Prediction device, prediction method, program, and power management device

Abstract

To provide a prediction device capable of improving not only prediction accuracy of transition of power demand but also prediction accuracy of a detailed state related to the power demand, a prediction method, a program, and a power management device.SOLUTION: A prediction device is configured to predict power demand, and includes a prediction unit. The prediction unit predicts power demand using a machine learning model. The machine learning model is generated by multi-task machine learning that simultaneously learns a function of predicting a daily transition in power demand and a function of predicting first sub-information. The first sub-information is information related to the power demand, and different from the daily transition of the power demand.SELECTED DRAWING: Figure 5

Description

Technical Field

[0001] The present disclosure relates to a prediction device, a prediction method, a program, and a power management device.

Background Art

[0002] There is a technology for predicting power demand in consumers. For example, a prediction device that predicts power demand using a machine learning model based on weather forecasts has been proposed (Patent Document 1, Patent Document 2). The machine learning model is generated by learning using past weather records and power demand record values. The machine learning model is configured to output a power demand prediction with a weather forecast as an input.

[0003] In addition, there is a power management device for managing the power supply cost related to the power supply to consumers using the predicted power demand. The power management device can reduce the power supply cost, for example, by creating a charge / discharge plan for a secondary battery (power storage facility) based on the discharge power unit price of the secondary battery (Patent Document 3).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Patent Document 3

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, although the above prediction device is configured to predict the trend of power demand (in other words, the overall change state of power demand), it does not always accurately predict the detailed state of power demand. For example, the above prediction device may not accurately predict the time period with the highest power demand during a day (in other words, the peak time period).

[0006] In particular, in a power management device, not only the prediction accuracy of the trend of power demand during a day but also the prediction accuracy of the peak time period of power demand will affect the management of power supply cost.

[0007] Therefore, it is desirable for the present disclosure to provide a prediction device, a prediction method, a program, and a power management device that can improve not only the prediction accuracy of the trend of power demand but also the prediction accuracy of the detailed state of power demand.

Means for Solving the Problem

[0008] A prediction device according to one aspect of the present disclosure is a prediction device configured to predict power demand, and includes a prediction unit. The prediction unit is configured to predict power demand using a machine learning model. The machine learning model is generated by multi-task machine learning that simultaneously learns a function of predicting the daily trend of power demand and a function of predicting first sub-information. The first sub-information is information related to power demand and is different from the daily trend of power demand.

[0009] Since this machine learning model is generated by multi-task machine learning, the learning effect based on the daily trend of power demand and the learning effect based on the first sub-information are reflected. Therefore, this machine learning model has improved prediction accuracy of power demand compared to a machine learning model generated by single-task machine learning based only on the function of predicting the daily trend of power demand.

[0010] Therefore, according to this prediction device, not only the prediction accuracy of the trend of power demand but also the prediction accuracy of the detailed state regarding power demand can be improved. Note that the information regarding power demand used as the first sub-information may be information that changes in relation to power demand.

[0011] Next, in the above-described prediction device, the first sub-information may be a peak time zone corresponding to the time zone in which the power demand is the largest during a day in the power demand. Such a prediction device can improve the prediction accuracy regarding the daily trend of power demand and can also improve the prediction accuracy of the peak time zone. As a result, by using the prediction result by this prediction device, appropriate measures can be implemented at the peak of power demand.

[0012] Next, in the above-described prediction device, the machine learning model may be configured to output at least the daily trend of power demand in response to the input of predetermined input information. The input information may include a weather forecast.

[0013] A prediction device equipped with such a machine learning model can predict the daily trend of power demand based on the weather forecast. Next, in the above-described prediction device, the machine learning model may be configured to predict the daily trend of power demand by dividing one day into a plurality of predetermined individual periods and predicting the power demand for each individual period.

[0014] The daily trend of power demand is not limited to the change state of power demand every short time (for example, in seconds), and may be, for example, a plurality of individual periods divided every 30 minutes (in other words, 48 individual periods). Alternatively, it may be a plurality of individual periods divided every hour (in other words, 24 individual periods).

[0015] The prediction method in another aspect of the present disclosure is a prediction method for predicting power demand using a machine learning model. The machine learning model is generated by multi-task machine learning that simultaneously learns a function for predicting the daily variation of power demand and a function for predicting first sub-information. The first sub-information is information that changes in relation to power demand and is different from the daily variation of power demand.

[0016] According to this prediction method, similar to the above-described prediction device, it is possible to improve not only the prediction accuracy of the variation of power demand but also the prediction accuracy of the detailed state regarding power demand. The program in another aspect of the present disclosure is a program that causes a computer to realize a prediction function. The prediction function is a function of predicting power demand using a machine learning model.

[0017] The machine learning model is generated by multi-task machine learning that simultaneously learns a function for predicting the daily variation of power demand and a function for predicting first sub-information. The first sub-information is information related to power demand and is different from the daily variation of power demand.

[0018] According to this program, by causing a computer to realize the prediction function, similar to the above-described prediction device, it is possible to improve not only the prediction accuracy of the variation of power demand but also the prediction accuracy of the detailed state regarding power demand.

[0019] The power management device in another aspect of the present disclosure includes a prediction device, a monitoring unit, and a simulation unit. The prediction device is the above-described prediction device, and is configured to predict predicted power demand corresponding to the predicted value of the power transmitted between the power generation device and the consumer as the power demand.

[0020] The monitoring unit is configured to monitor the dischargeable power amount corresponding to a specific type of power. The specific type of power is a specific power among the powers stored in a secondary battery capable of charging and discharging.

[0021] The simulation unit is configured to create a power supply and demand plan based on the predicted power demand during the prediction period and the dischargeable power amount at the time of plan creation. The power supply and demand plan corresponds to the plan for power demand and supply in the consumer.

[0022] In this power management device, the prediction device can accurately predict information related to power demand. Therefore, the power management device can use a predicted value of power demand with excellent prediction accuracy, can create a power supply and demand plan accurately, and can reduce the power supply cost.

[0023] In the above power management device, a specific type of power may include first generated power and second generated power. The first generated power may be power generated by a first power generation method in which the CO2 emission amount is smaller than a predetermined emission upper limit value. The second generated power may be power generated by a second power generation method in which the power generation cost is smaller than a predetermined cost upper limit value.

[0024] This power management device can suppress an increase in the emission amount of energy source CO2 while reducing the power supply cost in power supply to consumers. The above power management device may include a correction unit configured to calculate a corrected predicted power demand. The correction unit may correct the predicted power demand based on the difference value between the actual value corresponding to the measured value of power demand and the predicted power demand. The corrected predicted power demand corresponds to the predicted power demand after correction.

[0025] Since this power management device corrects the predicted power demand, it can predict the predicted value of power demand more accurately. Therefore, the power management device can create a power supply and demand plan with higher accuracy and reduce the power supply cost.

Brief Description of the Drawings

[0026]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Mode for Carrying Out the Invention

[0027] Hereinafter, embodiments to which the present disclosure is applied will be described with reference to the drawings. Note that the present disclosure is not limited to the following embodiments, and it goes without saying that various forms can be adopted as long as they belong to the technical scope of the present disclosure.

[0028] [1. First Embodiment] [1-1. Overall Configuration] As shown in FIG. 1, the power control system 1 includes a power supply and demand system 11, a power management device 31, a data collection device 51, and a weather forecast system 53.

[0029] The power supply and demand system 11 includes a first power supply and demand system 11a and a second power supply and demand system 11b. The first power supply and demand system 11a is provided separately from the second power supply and demand system 11b. The first power supply and demand system 11a and the second power supply and demand system 11b each include similar components. Therefore, the first power supply and demand system 11a will be described, and the description of the second power supply and demand system 11b will be omitted.

[0030] As shown in FIG. 2, the first power supply and demand system 11a includes a renewable energy power generation unit 13, a power conditioner 15, a secondary battery 17, a bidirectional power conversion device 19, a grid power network 21, a load facility 23, a first power system 25, and a second power system 27.

[0031] The first power system 25 is configured to transmit AC power. The first power system 25 is connected to the bidirectional power conversion device 19, the grid power network 21, and the load facility 23. The second power system 27 is configured to transmit DC power. The second power system 27 is connected to the bidirectional power conversion device 19, the power conditioner 15, and the secondary battery 17.

[0032] The renewable energy power generation unit 13 includes a power generation facility configured to generate power using renewable energy. The renewable energy power generation unit 13 generates first generated power PW1 which is power of a DC voltage. Examples of the power generation facility include a solar power generation facility, a wind power generation unit, a hydroelectric power generation facility, and a geothermal power generation facility.

[0033] The power conditioner 15 is connected to the renewable energy power generation unit 13. The power conditioner 15 includes a voltage device that converts the voltage value of the first generated power PW1 into a voltage value that can be used for charging the secondary battery 17. In the power supply and demand system 11, stable power with small voltage fluctuations can be supplied to the secondary battery 17 by voltage conversion by the power conditioner 15.

[0034] The secondary battery 17 includes a battery configured to be chargeable and dischargeable. The secondary battery 17 stores a predetermined amount of power by being charged with power supplied from the outside. The secondary battery 17 can store power using, for example, the first generated power PW1 generated by the renewable energy power generation unit 13 or the grid power PWa supplied from the grid power network 21 as a power source. The secondary battery 17 can supply power to other devices by discharging the stored power. The secondary battery 17 can supply power to the load facility 23 via, for example, the bidirectional power conversion device 19.

[0035] The bidirectional power conversion device 19 is configured to mutually convert DC power and AC power. The bidirectional power conversion device 19 is connected to each of the first power system 25 and the second power system 27.

[0036] The bidirectional power conversion device 19 can convert the AC power supplied from the first power system 25 into DC power and supply the DC power to the second power system 27. For example, the AC power supplied by the grid power network 21 can be converted into DC power by the bidirectional power conversion device 19 and used for charging the secondary battery 17. The bidirectional power conversion device 19 can convert the DC power supplied from the second power system 27 into AC power and supply the AC power to the first power system 25. For example, the DC power supplied by the secondary battery 17 can be converted into AC power by the bidirectional power conversion device 19 and used for operating the load equipment 23.

