Control method, control device, and program

Abstract

This control device for controlling supply of power to a load acquires a lower limit value as the confidence interval of a predicted value of power demand by the load, acquires an upper limit value as the confidence interval of a predicted value of solar power generation by a solar cell, plans the output of at least one of a fuel cell and a storage battery for supplying power corresponding to the difference between the lower limit value and the upper limit value to the load, and controls, according to the plan, the output of the at least one of the fuel cell and the storage battery.

Description

Control method, control device and program

[0001] The present disclosure relates to a technique for controlling the supply of power from a battery to a load.

[0002] In recent years, in order to achieve self-sufficiency in electricity at facilities such as factories, a technology has been developed to control the power supply from power generation devices such as solar cells and fuel cells and storage batteries to loads within the facility. This technology performs control to offset the difference between the power demand of the load and the amount of power generated by the power generation device by charging and discharging the storage batteries (hereinafter referred to as multiple battery control).

[0003] Supplying power to a power grid (hereinafter referred to as "reverse power flow") may be prohibited by grid interconnection regulations. For this reason, in multiple battery control, it is necessary to suppress the amount of power generated to avoid the occurrence of reverse power flow, but at the same time, it is also necessary to avoid excessive suppression. For example, Patent Document 1 describes a method in which a margin is added to the predicted value of power demand, and an upper limit of power generation is set based on the predicted value of power demand, the margin, and the predicted value of power generation by renewable energy power generation facilities.

[0004] However, Patent Document 1 does not consider adding a margin to the predicted value of the amount of power generated by the renewable energy power generation facility. Therefore, if the predicted value of the amount of power generated by the solar cell is smaller than the actual amount of power generated, the upper limit of the amount of power generated by the solar cell may be set higher than the upper limit that should be set. In this case, the amount of power generated by the solar cell may not be suppressed, and reverse power flow may occur.

[0005] Japanese Patent Application Laid-Open No. 2022-25698

[0006] The present disclosure has been made to solve such problems, and aims to provide a technology that can realize power supply from a battery to a load without causing reverse power flow.

[0007] A control method in one aspect of the present disclosure is a control method in a control device that controls the supply of power to a load, which obtains a lower limit value as a confidence interval for a predicted value of power demand by the load, obtains an upper limit value as a confidence interval for a predicted value of solar power generation by a solar cell, plans the output of at least one of a fuel cell and a storage battery that supplies power to the load corresponding to the difference between the lower limit value and the upper limit value, and controls the output of the at least one battery in accordance with the plan.

[0008] FIG. 1 is a diagram showing an example of the configuration of a control system. FIG. 2 is a flowchart showing an example of processing performed by a control device. FIG. 3 is a flowchart showing an example of processing to acquire a lower limit value as a confidence interval of a predicted value of power demand. FIG. 4 is a flowchart showing an example of processing to acquire an upper limit value as a confidence interval of a predicted value of photovoltaic power generation. FIG. 5 is a diagram showing an example of a plan formulated by a planner. FIG. 6 is a flowchart showing another example of processing to acquire a lower limit value as a confidence interval of a predicted value of power demand. FIG. 7 is a flowchart showing another example of processing to acquire an upper limit value as a confidence interval of a predicted value of photovoltaic power generation.

[0009] (Knowledge underlying the present disclosure) As described above, in recent years, in order to achieve self-sufficiency in electricity at facilities such as factories, multiple battery control has been implemented in which the difference between the electricity demand from the load and the amount of electricity generated by power generation devices such as solar cells and fuel cells is offset by charging and discharging storage batteries.

[0010] However, it is difficult to accurately predict the power demand from the load and the power generated by the solar cell (hereinafter referred to as photovoltaic power generation). In addition, generally, a control delay occurs between instructing a battery to output power and actually outputting the power due to the generation and input / output of a control signal instructing the battery to output the power, as well as the rise time until the instructed power is output.

