steps of:
The information processing device,
storing a relationship between the number of main switches receiving power from the transformer and an actual measured value of the maximum power of the transformer;
determining a relationship between the number of main switches and a predicted value of maximum power of the transformer based on the load prediction model;
generating an approximation equation indicating a relationship between the number of main switches and a difference between the predicted value and the actual measured value;
outputting information indicating the difference calculated using the generated approximation formula;
A transformer load forecasting method.

Images