[0037] The grid power network 21 is configured to supply AC power. The grid power network 21 includes, for example, a thermal power plant, a hydroelectric power plant, a nuclear power plant, etc. The grid power network 21 may supply, for example, AC 100V.

[0038] The load equipment 23 includes devices that operate when power is supplied. The load equipment 23 receives power from the first power system 25. The load equipment 23 may include, for example, air conditioning equipment, lighting equipment, etc.

[0039] As shown in FIG. 1, the power management device 31 includes a demand prediction device 39b configured to predict the power demand. The power management device 31 is configured to create a power supply and demand plan PD in the power supply and demand system 11 using the predicted power demand. Details of the power management device 31 will be described later.

[0040] The data collection device 51 is configured to collect and store various types of data (hereinafter also referred to as power data DE) in the power supply and demand system 11. The power data DE includes information on past power generation performance (hereinafter also referred to as power generation performance information DF), information on past power demand performance (hereinafter also referred to as power demand information DM), and the like. The data collection device 51 is provided with a non-volatile storage medium (such as a hard disk drive (HDD), a solid state drive (SSD), etc.) for storing data. The data collection device 51 is configured to read out data corresponding to a request from a storage medium in response to a request from an external device and transmit the read data to the external device.

[0041] The weather forecast system 53 is configured to hold information related to the weather (hereinafter also referred to as weather information DW). The weather forecast system 53 is configured to transmit weather information DW corresponding to a request in response to a request from an external device. The weather information DW includes weather information (such as sunny, cloudy, rainy, snowy, etc.), sunshine duration, temperature information, wind speed information, and the like. The weather information DW includes future weather information DW (hereinafter also referred to as weather forecast information DW1) and actual weather information DW in the past and present (hereinafter also referred to as weather performance information DW2). The weather forecast system 53 has a function of predicting weather forecast information DW1 based on a known prediction method.

[0042] [1-2. Power Management Device] As shown in FIG. 1, the power management device 31 includes a data collection function group 33 and a charge and discharge control function group 35.

[0043] The data collection function group 33 includes a display unit 33a, a collection unit 33b, and a database storage unit 33c. The display unit 33a is configured to display data according to a command from a user of the power control system 1. When a data display request command is input from the user via an operation unit (not shown), the display unit 33a displays data corresponding to the data display request command. The display unit 33a may display data according to a predetermined schedule command instead of a command from the user. The schedule command may include information regarding the correspondence relationship between the set time and the data. The display unit 33a may display the corresponding data when the set time is reached. The display unit 33a may include a liquid crystal display, an organic EL display, a cathode ray tube display, or the like.

[0044] The collection unit 33b is configured to acquire various power data DE from the data collection device 51. For example, the collection unit 33b can transmit a period command representing a past specific period to the data collection device 51 and acquire power data DE (power generation performance information DF, power demand information DM) corresponding to the specific period.

[0045] The database storage unit 33c is configured to store various information. The various information includes the power data DE acquired from the data collection device 51. The database storage unit 33c includes a non-volatile storage medium (such as a hard disk drive (HDD), a solid state drive (SSD), etc.) for storing various information. The database storage unit 33c is configured to store the power data DE acquired from the data collection device 51 in the storage medium according to a command from the collection unit 33b. The database storage unit 33c is configured to read out the power data DE corresponding to the request from the storage medium according to a request from the collection unit 33b and transmit the read power data DE to the collection unit 33b.

[0046] The charge / discharge control function group 35 includes a weather information acquisition unit 37, a prediction unit 39, a simulation unit 41, a control unit 43, and a monitoring unit 45. The weather information acquisition unit 37 is configured to receive weather information DW from the weather forecasting system 53. The weather information acquisition unit 37 requests the weather forecasting system 53 for weather information DW (weather forecast information DW1) by specifying a prediction period, and receives the weather information DW (weather forecast information DW1) corresponding to the prediction period from the weather forecasting system 53. The prediction period may be set to any period, for example, a period up to 24 hours after the current time, a period up to 48 hours after the current time, a period up to 60 hours after the current time, a period up to 24:00 on the same day as the current time, and so on.

[0047] The prediction unit 39 includes a power generation prediction unit 39a and a demand prediction device 39b. The power generation prediction unit 39a includes a predicted power generation amount calculation unit 39a1 and a correction unit 39a2. The predicted power generation amount calculation unit 39a1 calculates a predicted value of the power generation amount of the renewable energy power generation unit 13 during the prediction period (hereinafter, also referred to as the predicted power generation amount WD1) based on the weather forecast information DW1, the weather actual result information DW2, and the power generation actual result information DF. First, the predicted power generation amount calculation unit 39a1 receives the weather forecast information DW1 and the weather actual result information DW2 from the weather forecasting system 53 via the weather information acquisition unit 37. The predicted power generation amount calculation unit 39a1 receives the power generation actual result information DF from the data collection function group 33 (specifically, the database storage unit 33c). Then, the predicted power generation amount calculation unit 39a1 executes arithmetic processing for predicting the predicted power generation amount WD1 during the prediction period based on the weather forecast information DW1, the weather actual result information DW2, and the power generation actual result information DF using a known prediction method.

[0048] For example, when the renewable energy power generation unit 13 is a solar power generation facility, the predicted power generation amount calculation unit 39a1 may predict the power generation amount of the solar power generation facility during the prediction period based on the predicted information of the weather information and the predicted information of the sunshine duration included in the weather forecast information DW1. When the renewable energy power generation unit 13 is a wind power generation facility, the predicted power generation amount calculation unit 39a1 may predict the power generation amount of the wind power generation facility during the prediction period based on the predicted information of the wind direction included in the weather forecast information DW1.

[0049] The predicted power generation amount calculation unit 39a1 may divide the prediction period into a plurality of individual periods, calculate an individual predicted power generation amount WGi for each individual period, and calculate the total value of all the individual predicted power generation amounts WGi (i is an integer of 1 or more) as the predicted power generation amount WD1. The predicted power generation amount WD1 obtained by summing up n individual periods can be expressed as in Equation (1) using the individual predicted power generation amount WGi.

[0050]

Equation

[0051] Regarding the individual periods, for example, as shown in the upper region of FIG. 3, with the operation execution time t of the predicted power generation amount WD1 as a reference point, a plurality of periods of one hour each starting from the exact hour after the operation execution time t may be set as individual periods respectively. In FIG. 3, the period from the operation execution time t to 24:00 on the same day is the prediction period. The predicted power generation amount calculation unit 39a1 may calculate an individual predicted power generation amount WGi for each of the plurality of individual periods, and calculate the total value of all the individual predicted power generation amounts WGi as the predicted power generation amount WD1. In FIG. 3, among the plurality of individual predicted power generation amounts WGi, those with a power generation amount exceeding 0 are from the individual predicted power generation amount WG1 to the individual predicted power generation amount WG6. Therefore, the predicted power generation amount WD1 is the total value of the individual predicted power generation amounts from WG1 to WG6 (=WG1 + WG2 + WG3 + WG4 + WG5 + WG6).

[0052] The correction unit 39a2 corrects the predicted power generation amount WD1 to calculate the corrected predicted power generation amount WD1 (hereinafter, the corrected predicted power generation amount WD11). The correction unit 39a2 first calculates a difference value DP (=WD12 - PWd) between the predicted power generation amount WD1 predicted in the past (hereinafter, also referred to as the past predicted power generation amount WD12) and the actual power generation amount PWd at the same time on the same day. This difference value DP corresponds to the error between the past predicted power generation amount WD12 and the actual power generation amount PWd (hereinafter, also referred to as the power generation prediction error ER).

[0053] The correction unit 39a2 executes a process of estimating a distribution function of the difference value DP (in other words, the power generation prediction error ER) using the power generation performance information DF received from the data collection function group 33. As a result, for example, a distribution function as shown in FIG. 4 is obtained. In FIG. 4, in the coordinate plane with the horizontal axis being the difference value DP and the vertical axis being the probability density, the error distribution is represented. As the method used for estimating the distribution function, a known method can be utilized, and for example, kernel density estimation or the like may be used. The correction unit 39a2 determines a desired error rate ER1 from the obtained distribution function. The desired error rate ER1 is determined using the quality level expected by the user, such as the lower probability corresponding to, for example, the standard deviation 3σ. Information such as the distribution function, the error distribution, and the error rate ER1 is stored (memorized) in the data collection function group 33.

[0054] Next, the correction unit 39a2 executes a process of calculating a corrected predicted power generation amount WD11 using the predicted power generation amount WD1 in the prediction period to be predicted this time and the determined error rate ER1. For example, a value obtained by multiplying the predicted power generation amount WD1 by the error rate ER1 may be set as the error value ER2, and a value obtained by subtracting the error value ER2 from the predicted power generation amount WD1 may be calculated as the corrected predicted power generation amount WD11 (= WD1 - ER2).

[0055] In this way, the power generation prediction unit 39a is configured to calculate the corrected predicted power generation amount WD11 in the prediction period. The demand prediction device 39b predicts the power demand PWc in the prediction period using the machine learning model MLM1.

[0056] As shown in FIG. 5, the machine learning model MLM1 is configured to take a predetermined input information X as an input and output at least one output information Y. In the present embodiment, the input information X includes a weather forecast. In the present embodiment, the output information Y includes a predicted value Y1 of the daily transition of the power demand. Further, the output information Y includes a predicted value Y2 of the peak time zone of the power demand. The machine learning model MLM1 outputs the predicted value Y1 (in other words, the daily transition of the power demand) as the predicted demand PWg.

[0057] The machine learning model MLM1 includes a time series prediction layer F1, a first fully-connected layer F2a, and a second fully-connected layer F2b. The time series prediction layer F1 is configured to perform an operation of predicting predetermined time series information based on the input information X. The time series prediction layer F1 transmits the prediction result of the predicted time series information to each of the first fully-connected layer F2a and the second fully-connected layer F2b. The first fully-connected layer F2a is configured to perform an operation of predicting a predicted value Y1 (in other words, the daily trend of power demand). The second fully-connected layer F2b is configured to perform an operation of predicting a predicted value Y2 (in other words, the peak time zone of power demand) based on the time series information.