[0011] For this reason, with multi-battery control, the balance between power supply and demand may not be maintained in real time. In this case, reverse power flow, which involves purchasing power from the power grid and supplying power to the power grid, may occur. Because reverse power flow may be prohibited by grid interconnection regulations, multi-battery control requires suppressing solar power generation to avoid reverse power flow. Furthermore, with multi-battery control using fuel cells, it is necessary to determine the amount of power generation of the fuel cell based on the amount of power generated by the solar cells and to command the fuel cell to generate that amount of power. In this case, it is economically rational to operate the system without suppressing solar power generation as much as possible. As such, with multi-battery control, it is necessary to suppress solar power generation to avoid reverse power flow, but at the same time, it is also necessary to avoid excessive suppression of solar power generation.

[0012] For example, the above-mentioned Patent Document 1 describes that the amount of power generated by a renewable energy power generation facility and the amount of power demand required by a load are predicted, and a margin is provided in the predicted value of the amount of power demand. ＊ , margin Mt and predicted value Pt of power generation from renewable energy power generation facilities ＊ It is described that the upper limit value Xt of the power generation amount is calculated based on the above.

[0013] Specifically, the predicted value of the power generation amount Pt ＊ The smaller the margin Mt, the lower the risk of reverse power flow (the larger the index RR for this risk). ＊ Therefore, the predicted value of the power generation amount Pt ＊ The smaller the value of the upper limit Xt of the power generation amount is set to be, the larger the value of the upper limit Xt of the power generation amount is set to be.

[0014] However, in Patent Document 1, the predicted value Pt ＊ Therefore, the predicted value of the power generation amount Pt ＊is smaller than the actual amount of power generation, the risk of reverse power flow occurring may be determined to be lower than the actual amount, and the upper limit value Xt of the amount of power generation may be set to a value greater than the upper limit value that should be set. In this case, there is a risk that reverse power flow may occur without suppressing the amount of power generation.

[0015] Therefore, the present inventors have conducted extensive research into techniques that can realize multiple battery control that does not cause reverse power flow, and have come up with the following aspects of the present disclosure.

[0016] (1) A control method according to one aspect of the present disclosure is a control method in a control device that controls the supply of power to a load, which obtains a lower limit value as a confidence interval for a predicted value of power demand by the load, obtains an upper limit value as a confidence interval for a predicted value of solar power generation by a solar cell, plans the output of at least one of a fuel cell and a storage battery that supplies power to the load corresponding to the difference between the lower limit value and the upper limit value, and controls the output of the at least one battery according to the plan.

[0017] The power corresponding to the difference between the lower limit of the confidence interval for the predicted value of power demand and the upper limit of the confidence interval for the predicted value of solar power generation indicates the minimum power predicted to be required to supply to the load. In this configuration, the output of at least one of the fuel cell and the storage battery is planned so that the minimum power predicted to be required to supply to the load is supplied to the load. Therefore, the minimum power required by the load can be met by the power output from the at least one of the batteries according to the plan.

[0018] Furthermore, with this configuration, the at least one battery is not scheduled to output power exceeding the minimum power required for the load, so it is possible to avoid charging the storage battery with photovoltaic power generation that exceeds the minimum power required for the load. This prevents the output of the solar cell and the storage battery from exceeding the power demand. As a result, it is possible to realize power supply from the solar cell and the at least one battery to the load without causing reverse power flow.

[0019] (2) In the control method described in (1) above, if the upper limit value is greater than the lower limit value, the plan may include stopping the output of the fuel cell, and further including planning charging of the storage battery by solar power generation corresponding to the difference between the upper limit value and the lower limit value, and the control may further include controlling the output of the solar cell according to the plan.

[0020] In this configuration, when the upper limit of the confidence interval for the predicted value of solar power generation is greater than the lower limit of the confidence interval for the predicted value of power demand, the storage battery is charged by solar power generation corresponding to the difference between the upper limit and the lower limit, thereby reducing the possibility of reverse power flow due to solar power generation.

[0021] (3) In the control method described in (2) above, the plan may further include acquiring the remaining amount of power in the storage battery, and determining whether the storage battery is in a predetermined state close to a fully charged state based on the remaining amount of power. If it is determined that the storage battery is in the predetermined state, a plan may be made to suppress solar power generation. If it is determined that the storage battery is in the predetermined state, the control may further control the output of the solar cell in accordance with the plan.