[0058] The machine learning model MLM1 is generated by performing learning using past weather forecasts and past power demand records as teacher data. The past power demand records include the record of the daily trend of power demand and the record of the peak time zone of power demand. That is, the machine learning model MLM1 is generated by multi-task machine learning that simultaneously learns the function of predicting the daily trend of power demand and the function of predicting the peak time zone of power demand.

[0059] The machine learning model MLM1 is configured to predict the daily trend of power demand by dividing one day into a plurality of predetermined individual periods and predicting the power demand for each individual period. The length of each individual period can be arbitrarily set, such as every 30 minutes or every hour. For example, one day (24 hours) may be divided into 48 individual periods every 30 minutes, or one day may be divided into 24 individual periods every hour.

[0060] The demand prediction device 39b predicts, as the power demand PWc during the prediction period, at least the power reception demand PWcp, the first demand PWc1, and the second demand PWc2. The power reception demand PWcp corresponds to the power demand PWc supplied from the grid power network 21 to the first power supply and demand system 11a and the second power supply and demand system 11b. The first demand PWc1 corresponds to the power demand PWc required by the load equipment 23 of the first power supply and demand system 11a. The second demand PWc2 corresponds to the power demand PWc required by the load equipment 23 of the second power supply and demand system 11b.

[0061] The demand prediction device 39b predicts, as the power reception demand PWcp, at least the power reception demand PWcp of the first power supply and demand system 11a (hereinafter also referred to as the first power reception demand PWcp1) and the power reception demand PWcp of the second power supply and demand system 11b (hereinafter also referred to as the second power reception demand PWcp2).

[0062] The demand prediction device 39b may include an individual machine learning model MLM1 in order to predict the power reception demand PWcp, the first demand PWc1, and the second demand PWc2 respectively. The demand prediction device 39b may include an individual machine learning model MLM1 in order to predict the first power reception demand PWcp1 and the second power reception demand PWcp2 respectively.

[0063] The monitoring unit 45 executes a process of specifying the dischargeable power amount PA1 among the power amounts stored in the secondary battery 17. The dischargeable power amount PA1 is the power amount corresponding to a specific type of power among the power stored in the secondary battery 17. The specific type of power includes the first generated power and the second generated power. The first generated power corresponds to the power generated by the first power generation method. The first power generation method is a power generation method in which the CO2 emission amount is smaller than a predetermined emission upper limit value. Examples of the first power generation method include solar power generation, wind power generation, hydraulic power generation, geothermal power generation, and the like. The second generated power corresponds to the power generated by the second power generation method. The second power generation method is a power generation method in which the power generation cost is smaller than a predetermined cost upper limit value. The power generation cost in each power generation method may be the power generation cost on a predetermined date of each month, or may be the power generation cost at the time of calculating the dischargeable power amount PA1 by the monitoring unit 45. The cost upper limit value may be a fixed value or a value arbitrarily set by the user.

[0064] The dischargeable power amount PA1 varies, for example, as shown in the lower region of FIG. 3, in accordance with the change in the state of charge (SOC) of the secondary battery 17. In FIG. 3, the specific type of power is represented by Wen (hereinafter also referred to as specific power Wen), and the power that is not of the specific type is represented by Wdi (hereinafter also referred to as non-specific power Wdi). When the rated capacity of the secondary battery 17 is Wc and the state of charge at the time of executing the calculation is SOC, the dischargeable power amount PA1 is calculated using Equation (2).

[0065]

Equation

[0066] Note that the monitoring unit 45 monitors the charging and discharging status of the secondary battery 17, and records the specific power Wen and the specific external power Wdi over time. Further, the monitoring unit 45 records the state of charge SOC of the secondary battery 17. The monitoring unit 45 can calculate the dischargeable power amount PA1 based on equation (2) using the specific power Wen, the specific external power Wdi, the state of charge SOC, and the rated capacity Wc. The monitoring unit 45 repeatedly calculates and records the dischargeable power amount PA1 over time, thereby recording the time-series data of the dischargeable power amount PA1. Therefore, the monitoring unit 45 can specify the dischargeable power amount PA1 at the operation execution time t.

[0067] The dischargeable power amount PA1 is the power amount generated by a power generation method with a CO2 emission amount smaller than the emission upper limit value, or the power amount generated by a power generation method with a generation cost smaller than the cost upper limit value. Therefore, when a consumer desires to reduce the power consumption cost and / or suppress an increase in the emission amount of energy source CO2, the consumer's desire can be met by using the dischargeable power amount PA1.

[0068] The simulation unit 41 is configured to create a power supply and demand plan PD for a prediction period. The power supply and demand plan PD corresponds to the plan of power demand and supply in the power supply and demand system 11. First, the simulation unit 41 receives the corrected predicted power generation amount WD11 and the received power demand amount PWcp from the prediction unit 39, and receives the dischargeable power amount PA1 from the monitoring unit 45.

[0069] The simulation unit 41 calculates a dischargeable amount Wa based on the corrected predicted power generation amount WD11 in the prediction period and the dischargeable power amount PA1 at the time of plan creation (in other words, the operation execution time t). The dischargeable amount Wa can be calculated using equation (3). The dischargeable amount Wa is the power amount that can be supplied by the renewable energy power generation unit 13 and the secondary battery 17 in the prediction period, and is the power generated by the first power generation method or the second power generation method.

[0070]

Equation

[0071] The simulation unit 41 creates a power supply and demand plan PD so as to reduce the electricity usage fee based on the dischargeable amount Wa during the prediction period and the power reception demand amount PWcp during the prediction period.

[0072] For example, as shown in FIG. 6, in the predicted waveform of the power reception demand amount PWcp, by suppressing the peak value of the power reception demand amount PWcp from exceeding the fee boundary value Th1, an increase in the electricity usage fee due to the excess of the fee boundary value Th1 can be suppressed. Therefore, the simulation unit 41 creates a power supply and demand plan PD so as to utilize the dischargeable amount Wa in the time zone predicted that the power reception demand amount PWcp will exceed the fee boundary value Th1. According to such a power supply and demand plan PD, it is possible to suppress the power reception demand amount PWcp from exceeding the fee boundary value Th1 and suppress an increase in the electricity usage fee. The simulation unit 41 may create the power supply and demand plan PD using known peak shaving.

[0073] In FIG. 6, the waveform of the dischargeable amount Wa, which is the total value of the corrected predicted power generation amount WD11 calculated by the correction unit 39a2 and the dischargeable power amount PA1 calculated by the monitoring unit 45, is illustrated as "Wa (after correction)". Further, the waveform of the total value of the predicted power generation amount WD1 calculated by the predicted power generation amount calculation unit 39a1 and the dischargeable power amount PA1 calculated by the monitoring unit 45 is illustrated as "Wa (before correction)". That is, the waveform of "Wa (before correction)" corresponds to the waveform obtained by removing the effect of the correction by the correction unit 39a2 from the dischargeable amount Wa. FIG. 6 shows an example in which the dischargeable amount Wa after correction has a reduced power amount compared to the dischargeable amount Wa before correction by the correction unit 39a2.

[0074] The control unit 43 is a computer system that executes various processes in the power management device 31. The various processes are executed by operating each unit provided in the power management device 31. The control unit 43 is a computer system having a central processing unit (CPU), a ROM, a RAM, a hard disk, an input / output interface, etc. The CPU, ROM, RAM, hard disk, and input / output interface are not shown in the figure.

[0075] The control program stored in the ROM or the like causes the CPU to function at least as an instruction unit for performing calculations on various information, causes the input / output interface or the like to function as an acquisition unit and an output unit, and causes the hard disk or the like to function as a storage unit in which various information input from the outside is stored. Further, the program stored in the storage device such as the above-mentioned ROM causes the CPU, ROM, RAM, input / output interface, etc. to cooperate, and causes the control unit 43 (in other words, the computer) to function as a predicted power generation amount calculation unit 39a1, a correction unit 39a2, a demand prediction device 39b, a monitoring unit 45, and a simulation unit 41.

[0076] [1-3. Processes Executed in the Power Control System] Next, a series of processes for creating the power supply and demand plan PD in the power control system 1 will be described.

[0077] As shown in FIG. 7, first, in the data collection device 51, the processes of S110, S120, and S130 are executed respectively. The data collection device 51 is provided with a data acquisition unit and a control unit (not shown) and is configured to execute various processes.

[0078] In S110, the data collection device 51 executes a process of acquiring power supply and demand system information from the power supply and demand system 11. The power supply and demand system information corresponds to the above-mentioned power data DE.

[0079] In S120, the data collection device 51 executes a process of acquiring multimeter information from a multimeter (not shown). The multimeter is installed at each part of the power supply and demand system 11 and is configured to measure current values, voltage values, active power, reactive power, frequency, power factor, demand power, demand current, etc. at the installation location. The multimeter information includes various information measured by the multimeter (hereinafter also referred to as multimeter information).

[0080] In S130, the data collection device 51 executes a process of acquiring weather information DW from the weather forecast system 53. The data collection device 51 is configured to acquire the weather information DW from the weather forecast system 53 via the power management device 31. The data collection device 51 may be directly connected to the weather forecast system 53 without going through the power management device 31 to acquire the weather information DW.

[0081] In the power management device 31, the database storage unit 33c of the data collection function group 33 executes a process of storing (memorizing) various information acquired by the data collection device 51 in S140. Also, in S140, the data collection function group 33 executes a process of reading out the stored various information and transmitting it to the prediction unit 39, etc.

[0082] The demand prediction device 39b of the prediction unit 39 executes a process of predicting the power reception demand amount PWcp during the prediction period, the first power reception demand amount PWcp1 during the prediction period, and the second power reception demand amount PWcp2 during the prediction period, respectively, in S150.

[0083] The demand prediction device 39b executes a process of predicting the first demand amount PWc1 during the prediction period in S160. The power generation prediction unit 39a of the prediction unit 39 executes a process of calculating the corrected predicted power generation amount WD111 of the first power supply and demand system 11a among the corrected predicted power generation amounts WD11 during the prediction period (hereinafter also referred to as the first corrected power generation amount WD111) in S170.

[0084] The demand prediction device 39b executes a process of predicting the second demand amount PWc2 during the prediction period in S180. The power generation prediction unit 39a of the prediction unit 39 executes a process of calculating the corrected predicted power generation amount WD112 of the second power supply and demand system 11b (hereinafter also referred to as the second corrected power generation amount WD112) among the corrected predicted power generation amounts WD11 during the prediction period in S190.

[0085] The simulation unit 41 executes a process of creating a power supply and demand plan PD (hereinafter also referred to as the first power supply and demand plan PD1) during the prediction period of the first power supply and demand system 11a in S200. First, the simulation unit 41 calculates the dischargeable amount Wa (hereinafter also referred to as the first dischargeable amount Wa1) of the first power supply and demand system 11a according to the above-described procedure. The simulation unit 41 creates the first power supply and demand plan PD1 so as to reduce the electricity usage fee using the first received power demand PWcp1 and the first dischargeable amount Wa1.

[0086] The simulation unit 41 executes a process of creating a power supply and demand plan (hereinafter also referred to as the second power supply and demand plan PD2) during the prediction period of the second power supply and demand system 11b in S210. First, the simulation unit 41 calculates the dischargeable amount Wa (hereinafter also referred to as the second dischargeable amount Wa2) of the second power supply and demand system 11b according to the above-described procedure.

[0087] Furthermore, the simulation unit 41 calculates the change amount CF of the first received power demand PWcp1 before and after creating the first power supply and demand plan PD1 in S200. When the change amount CF is a positive value, the amount of power corresponding to the change amount CF among the first received power demand PWcp1 that was planned to be consumed in the first power supply and demand system 11a becomes surplus. This surplus amount of power can be used in the second power supply and demand system 11b. When the change amount CF is a negative value, the amount of power corresponding to the change amount CF with respect to the first received power demand PWcp1 that was planned to be consumed in the first power supply and demand system 11a becomes a shortage. Regarding this shortage amount of power, it is desirable to also consider using a part of the second received power demand PWcp2 of the second power supply and demand system 11b.

[0088] The simulation unit 41 creates a second power supply and demand plan PD2 so as to reduce the electricity usage fee by using the second power reception demand amount PWcp2, the second dischargeable amount Wa2, and the change amount CF. The database storage unit 33c of the data collection function group 33 executes a process of storing (memorizing) the first demand amount PWc1 predicted in S160, the second demand amount PWc2 predicted in S180, the first power supply and demand plan PD1 created in S200, and the second power supply and demand plan PD2 created in S210 in S220. Further, in S220, the data collection function group 33 executes a process of reading out the stored various information and transmitting the various information to the data collection device 51.

[0089] The data collection device 51 executes a process of changing various settings related to the first power supply and demand system 11a in S230. First, the data collection device 51 executes a process of acquiring various information related to the first power supply and demand system 11a from the data collection function group 33. The various information here includes the first demand amount PWc1 and the first power supply and demand plan PD1. The data collection device 51 changes various settings such as the power supply and demand plan related to the first power supply and demand system 11a by using the various information.

[0090] As a result, the first power supply and demand system 11a executes a power supply and demand plan based on the set content. The data collection device 51 executes a process of changing various settings related to the second power supply and demand system 11b in S240. First, the data collection device 51 executes a process of acquiring various information related to the second power supply and demand system 11b from the data collection function group 33. The various information here includes the second demand amount PWc2 and the second power supply and demand plan PD2. The data collection device 51 changes various settings such as the power supply and demand plan related to the second power supply and demand system 11b by using the various information. As a result, the second power supply and demand system 11b executes a power supply and demand plan based on the set content.

[0091] [1-4. Machine learning model of demand prediction device] The machine learning model MLM1 used in the prediction device 39b will be described. As described above, the machine learning model MLM1 is generated by performing multi-task machine learning using, as teacher data, past weather forecasts, the actual daily trend of power demand, and the actual peak time zone of power demand.

[0092] And when performing machine learning of the machine learning model MLM1, when learning the function of predicting the peak time zone of power demand, Equation (4) is used to calculate the error L between the predicted value and the actual value.

[0093]

Equation

[0094] However, N represents the number of data points in a day, D represents the actual trend of power demand (in other words, the N actual data of the actual power demand), the argsort operation returns the array indices when sorting the actual trend of power demand in descending order. P represents the top k (time normalized with 1 day as 1.0) of the predicted peak time zone.

[0095] Next, the prediction accuracy of power demand is compared using the machine learning model MLM1 and a comparative example. As the comparative example, a machine learning model (hereinafter also referred to as a comparison model) generated by single-task machine learning that learns based only on the function of predicting the daily trend of power demand is used.

[0096] Fig. 8 shows the comparison results between the predicted values predicted using the machine learning model MLM1 and the actual values regarding the daily trend of power demand. Fig. 9 shows the comparison results between the predicted values predicted using the comparison model and the actual values regarding the daily trend of power demand.

[0097] First, regarding the peak time zone of power demand, when using the machine learning model MLM1 (in other words, the model generated by multi-task machine learning), the peak time zone of power demand almost coincides between the predicted value and the actual value (see Fig. 8). In contrast, when using the comparison model (in other words, the model generated by single-task machine learning), there is an error between the predicted value and the actual value in the peak time zone of power demand (see Fig. 9).

[0098] Also, regarding the error of the peak value of power demand, when using the machine learning model MLM1, the error between the predicted value and the actual value becomes smaller compared to the case of using the comparison model (see Figs. 8 and 9).

[0099] According to these comparison results, when using the machine learning model MLM1, the prediction accuracy of power demand is improved compared to the case of using the comparison model. Next, the measurement results of comparing the prediction accuracy of the peak time zone will be described for each of the case of using the machine learning model MLM1 and the case of using the comparison model.

[0100] In this measurement, for the multiple predicted values obtained from multiple predictions, the error between the predicted value and the actual value regarding the peak time zone was measured. Specifically, one day (24 hours) was divided into 48 individual periods every 30 minutes, and the error between the peak time zone in the predicted value (the transition of power demand) of the machine learning model MLM1 and the peak time zone in the actual value was measured in units of individual periods.

[0101] As shown in Fig. 10, (a) in the case of multi-task machine learning, compared to (b) the case of single-task machine learning, the total value of the number of samples included in the top three individual periods with smaller errors in the peak time zone is larger. That is, it can be seen that the predicted value using the machine learning model MLM1 generated by multi-task machine learning is likely to obtain a predicted value with a smaller error in the peak time zone compared to the predicted value using the comparison model generated by single-task machine learning.

[0102] Also, using the measurement data of the error in the peak time zone between the above predicted value and the actual value, the ratio Pr was calculated for the peak time zone of the predicted value to be included in any of the peak time zones of the top three in terms of large power demand among the actual values. As shown in the upper part of FIG. 11, the ratio Pr is 53.3% in the case of multi-task machine learning and 52.4% in the case of single-task machine learning.

[0103] From this also, it can be seen that the predicted value using the machine learning model MLM1 generated by multi-task machine learning has higher prediction accuracy in the peak time zone than the predicted value using the comparison model generated by single-task machine learning.

[0104] Next, the measurement results of comparing the prediction accuracy of the transition of power demand for each of the case of using the machine learning model MLM1 and the case of using the comparison model will be described. In this measurement, for the transition of power demand, the root mean square error (RMSE) was calculated for the predicted value and the actual value.

[0105] As shown in the lower part of FIG. 11, the root mean square error is 0.0726 in the case of multi-task machine learning and 0.0728 in the case of single-task machine learning. A smaller root mean square error indicates that the difference between the predicted value and the actual value is smaller. That is, it can be seen that the predicted value using the machine learning model MLM1 generated by multi-task machine learning is closer to the actual value in the transition of power demand than the predicted value using the comparison model generated by single-task machine learning.

[0106] [1-5. Effect] As described above, the demand prediction device 39b predicts the power demand amount PWc in the prediction period using the machine learning model MLM1.

[0107] Since the machine learning model MLM1 is generated by multi-task machine learning, the learning effect based on the daily variation of power demand and the learning effect based on the peak time zone of power demand are reflected. Therefore, compared with the machine learning model generated by single-task machine learning based only on the function of predicting the daily variation of power demand, the prediction accuracy of power demand of the machine learning model MLM1 is improved.

[0108] Therefore, according to the demand prediction device 39b, the prediction accuracy regarding the daily variation of power demand can be improved, and the prediction accuracy of the peak time zone can be improved. Next, the machine learning model MLM1 is configured to output, as output information Y, the daily variation of power demand (predicted value Y1) and the peak time zone of power demand (predicted value Y2) in response to the input of weather forecast as input information X.

[0109] Therefore, the demand prediction device 39b can predict at least the daily variation of power demand based on the weather forecast. Next, the power management device 31 creates a power supply and demand plan using the prediction result by the demand prediction device 39b. For this reason, the power management device 31 can create a power supply and demand plan with high accuracy and can appropriately implement countermeasures at the peak of power demand.

[0110] Therefore, the power management device 31 can use a predicted value of power demand with excellent prediction accuracy and can create a power supply and demand plan with high accuracy, so that the power supply cost can be reduced. Next, the power management device 31 can suppress an increase in the emission amount of energy source CO2 while reducing the power supply cost in the power supply to each of the first power supply and demand system 11a and the second power supply and demand system 11b.

[0111] The predicted power generation amount calculation unit 39a1 can obtain a predicted power generation amount WD1 in which weather forecast information DW1, weather actual result information DW2, and power generation actual result information DF are reflected. Thereby, the power management device 31 can improve the prediction accuracy of the predicted power generation amount WD1 and can more accurately realize the reduction of the power supply cost and the suppression of the increase in the CO2 emission amount.

[0112] The correction unit 39a2 calculates the corrected predicted power generation amount WD11 using the distribution function of the difference values specified based on the past difference values DP. That is, the correction unit 39a2 can accurately calculate the corrected predicted power generation amount WD11 based on past facts. Thereby, the power management device 31 can improve the correction accuracy of the corrected predicted power generation amount WD11, and can more accurately reduce the power supply cost and suppress the increase in CO2 emissions.