[0022] In this configuration, when the storage battery is being charged by solar power generation corresponding to the difference between the upper limit and the lower limit, the solar power generation is suppressed when the storage battery reaches a predetermined state close to a fully charged state. This reduces the possibility of reverse power flow occurring due to continued solar power generation even after the storage battery reaches a fully charged state.

[0023] (4) In the control method described in any one of (1) to (3) above, if the power corresponding to the difference between the lower limit value and the upper limit value is greater than the sum of the rated outputs of the fuel cell and the storage battery, the plan may further include planning to purchase from the power grid the shortage power, which is the power corresponding to the difference between the difference between the lower limit value and the upper limit value and the sum, and the control may further include supplying the shortage power purchased from the power grid to the load in accordance with the plan.

[0024] Since there is a considerable control delay in the power output of the fuel cell and the storage battery, it is difficult to obtain the rated output of the fuel cell and the storage battery at the desired timing. However, with this configuration, if the power corresponding to the difference between the lower limit value and the upper limit value is greater than the sum of the rated outputs of the fuel cell and the storage battery, the shortage of power that cannot be supplied to the load by the solar cell, fuel cell, and storage battery alone can be purchased from the power grid and the shortage of power can be appropriately supplied to the load.

[0025] (5) In the control method described in (1) above, the plan may plan the output of the fuel cell and the output of the storage battery so that the frequency of output fluctuations of the fuel cell is lower than the frequency of output fluctuations of the storage battery.

[0026] In this configuration, the outputs of the fuel cell and the storage battery are planned so that the frequency of output fluctuations of the fuel cell is lower than the frequency of output fluctuations of the storage battery. Therefore, taking into consideration that the more frequently the output fluctuations of the fuel cell, the shorter the lifespan of the fuel cell becomes, the more the outputs of the fuel cell and the storage battery can be planned to extend the lifespan of the fuel cell.

[0027] (6) In the control method described in (1) above, when obtaining the lower limit value, the current power demand may be predicted based on past actual values ​​of the power demand, a time progression of a prediction error actual value, which is the difference between the previously predicted power demand and the actual value of the power demand at the time of the prediction, a standard deviation of the prediction error actual value may be calculated based on the time progression of the prediction error actual value, and the result of subtracting the product of the standard deviation of the prediction error actual value and a predetermined coefficient from the current power demand may be obtained as the lower limit value.

[0028] According to this configuration, the result of subtracting the product of the standard deviation of the actual forecast error and a predetermined coefficient from the current power demand predicted based on the actual values ​​of past power demand can be appropriately obtained as the lower limit of the confidence interval for the predicted value of power demand.

[0029] (7) In the control method described in (1) above, when obtaining the lower limit value, the current power demand may be predicted based on past actual values ​​of the power demand, and the relationship between a prediction error actual value, which is the difference between the previously predicted power demand and the actual value of the power demand at the time of the prediction, and one or more predetermined feature values ​​corresponding to the time of the prediction, may be machine-learned through a trained model, and the current actual prediction error may be estimated by inputting the one or more predetermined feature values ​​corresponding to the time of the prediction into the trained model.The lower limit value may be obtained by subtracting the product of the estimated value of the current actual prediction error and a predetermined coefficient from the current power demand.

[0030] According to this configuration, the product of the actual prediction error at the current time, estimated by inputting one or more features corresponding to the current time into the trained model, and a predetermined coefficient, can be subtracted from the current power demand predicted based on actual values ​​of past power demand, and the result can be appropriately obtained as the lower limit value of the confidence interval for the predicted value of power demand.

[0031] (8) In the control method described in (7) above, the one or more features corresponding to the current time and the predicted time may include at least one of the following: the days of the week corresponding to the current time and the predicted time; information indicating whether the current time and the predicted time are operating days of the facility in which the load is installed; information indicating the temperature and production plan at the facility at the current time and the predicted time; and information indicating whether the production plan at the facility at the current time and the predicted time has been set.