[0113] Since the above-described first power generation method is a power generation method using renewable energy, the CO2 emissions of the first generated power can be reduced. Thereby, the power management device 31 can more accurately suppress the increase in CO2 emissions.

[0114] [1-6. Corresponding relationships of expressions] Here, the corresponding relationships of expressions will be described. The demand prediction device 39b corresponds to an example of the prediction device in the present disclosure, and the demand prediction device 39b that executes S150 corresponds to an example of the prediction unit in the present disclosure, and the peak time zone of the power demand corresponds to an example of the first sub-information in the present disclosure.

[0115] The first power supply and demand system 11a and the second power supply and demand system 11b correspond to an example of consumers. [2. Second Embodiment] [2-1. Overall Configuration] As a second embodiment, the second power control system 101 will be described.

[0116] As shown in FIG. 12, the second power control system 101 includes a power supply and demand system 11, a second power management device 131, a data collection device 51, and a weather forecast system 53. The second power control system 101 is different from the power control system 1 in that it includes a second power management device 131 instead of the power management device 31.

[0117] In the following description, the differences between the second power control system 101 and the power control system 1 will be mainly described, and the common points will be omitted from the description. Similar to the power management device 31, the second power management device 131 is configured to create a power supply and demand plan PD in the power supply and demand system 11.

[0118] [2-2. Second Power Management Device] As shown in FIG. 12, the second power management device 131 includes a data collection function group 33 and a second charge and discharge control function group 135.

[0119] Compared with the power management device 31, the second power management device 131 is different in that it includes at least a second charge and discharge control function group 135 instead of the charge and discharge control function group 35. The second charge and discharge control function group 135 includes a weather information acquisition unit 37, a second prediction unit 139, a simulation unit 41, a control unit 43, a monitoring unit 45, and an accuracy monitoring unit 47. Compared with the charge and discharge control function group 35, the second charge and discharge control function group 135 is different in that it includes at least a second prediction unit 139 instead of the prediction unit 39. In addition, compared with the charge and discharge control function group 35, the second charge and discharge control function group 135 is different in that it includes an accuracy monitoring unit 47. Furthermore, compared with the charge and discharge control function group 35, the processing content in the simulation unit 41 of the second charge and discharge control function group 135 is partially different.

[0120] The second prediction unit 139 includes a power generation prediction unit 39a and a second demand prediction device 139b. Compared with the prediction unit 39, the second prediction unit 139 is different in that it includes a second demand prediction device 139b instead of the demand prediction device 39b.

[0121] The second demand prediction device 139b includes a predicted demand calculation unit 139b1 and a demand correction unit 139b2. The predicted demand calculation unit 139b1 calculates a predicted value of the power demand PWc (hereinafter also referred to as the predicted demand PWg) during the prediction period using the machine learning model MLM1.

[0122] The machine learning model MLM1 is the same as that in the first embodiment, so the description thereof is omitted. The predicted demand calculation unit 139b1 calculates, as the predicted demand PWg during the prediction period, at least the predicted value of the power reception demand PWcp (hereinafter also referred to as the predicted power reception demand PWgp), the predicted value of the first demand PWc1 (hereinafter also referred to as the predicted first demand PWg1), and the predicted value of the second demand PWc2 (hereinafter also referred to as the predicted second demand PWg2).

[0123] The predicted demand calculation unit 139b1 calculates, as the predicted power reception demand PWgp, at least the predicted value of the power reception demand PWcp of the first power supply and demand system 11a (in other words, the first power reception demand PWcp1) (hereinafter also referred to as the predicted first power reception demand PWgp1). Further, the predicted demand calculation unit 139b1 calculates, as the predicted power reception demand PWgp, at least the predicted value of the power reception demand PWcp of the second power supply and demand system 11b (in other words, the second power reception demand PWcp2) (hereinafter also referred to as the predicted second power reception demand PWgp2).

[0124] The demand correction unit 139b2 corrects the predicted demand PWg to calculate the corrected predicted demand PWg (hereinafter also referred to as the corrected predicted demand PWe). When calculating the corrected predicted demand PWe during the prediction period, the demand correction unit 139b2 corrects at least the predicted power reception demand PWgp, the predicted first demand PWg1, and the predicted second demand PWg2. That is, the demand correction unit 139b2 calculates the corrected predicted power reception demand PWgp (hereinafter also referred to as the corrected predicted power reception demand PWep), the corrected predicted first demand PWg1 (hereinafter also referred to as the corrected first demand PWe1), and the corrected predicted second demand PWg2 (hereinafter also referred to as the corrected second demand PWe2).

[0125] Furthermore, the demand correction unit 139b2 corrects the predicted first power reception demand PWgp1 to calculate the corrected predicted first power reception demand PWgp1 (hereinafter, also referred to as the corrected first power reception demand PWep1). The demand correction unit 139b2 corrects the predicted second power reception demand PWgp2 to calculate the corrected predicted second power reception demand PWgp2 (hereinafter, also referred to as the corrected second power reception demand PWep2).

[0126] The demand correction unit 139b2 first calculates a demand difference value DC (= PWgc - PWcr), which is the difference between the predicted demand PWg predicted in the past (hereinafter, also referred to as the past predicted demand PWgc) and the actual power demand PWc (hereinafter, also referred to as the actual power demand PWcr) in the same time period and on the same day as the prediction period. This demand difference value DC corresponds to the error between the past predicted demand PWgc and the actual power demand PWcr (hereinafter, also referred to as the demand prediction error ERc).

[0127] The demand correction unit 139b2 executes a process of estimating the distribution function of the demand difference value DC (in other words, the demand prediction error ERc) using the power demand information DM received from the data collection function group 33. As a result, for example, in a coordinate plane with the demand difference value DC on the horizontal axis and the probability density on the vertical axis, an error distribution corresponding to the distribution function is obtained. The distribution function of the demand prediction error ERc is approximated to, for example, the distribution function shown in FIG. 4, although not shown in the figure. As a method for estimating the distribution function, a known method can be used, such as kernel density estimation. The demand correction unit 139b2 determines a desired demand error rate ERc1 from the obtained distribution function. The desired demand error rate ERc1 is determined using the quality level expected by the user, such as the lower probability corresponding to the standard deviation 3σ. Information such as the distribution function, error distribution, and demand error rate ERc1 is stored (memorized) in the data collection function group 33.

[0128] Next, the demand correction unit 139b2 executes a process of calculating a corrected predicted demand PWe using the predicted demand PWg in the prediction period to be predicted this time and the determined demand error rate ERc1. For example, a value obtained by multiplying the predicted demand PWg by the demand error rate ERc1 is set as a demand error value ERv (= PWg × ERc1), and a value obtained by subtracting the demand error value ERv from the predicted demand PWg may be calculated as the corrected predicted demand PWe (= PWg - ERv).

[0129] The demand correction unit 139b2 calculates a corrected predicted received power demand PWep, a corrected first demand PWe1, a corrected second demand PWe2, a corrected first received power demand PWep1, and a corrected second received power demand PWep2, respectively, using the same method as the method for calculating the corrected predicted demand PWe as described above.

[0130] In this way, the second demand prediction device 139b is configured to calculate the corrected predicted received power demand PWep, the corrected first demand PWe1, the corrected second demand PWe2, the corrected first received power demand PWep1, and the corrected second received power demand PWep2, respectively, as the corrected predicted demand PWe in the prediction period.

[0131] That is, the second demand prediction device 139b is different from the demand prediction device 39b in that it corrects the predicted value of the power demand PWc in the prediction period (in other words, the predicted demand PWg). And the second demand prediction device 139b is configured to transmit the corrected predicted values (for example, the corrected predicted received power demand PWep, the corrected first demand PWe1, the corrected second demand PWe2, the corrected first received power demand PWep1, the corrected second received power demand PWep2) as the predicted value of the power demand PWc in the prediction period to the simulation unit 41 and the like.

[0132] In the second embodiment, the simulation unit 41 is configured to create a power supply and demand plan PD during the prediction period, as in the first embodiment. In the second embodiment, the simulation unit 41 receives the corrected predicted power generation amount WD11, the corrected predicted power reception demand amount PWep, the corrected first demand amount PWe1, and the corrected second demand amount PWe2 from the second prediction unit 139, and receives the dischargeable power amount PA1 from the monitoring unit 45. That is, the simulation unit 41 of the second embodiment is different from that of the first embodiment in that it receives the corrected predicted power reception demand amount PWep instead of the power reception demand amount PWcp. Further, the simulation unit 41 of the second embodiment is different from that of the first embodiment in that it receives the corrected first demand amount PWe1 and the corrected second demand amount PWe2.

[0133] The simulation unit 41 of the second embodiment calculates the dischargeable amount Wa based on the corrected predicted power generation amount WD11 during the prediction period and the dischargeable power amount PA1 at the time of plan creation (in other words, the calculation execution time t), as in the first embodiment.

[0134] The simulation unit 41 of the second embodiment creates a power supply and demand plan PD so as to reduce the electricity usage fee based on the dischargeable amount Wa during the prediction period and the corrected predicted power reception demand amount PWep during the prediction period. The simulation unit 41 of the second embodiment is different from that of the first embodiment in that it uses the corrected predicted power reception demand amount PWep instead of the power reception demand amount PWcp when creating the power supply and demand plan PD.

[0135] Note that the predicted waveform of the corrected predicted power reception demand PWep in the second embodiment corresponds to the predicted waveform of the power reception demand PWcp shown in FIG. 6 in the first embodiment. By suppressing the peak value of the corrected predicted power reception demand PWep from exceeding the tariff boundary value Th1, an increase in the electricity usage charge due to the exceeding of the tariff boundary value Th1 can be suppressed. When the simulation unit 41 of the second embodiment predicts that the corrected predicted power reception demand PWep will exceed the tariff boundary value Th1, the power supply and demand plan PD is created so as to utilize the dischargeable amount Wa. According to such a power supply and demand plan PD, it is possible to suppress the corrected predicted power reception demand PWep from exceeding the tariff boundary value Th1 and suppress an increase in the electricity usage charge. Similar to the first embodiment, the simulation unit 41 may create the power supply and demand plan PD using known peak shaving.

[0136] Similar to the first embodiment, the simulation unit 41 creates a first power supply and demand plan PD1 and a second power supply and demand plan PD2 as the power supply and demand plan PD. The simulation unit 41 creates the first power supply and demand plan PD1 using the corrected first power reception demand PWep1 and the first dischargeable amount Wa1. The simulation unit 41 creates the second power supply and demand plan PD2 using the corrected second power reception demand PWep2 and the first dischargeable amount Wa1.

[0137] Next, the accuracy monitoring unit 47 is configured to execute an accuracy determination process and a fitness verification process. The accuracy determination process is a process for determining the prediction accuracy of the power generation amount (or demand amount) by the second prediction unit 139. Specifically, the prediction accuracy of the power generation amount by the second prediction unit 139 is the prediction accuracy of the predicted power generation amount WD1 by the predicted power generation amount calculation unit 39a1 of the power generation prediction unit 39a. Specifically, the prediction accuracy of the demand amount by the second prediction unit 139 is the prediction accuracy of the predicted demand amount PWg by the predicted demand amount calculation unit 139b1 of the second demand prediction device 139b.

[0138] When executing the accuracy determination process, the accuracy monitoring unit 47 first receives the actual power generation amount PWd and the predicted power generation amount WD1 during a predetermined past accuracy determination period TE from the data collection function group 33. The historical data of the actual power generation amount PWd and the predicted power generation amount WD1 are stored in the data collection function group 33 as power generation actual result information DF and power generation prediction information Dpw. The accuracy determination period TE may be set to, for example, one month, or it may be set to two months, three months, half a year, or one year.

[0139] Next, the accuracy monitoring unit 47 determines the prediction accuracy of the predicted power generation amount WD1 by the predicted power generation amount calculation unit 39a1 based on the comparison result between the actual power generation amount PWd and the predicted power generation amount WD1 during the accuracy determination period TE.

[0140] At this time, the accuracy monitoring unit 47 compares the actual power generation amount PWd and the predicted power generation amount WD1 using a known evaluation function. Examples of the evaluation function include the mean absolute percentage error (MAPE), the root mean square error (RMSE), etc. When using the mean absolute percentage error, the error E1 is calculated using equation (5). The error E1 corresponds to the prediction accuracy of the predicted power generation amount WD1 by the predicted power generation amount calculation unit 39a1.

[0141]

Equation

[0142] In equation (5), At is the actual value at time t, Ft is the predicted value at time t, and n is the total number of data in the accuracy determination period TE. The smaller the error E1, the higher the prediction accuracy of the power generation amount by the second prediction unit 139, and the larger the error E1, the lower the prediction accuracy of the power generation amount by the second prediction unit 139.

[0143] The accuracy monitoring unit 47 determines whether or not the error E1 deviates from a predetermined allowable range Ra. When the accuracy monitoring unit 47 determines that the error E1 has deviated from the allowable range Ra, it executes a process for improving the prediction method by the predicted power generation amount calculation unit 39a1. The allowable range Ra is set as a range defined by, for example, an allowable upper limit value Rau and an allowable lower limit value Rab. The allowable range Ra may be set, for example, to a range where the allowable upper limit value Rau is 5.0% and the allowable lower limit value Rab is 0%.

[0144] An example of the change state of the prediction accuracy (in other words, the error E1) over time is shown in FIG. 14. In this example, from time t1 to time t5, the error E1 is within the allowable range Ra (Rab ≦ E1 ≦ Rau), and at time t6, the error E1 is in a state of deviating from the allowable range Ra (in other words, the error E1 exceeds the allowable upper limit value Rau). This example shows the case where the allowable lower limit value Rab is 0.

[0145] Note that when executing the accuracy determination process, the accuracy monitoring unit 47 determines the prediction accuracy of the demand amount (predicted demand amount PWg) by the second prediction unit 139 by the same method as the determination of the prediction accuracy of the power generation amount described above.

[0146] Next, the goodness-of-fit test process is a process for testing the correction accuracy of the power generation amount (or demand amount) by the second prediction unit 139. Specifically, the correction accuracy of the power generation amount by the second prediction unit 139 is the correction accuracy of the predicted power generation amount WD1 by the correction unit 39a2 of the power generation prediction unit 39a. Specifically, the correction accuracy of the demand amount by the second prediction unit 139 is the correction accuracy of the predicted demand amount PWg by the demand amount correction unit 139b2 of the second demand prediction device 139b.

[0147] During the execution of the fitness test process, the accuracy monitoring unit 47 first obtains information on the error distribution (hereinafter also referred to as the first error distribution information Ba) used for correcting the predicted power generation amount WD1 in the correction unit 39a2 from the data collection function group 33 (specifically, the database storage unit 33c). The first error distribution information Ba can be represented as a waveform as shown in the upper left region of FIG. 15, for example. The first error distribution information Ba corresponds to a function (in other words, a distribution function) representing the waveform of the error distribution in FIG. 4.

[0148] Next, the accuracy monitoring unit 47 acquires the power generation performance information DF for a predetermined comparison period Tc from the data collection function group 33, and executes a process of calculating distribution information (hereinafter also referred to as the second error distribution information Bb) of the difference value DP (in other words, the power generation prediction error ER) in the comparison period Tc using the power generation performance information DF. The second error distribution information Bb can be represented as a bar graph as shown in the lower left region of FIG. 15, for example.

[0149] It should be noted that the comparison period Tc is set to a period after the period in which the power generation performance information DF used for calculating the first error distribution information Ba was collected. Next, the accuracy monitoring unit 47 tests whether the first error distribution information Ba and the second error distribution information Bb match each other. In other words, the accuracy monitoring unit 47 calculates an index (hereinafter also referred to as the fitness) indicating the degree of fitness between the first error distribution information Ba and the second error distribution information Bb. The fitness may be, for example, a test statistic ST calculated using a predetermined test method. As the test method, for example, known methods such as the chi-square test and the Kolmogorov-Smirnov test can be used. When the chi-square test is used as the test method, the chi-square value calculated by the test corresponds to the test statistic ST. As the test statistic ST, known numerical values can be used, and for example, a p-value can be used.

[0150] Whether it is suitable may be determined based on the comparison result between the calculated test statistic ST and a predetermined fitness tolerance range Rb. For example, as the fitness tolerance range Rb, a numerical range that the test statistic ST can take when the first error distribution information Ba and the second error distribution information Bb are compatible with each other may be set. The fitness tolerance range Rb may be determined based on a predetermined significance level Ls. For example, a numerical range larger than the significance level Ls may be set as the fitness tolerance range Rb.

[0151] Here, when the first error distribution information Ba has a waveform as shown in the upper left region of FIG. 15 and the second error distribution information Bb is a bar graph as shown in the lower left region of FIG. 15, when the two are superimposed, it becomes as shown in the right region of FIG. 15. In this example, when the first error distribution information Ba and the second error distribution information Bb are compared, the prediction errors (in other words, the difference values DP (power generation prediction errors DR)) when the probability density peaks are different from each other, and the change tendencies (waveforms) of the probability density (vertical axis) with respect to the prediction error (horizontal axis) are different from each other. That is, FIG. 15 shows an example where the first error distribution information Ba and the second error distribution information Bb are not compatible with each other.

[0152] When the accuracy monitoring unit 47 determines that the first error distribution information Ba and the second error distribution information Bb are not compatible with each other, it executes a process for improving the correction method by the correction unit 39a2. For example, the correction method by the correction unit 39a2 may be improved by updating the first error distribution information Ba of the correction unit 39a2 using the second error distribution information Bb. Specifically, a distribution function based on the second error distribution information Bb is estimated by a known method such as kernel density estimation, and a desired correction error rate ERa is determined from the obtained distribution function. The correction error rate ERa is determined using a quality level expected by the user, such as a lower probability corresponding to, for example, a standard deviation of 3σ. The accuracy monitoring unit 47 can improve the correction method by the correction unit 39a2 by overwriting the error rate ER1 in the correction unit 39a2 with the correction error rate ERa.

[0153] Note that when the conformity test process is executed, the accuracy monitoring unit 47 detects the correction accuracy of the demand amount (predicted demand amount PWg) by the second prediction unit 139 by the same method as the test of the correction accuracy of the power generation amount.

[0154] [2-3. Power supply and demand plan creation process] Next, a series of processes for creating a power supply and demand plan PD in the second power control system 101 will be described.

[0155] As shown in FIG. 13, a series of processes in the second power control system 101 include parts common to a series of processes in the power control system 1 of the first embodiment, and some processes such as S150, S160, S180, S200, and S210 are different. Therefore, the description will focus on the different processes.

[0156] In S150, the second demand prediction device 139b corrects the received power demand amount PWcp, the first received power demand amount PWcp1, and the second received power demand amount PWcp2 in the prediction period, respectively, and calculates the corrected predicted received power demand amount PWep, the corrected first received power demand amount PWep1, and the corrected second received power demand amount PWep2 in the prediction period, respectively.

[0157] In S160, the second demand prediction device 139b corrects the predicted first demand amount PWg1 to calculate the corrected first demand amount PWe1. Then, the second demand prediction device 139b transmits the corrected first demand amount PWe1 to the simulation unit 41 instead of the database storage unit 33c.

[0158] In S180, the second demand prediction device 139b corrects the predicted second demand amount PWg2 to calculate the corrected second demand amount PWe2. Then, the second demand prediction device 139b transmits the corrected second demand amount PWe2 to the simulation unit 41 instead of the database storage unit 33c.

[0159] In S200, the simulation unit 41 executes a process of creating a power supply and demand plan PD (in other words, the first power supply and demand plan PD1) for the prediction period of the first power supply and demand system 11a. First, similar to the first embodiment, the simulation unit 41 calculates the dischargeable amount Wa (in other words, the first dischargeable amount Wa1) of the first power supply and demand system 11a. The simulation unit 41 creates the first power supply and demand plan PD1 so as to reduce the electricity usage fee, using the corrected first power reception demand amount PWep1 and the first dischargeable amount Wa1.