[0032] According to this configuration, by inputting at least one of the following into the trained model: the day of the week corresponding to the current time, information indicating whether the current time is an operating day for the facility, information indicating the current temperature and production plan at the facility, and information indicating whether a current production plan has been set for the facility, the actual prediction error at the current time can be appropriately estimated.

[0033] (9) In the control method described in (1) above, in acquiring the upper limit value, the solar power generation at the current time may be predicted based on actual values ​​of the solar power generation in the past, a time transition of a prediction error actual value, which is the difference between the solar power generation predicted in the past and the actual value of the solar power generation at the time of the prediction, may be acquired, a standard deviation of the prediction error actual value may be calculated based on the time transition of the prediction error actual value, and the upper limit value may be acquired as the sum of the solar power generation at the current time and the product of the standard deviation of the prediction error actual value and a predetermined coefficient.

[0034] According to this configuration, the sum of the current solar power generation predicted based on the past actual solar power generation values ​​and the product of the standard deviation of the actual prediction error and a predetermined coefficient can be appropriately obtained as the upper limit value of the confidence interval for the predicted solar power generation value.

[0035] (10) In the control method described in (1) above, in acquiring the upper limit value, the solar power generation at the current time is predicted based on past actual values ​​of the solar power generation, and the relationship between a prediction error actual value, which is the difference between the previously predicted solar power generation and the actual value of the solar power generation at the time of the prediction, and one or more predetermined feature values ​​corresponding to the time of the prediction, is learned by machine learning, and the one or more predetermined feature values ​​corresponding to the time of the prediction are input into a trained model, where the relationship between the prediction error actual value, which is the difference between the previously predicted solar power generation and the actual value of the solar power generation at the time of the prediction, is estimated, and the upper limit value may be acquired as the sum of the solar power generation at the current time and the product of the estimated value of the prediction error actual value at the current time and a predetermined coefficient.

[0036] According to this configuration, the sum of the product of the actual prediction error at the current time, estimated by inputting one or more features corresponding to the current time into the trained model, and a predetermined coefficient, and the solar power generation at the current time, predicted based on the actual values ​​of solar power generation in the past, can be appropriately obtained as the upper limit value of the confidence interval for the predicted value of solar power generation.

[0037] (11) In the control method described in (10) above, the one or more features corresponding to the current time and the predicted time may include at least one of the probability of precipitation, solar radiation, wind speed, wind direction, and a panoramic sky image corresponding to the current time and the predicted time.

[0038] According to this configuration, by inputting at least one of the current precipitation probability, solar radiation, wind speed, wind direction, and all-sky image into the trained model, the actual forecast error at the current time can be appropriately estimated.

[0039] (12) In the control method described in any one of (6) to (11) above, the predetermined coefficient may be set to a larger value as the regulations for determining whether or not power is being supplied to the power grid in the facility where the load is installed are stricter.

[0040] In this configuration, the stricter the rule for determining whether a reverse power flow, which is a supply of power to the power grid in a facility where a load is installed, is, the larger the predetermined coefficient is set. Therefore, the stricter the rule, the smaller the lower limit of the confidence interval for the predicted value of power demand, and the larger the upper limit of the confidence interval for the predicted value of photovoltaic power generation.

[0041] This configuration reduces the minimum amount of power predicted to be required to supply to the load, which corresponds to the difference between the lower limit and the upper limit. As a result, when power demand exceeds the lower limit or when solar power generation falls below the upper limit, the stricter the rule, the greater the amount of power supplied to the load from the grid. Therefore, this configuration can appropriately prevent reverse power flow in the facility depending on the strictness of the rule.

[0042] (13) In the control method described in any one of (6) to (11) above, the predetermined coefficient may be set to a larger value as the control delay time that occurs in the control device's control of outputting power to the storage battery becomes longer.

[0043] In this configuration, the longer the control delay time that occurs in the control device's control of the storage battery to output power, the larger the predetermined coefficient is set. Therefore, the longer the control delay time, the smaller the lower limit of the confidence interval for the predicted value of power demand and the larger the upper limit of the confidence interval for the predicted value of solar power generation.