[0160] In S210, the simulation unit 41 executes a process of creating a power supply and demand plan PD (in other words, the second power supply and demand plan PD2) for the prediction period of the second power supply and demand system 11b. First, similar to the first embodiment, the simulation unit 41 calculates the dischargeable amount Wa (in other words, the second dischargeable amount Wa2) of the second power supply and demand system 11b.

[0161] Furthermore, the simulation unit 41 calculates the change amount CF of the corrected first power reception demand amount PWep1 before and after the creation of the first power supply and demand plan PD1 in S200. When the change amount CF is a positive value, the amount of electric power corresponding to the change amount CF in the corrected first power reception demand amount PWep1 that was planned to be consumed in the first power supply and demand system 11a becomes surplus. This surplus amount of electric power can be utilized in the second power supply and demand system 11b. When the change amount CF is a negative value, the amount of electric power corresponding to the change amount CF with respect to the corrected first power reception demand amount PWep1 that was planned to be consumed in the first power supply and demand system 11a becomes a shortage. Regarding this shortage amount of electric power, it is desirable to also consider using a part of the corrected second power reception demand amount PWep2 of the second power supply and demand system 11b.

[0162] The simulation unit 41 creates the second power supply and demand plan PD2 so as to reduce the electricity usage fee, using the corrected second power reception demand amount PWep2, the second dischargeable amount Wa2, and the change amount CF.

[0163] In S220, the database storage unit 33c of the data collection function group 33 executes a process of storing (memorizing) the first power supply and demand plan PD1 created in S200 and the second power supply and demand plan PD2 created in S210. Also, in S220, the data collection function group 33 executes a process of reading out the stored various types of information and transmitting the various types of information to the data collection device 51.

[0164] Note that the database storage unit 33c of the data collection function group 33 may store the corrected first demand amount PWe1 calculated in S160, the first corrected power generation amount WD111 calculated in S170, the corrected second demand amount PWe2 calculated in S180, the second corrected power generation amount WD112 calculated in S190, and the like.

[0165] In the second embodiment, the data collection device 51 executes a process of changing various settings related to the first power supply and demand system 11a in S230, similar to the first embodiment. Note that, in the second embodiment, as a difference from the first embodiment, the data collection device 51 executes a process of acquiring the corrected first demand amount PWe1 from the data collection function group 33 as various information related to the first power supply and demand system 11a instead of the first demand amount PWc1.

[0166] In the second embodiment, the data collection device 51 executes a process of changing various settings related to the second power supply and demand system 11b in S240, similar to the first embodiment. Note that, in the second embodiment, as a difference from the first embodiment, the data collection device 51 executes a process of acquiring the corrected second demand amount PWe2 from the data collection function group 33 as various information related to the second power supply and demand system 11b instead of the second demand amount PWc2.

[0167] [2-4. Accuracy determination process] Next, the accuracy determination process executed by the second power control system 101 will be described. As described above, the accuracy determination process is a process for determining the prediction accuracy of the power generation amount (or demand amount) by the second prediction unit 139. In the second power control system 101, the accuracy determination process is mainly executed by the accuracy monitoring unit 47.

[0168] The accuracy determination process is executed at each predetermined determination period Tj. The determination period Tj may be set to, for example, one month, or may be set to two months, three months, six months, or one year. The process for determining the prediction accuracy of the power generation amount in the accuracy determination process will be described.

[0169] As shown in FIG. 16, in order to determine the prediction accuracy of the power generation amount, when the accuracy determination process is started, first, in S410, the database storage unit 33c of the data collection function group 33 reads out various stored information and executes a process of transmitting it to the accuracy monitoring unit 47 and the like. As a result, the accuracy monitoring unit 47 receives the actual power generation amount PWd and the predicted power generation amount WD1 in the past predetermined accuracy determination period TE from the data collection function group 33.

[0170] In S420, the accuracy monitoring unit 47 calculates the prediction accuracy (in other words, the error E1) of the predicted power generation amount WD1 by the predicted power generation amount calculation unit 39a1 based on the comparison result between the actual power generation amount PWd and the predicted power generation amount WD1 in the accuracy determination period TE. As described above, the accuracy monitoring unit 47 compares the actual power generation amount PWd and the predicted power generation amount WD1 using a known evaluation function to calculate the error E1.

[0171] In S430, the accuracy monitoring unit 47 determines whether or not the error E1 deviates from a predetermined allowable range Ra. If it is determined (YES) that the error E1 deviates from the allowable range Ra, the process proceeds to S440. If it is determined (NO) that the error E1 does not deviate from the allowable range Ra, the process proceeds to S460.

[0172] When a YES determination is made in S430, the accuracy monitoring unit 47 transmits a command to the second prediction unit 139 to improve the prediction method. In S440, the second prediction unit 139 executes a process for improving the prediction method by the predicted power generation amount calculation unit 39a1 based on a command from the accuracy monitoring unit 47. For example, the second prediction unit 139 executes a process for reconstructing the prediction model by the predicted power generation amount calculation unit 39a1. At this time, the second prediction unit 139 reconstructs the prediction model so that the error E1 falls within the allowable range Ra. That is, the second prediction unit 139 improves the prediction method by the predicted power generation amount calculation unit 39a1 so that the prediction accuracy (in other words, the error E1) is improved and the predicted value approaches the actual value.

[0173] In S450, the accuracy monitoring unit 47 calculates the prediction accuracy of the predicted power generation amount WD1 by the improved predicted power generation amount calculation unit 39a1 in S440. By the same process as S420 described above, the accuracy monitoring unit 47 compares the actual power generation amount PWd and the predicted power generation amount WD1 using a known evaluation function to calculate the error E1.

[0174] In S460, the database storage unit 33c of the data collection function group 33 executes a process of storing (memorizing) the error E1 calculated in S420 when the negative determination (NO) is made in S430, and storing (memorizing) the error E1 calculated in S450 when the positive determination (YES) is made in S430.

[0175] Note that in the accuracy determination process, similar to the determination of the prediction accuracy of the power generation amount, a process for determining the prediction accuracy of the demand amount is executed. When determining the prediction accuracy of the demand amount, in each of the above processes in the accuracy determination process, various numerical values related to the power generation amount are replaced with various numerical values related to the demand amount, and each process is executed.

[0176] In this way, by executing the accuracy determination process, when the prediction accuracy of the power generation amount (or demand amount) by the second prediction unit 139 deviates from the allowable range, the second power control system 101 improves the prediction method. Thereby, the second power control system 101 can improve the prediction accuracy of the power generation amount (or demand amount), and can more accurately achieve the reduction of the power supply cost and the suppression of the increase in the CO2 emission amount.

[0177] [2-5. Fitness test process] Next, the fitness test process executed by the second power control system 101 will be described. As described above, the fitness test process is a process for testing the correction accuracy of the power generation amount (or demand amount) by the second prediction unit 139. In the second power control system 101, the fitness test process is mainly executed by the accuracy monitoring unit 47.

[0178] The fitness test process is executed at every predetermined test period Ts. The test period Ts may be set to, for example, half a year, or may be set to 1 month, 2 months, 3 months, or 1 year. The process for testing the correction accuracy of the power generation amount in the fitness test process will be described.

[0179] As shown in FIG. 17, in order to test the correction accuracy of the power generation amount, when the fitness test process is started, first, in S510, the database storage unit 33c of the data collection function group 33 reads out various pieces of information stored therein and executes a process of transmitting the information to the accuracy monitoring unit 47 and the like. As a result, the accuracy monitoring unit 47 receives the first error distribution information Ba and the power generation actual result information DF for the comparison period Tc from the data collection function group 33.

[0180] In S520, the accuracy monitoring unit 47 first causes the correction unit 39a2 to execute a process of calculating the second error distribution information Bb using the power generation actual result information DF for the comparison period Tc. The accuracy monitoring unit 47 calculates a test statistic ST in order to test whether the first error distribution information Ba and the second error distribution information Bb match each other. As described above, as a method for testing whether they match or not, for example, a known method such as a chi-square test or a Kolmogorov-Smirnov test is used to test the fitness.

[0181] In S530, the accuracy monitoring unit 47 shifts to S560 when it is determined (YES) that the first error distribution information Ba and the second error distribution information Bb match each other according to the test result of whether they match each other, and shifts to S540 when it is determined (NO) that they do not match.

[0182] In S540, the second prediction unit 139 executes a process for improving the correction method by the correction unit 39a2 based on a command from the accuracy monitoring unit 47. As described above, for example, the second prediction unit 139 may improve the correction method by the correction unit 39a2 by updating the first error distribution information Ba of the correction unit 39a2 using the second error distribution information Bb. In other words, the second prediction unit 139 executes a process for improving the correction method by the correction unit 39a2 so that the corrected value approaches the actual value.

[0183] In S550, the accuracy monitoring unit 47 causes the improved correction unit 39a2 in S540 to execute a process of calculating the second error distribution information Bb using the power generation actual information DF for the comparison period Tc. The accuracy monitoring unit 47 tests whether the improved second error distribution information Bb and the first error distribution information Ba match each other. At this time, similar to the process in S520, the goodness-of-fit is tested by a known method. Further, the accuracy monitoring unit 47 causes the improved correction unit 39a2 to execute a process of determining a desired error rate ER1 from the improved second error distribution information Bb.

[0184] In S560, when a negative determination (NO) is made in S530, the database storage unit 33c of the data collection function group 33 stores (memorizes) the goodness-of-fit and the error rate ER1 calculated in S550, and when an affirmative determination (YES) is made in S530, stores (memorizes) the goodness-of-fit calculated in S520.

[0185] Note that in the goodness-of-fit test process, similar to the test of the correction accuracy of the power generation amount, a process for testing the correction accuracy of the demand amount is executed. When testing the correction accuracy of the demand amount, in each of the above processes in the goodness-of-fit test process, various numerical values related to the power generation amount are replaced with various numerical values related to the demand amount, and each process is executed.

[0186] In this way, by executing the fitness test process, when the correction accuracy of the power generation amount (or demand amount) by the second prediction unit 139 decreases, the second power control system 101 improves the correction method. As a result, the second power control system 101 can improve the correction accuracy of the power generation amount (or demand amount), and can more accurately reduce the power supply cost and suppress the increase in CO2 emissions.

[0187] [2-6. Effect] As described above, the second demand prediction device 139b is configured such that the demand amount correction unit 139b2 corrects the predicted demand amount PWg predicted by the predicted demand amount calculation unit 139b1 to calculate the corrected predicted demand amount PWe.

[0188] In this way, by further correcting the predicted demand amount PWg predicted using the machine learning model MLM1 to obtain the corrected predicted demand amount PWe, the prediction accuracy of the power demand can be further improved. The second power management device 131 performs prediction and correction for each of the power generation amount and the demand amount, and creates a power supply and demand plan PD using the predicted and corrected power generation amount and demand amount.

[0189] Therefore, the second power management device 131 can reduce the power supply cost and suppress the increase in the amount of CO2 emissions from the energy source while supplying power to each of the first power supply and demand system 11a and the second power supply and demand system 11b.

[0190] The demand amount correction unit 139b2 calculates the corrected predicted demand amount PWe using the distribution function of the demand amount difference value DC specified based on the past demand amount difference value DC (in other words, the demand prediction error ERc). That is, the demand amount correction unit 139b2 can accurately calculate the corrected predicted demand amount PWe based on past facts. As a result, the second power management device 131 can improve the correction accuracy of the corrected predicted demand amount PWe, and can more accurately reduce the power supply cost and suppress the increase in CO2 emissions.

[0191] Next, in the second power management device 131, by executing the accuracy determination process, the accuracy monitoring unit 47 calculates the prediction accuracy (in other words, the error E1) of the predicted power generation amount WD1 by the predicted power generation amount calculation unit 39a1 based on the comparison result between the actual power generation amount PWd and the predicted power generation amount WD1 during the accuracy determination period TE (S420). The accuracy monitoring unit 47 determines whether or not the error E1 deviates from a predetermined allowable range Ra (S430). When the error E1 deviates from the allowable range Ra (YES determination in S430), the accuracy monitoring unit 47 sends a command to the second prediction unit 139 to improve the prediction method. Based on the command from the accuracy monitoring unit 47, the second prediction unit 139 executes a process for improving the prediction method by the predicted power generation amount calculation unit 39a1 so that the prediction accuracy (in other words, the error E1) is improved and the predicted value approaches the actual value (S440).

[0192] When the prediction accuracy of the predicted power generation amount calculation unit 39a1 (or the predicted demand amount calculation unit 139b1) deteriorates due to some factor, such a second power management device 131 can improve the prediction method by the predicted power generation amount calculation unit 39a1 (or the predicted demand amount calculation unit 139b1). Therefore, the second power management device 131 can suppress the continuous deterioration of the prediction accuracy of the predicted value by the predicted power generation amount calculation unit 39a1 (or the predicted demand amount calculation unit 139b1).

[0193] Next, in the second power management device 131, by executing the goodness-of-fit test process, the accuracy monitoring unit 47 calculates a test statistic ST in order to test whether or not the first error distribution information Ba and the second error distribution information Bb match each other (S520). The accuracy monitoring unit 47 performs a test using the test statistic ST to determine whether or not the first error distribution information Ba and the second error distribution information Bb match each other (S530). When it is determined that the first error distribution information Ba and the second error distribution information Bb do not match each other (NO determination in S530), the accuracy monitoring unit 47 sends a command to the second prediction unit 139 to improve the correction method. Based on the command from the accuracy monitoring unit 47, the second prediction unit 139 executes a process for improving the correction method by the correction unit 39a2 so that the corrected value approaches the actual value (S540).

[0194] Further, the second power management device 131 also executes a process for improving the correction method by the demand correction unit 139b2 so that the correction numerical value approaches the actual numerical value with respect to the correction accuracy of the corrected predicted demand PWe by the demand correction unit 139b2 as well.

[0195] When the correction accuracy of the correction unit 39a2 (or the demand correction unit 139b2) deteriorates due to some factor, such a second power management device 131 can improve the correction method by the correction unit 39a2 (or the demand correction unit 139b2). Therefore, the second power management device 131 can suppress the continuous deterioration of the correction accuracy of the predicted numerical value by the correction unit 39a2 (or the demand correction unit 139b2).

[0196] [2-7. Corresponding relationships of languages] Here, the corresponding relationships of languages will be described. The second demand prediction device 139b corresponds to an example of the prediction device in the present disclosure, and the second demand prediction device 139b that executes S150 corresponds to an example of the prediction unit in the present disclosure.

[0197] The correction unit 39a2 and the demand correction unit 139b2 correspond to examples of the correction unit. [3. Other embodiments] As described above, the embodiments of the present disclosure have been described. However, the present disclosure is not limited to the above embodiments, and can be implemented in various modes without departing from the gist of the present disclosure.

[0198] (a) In the above embodiment, the prediction device in the form of using the peak time zone of power demand as the first sub-information has been described. However, the prediction device of the present disclosure is not limited to such a form. For example, other information related to power demand may be used as the first sub-information to execute multi-task machine learning to generate a machine learning model. Examples of other information related to power demand include the integrated value of power demand in a specific time zone and the maximum power demand.

[0199] (b) In the above-described embodiment, the form of using two pieces of information (the daily change in power demand and the first sub-information) when executing multitask machine learning was described. However, the prediction device of the present disclosure is not limited to such a form. For example, a form of executing multitask machine learning using three or more pieces of information (the daily change in power demand, the first sub-information, the second sub-information, etc.) may also be used.

[0200] (c) In the above-described embodiment, the form in which the power control system 1 includes two power supply and demand systems was described. However, the number of power supply and demand systems provided in the power control system is not limited to two. For example, the power control system of the present disclosure may be in a form including one power supply and demand system, or may be in a form including three or more power supply and demand systems. In such a case, the power management device 31 may create a power supply and demand plan PD for each of the power supply and demand systems.

[0201] (d) The functions of one component in each of the above-described embodiments may be shared among a plurality of components, or the functions of a plurality of components may be exerted by one component. Also, a part of the configuration of each of the above-described embodiments may be omitted. Further, at least a part of the configuration of each of the above-described embodiments may be added to, replaced with, etc., the configuration of other of the above-described embodiments. Note that all aspects included in the technical idea specified from the language described in the claims are embodiments of the present disclosure.

[0202] (e) In addition to the computer system described above, the present disclosure can also be realized in various forms such as a higher-level system having the computer system as a component, a program for causing a computer to function as the computer system, a non-transitory tangible recording medium such as a semiconductor memory recording this program, and a concentration calculation method.

Explanation of Reference Numerals

[0203] 1... Power control system, 11... Power supply and demand system, 11a... First power supply and demand system, 11b... Second power supply and demand system, 13... Renewable energy power generation unit, 15... Power conditioner, 17... Secondary battery, 19... Bidirectional power conversion device, 21... Grid power network, 23... Load equipment, 31... Power management device, 33... Data collection function group, 33a... Display unit, 33b... Collection unit, 33c... Database storage unit, 35... Charge and discharge control function group, 37... Weather information acquisition unit, 39... Prediction unit, 39a... Power generation prediction unit, 39a1... Predicted power generation amount calculation unit, 39a2... Correction unit, 39b... Demand prediction device, 41... Simulation unit, 43... Control unit, 45... Monitoring unit, 47... Accuracy monitoring unit, 51... Data collection device, 53... Weather forecast system, 101... Second power control system, 131... Second power management device, 135... Second charge and discharge control function group, 139... Second prediction unit, 139b... Second demand prediction device, 139b1... Predicted demand amount calculation unit, 139b2... Demand amount correction unit.

Claims

1. A prediction device configured to predict power demand, comprising a prediction unit configured to predict the power demand using a machine learning model, wherein the machine learning model is generated by multi-task machine learning that simultaneously learns a function of predicting the daily trend of the power demand and a function of predicting first sub-information, wherein the first sub-information is information related to the power demand and is different from the daily trend of the power demand, a prediction device.

2. The prediction device according to claim 1, wherein the first sub-information is a peak time zone corresponding to the time zone when the power demand is the largest during one day among the power demands, a prediction device.

3. The prediction device according to claim 2, wherein the machine learning model is configured to output at least the daily trend of the power demand in response to input of predetermined input information, wherein the input information includes weather forecasts, a prediction device.

4. The prediction device according to claim 3, wherein the machine learning model is configured to predict the daily trend of the power demand by dividing one day into a plurality of predetermined individual periods and predicting the power demand for each individual period, a prediction device.

5. A prediction method for predicting power demand using a machine learning model, wherein the machine learning model is generated by multi-task machine learning that simultaneously learns a function of predicting the daily trend of the power demand and a function of predicting first sub-information, wherein the first sub-information is information related to the power demand and is different from the daily trend of the power demand, a prediction method.

6. A program for causing a computer to realize a prediction function of predicting power demand using a machine learning model, wherein the machine learning model is generated by multi-task machine learning that simultaneously learns a function of predicting the daily trend of the power demand and a function of predicting first sub-information, wherein the first sub-information is information related to the power demand and is different from the daily trend of the power demand, a program.

7. The prediction device according to any one of claims 1 to 4, configured to predict a predicted power demand corresponding to a predicted value of power transmitted between a power generation device and a consumer as the power demand, A monitoring unit configured to monitor a dischargeable power amount corresponding to a specific type of power among the power stored in a secondary battery capable of charging and discharging; A simulation unit configured to create a power supply / demand plan corresponding to the power supply / demand plan of the consumer based on the predicted power demand during the prediction period and the dischargeable power amount at the time of plan creation; A power management device comprising the above.

8. The power management device according to claim 7, wherein the specific type of power includes first generated power generated by a first power generation method with a CO2 emission amount smaller than a predetermined emission upper limit value, and second generated power generated by a second power generation method with a generation cost smaller than a predetermined cost upper limit value; Power management device.

9. The power management device according to claim 8, a correction unit configured to correct the predicted power demand based on a difference value between an actual value corresponding to the measured value of the power demand and the predicted power demand, and calculate a corrected predicted power demand corresponding to the corrected predicted power demand; A power management device comprising the above.

Images