[0044] This configuration reduces the minimum amount of power predicted to be required to supply to the load, which corresponds to the difference between the lower limit and the upper limit. As a result, when power demand exceeds the lower limit or when photovoltaic power generation falls below the upper limit, the longer the control delay, the greater the power supplied to the load from the grid power. Therefore, this configuration can appropriately prevent reverse power flow in the facility depending on the length of the control delay.

[0045] (14) A control device in another aspect of the present disclosure is a control device that controls the supply of power to a load, and includes: a first acquisition unit that acquires a lower limit value as a confidence interval of a predicted value of power demand by the load; a second acquisition unit that acquires an upper limit value as a confidence interval of a predicted value of solar power generation by a solar cell; a planning unit that plans the output of at least one of a fuel cell and a storage battery that supplies power to the load corresponding to the difference between the lower limit value and the upper limit value; and an output control unit that controls the output of the at least one battery in accordance with the plan.

[0046] According to this configuration, the same effects as those of the control method described in (1) above can be obtained.

[0047] (15) In yet another aspect of the present disclosure, a program is a program for a control device that controls the supply of power to a load, and causes the control device to execute a process of acquiring a lower limit value as a confidence interval for a predicted value of power demand by the load, acquiring an upper limit value as a confidence interval for a predicted value of solar power generation by a solar cell, supplying power to the load corresponding to the difference between the lower limit value and the upper limit value, planning the output of at least one of a fuel cell and a storage battery, and controlling the output of the at least one of the batteries in accordance with the plan.

[0048] According to this configuration, the same effects as those of the control method described in (1) above can be obtained.

[0049] The present disclosure can also be realized as a control system operated by such a program. Needless to say, such a computer program can be distributed on a non-transitory computer-readable recording medium such as a CD-ROM or via a communication network such as the Internet.

[0050] Note that the embodiments described below each illustrate a specific example of the present disclosure. The numerical values, shapes, components, steps, and step orders shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, among the components in the following embodiments, components that are not described in an independent claim that represents a superordinate concept are described as optional components. Furthermore, in all embodiments, the respective contents can be combined.

[0051] (First Embodiment) The configuration of a control system 100 according to an embodiment of the present disclosure will be described below. FIG. 1 is a diagram illustrating an example of the configuration of the control system 100. The control system 100 includes a solar cell 20, a fuel cell 30, a storage battery 40, a load 50, and a control device 10 installed in a facility 1, and a power grid 90 that supplies system power to the facility 1. The facility 1 is, for example, a factory, a plant, or a residence. The solar cell 20, the fuel cell 30, the storage battery 40, the power grid 90, and the control device 10 are connected by power feeders (not shown). Note that while FIG. 1 illustrates one solar cell 20, one fuel cell 30, one storage battery 40, and one load 50, there may be a plurality of these.

[0052] Under the control of the control device 10 , the solar cell 20 supplies power generated by utilizing sunlight (hereinafter, referred to as solar power generation) to the load 50 or the storage battery 40 .

[0053] Under the control of the control device 10 , the fuel cell 30 generates electricity using fuel such as methanol or hydrogen, and supplies the generated electricity to the load 50 .

[0054] The storage battery 40 is a rechargeable secondary battery such as a lithium-ion battery or a nickel-metal hydride battery. The storage battery 40 is charged by solar power generation and system power purchased from the power grid 90 under the control of the control device 10. The storage battery 40 also supplies (discharges) the power stored in the storage battery 40 to the load 50 under the control of the control device 10.

[0055] The load 50 is an electrical device that operates on power supplied from the solar cell 20, the fuel cell 30, the storage battery 40, and / or the power grid 90. The load 50 includes, for example, an information processing device such as a personal computer, a manufacturing robot, an air conditioner, and the like.

[0056] The control device 10 controls the supply of power to the load 50 from the solar cell 20, the fuel cell 30, the storage battery 40, and the power grid 90. The control device 10 also controls the charging of the storage battery 40 using solar power generation and system power purchased from the power grid 90.

[0057] Specifically, the control device 10 includes a memory 12 , a communication unit 13 , an operation unit 14 , a sensor 15 and a processor 11 .

[0058] The memory 12 is configured by, for example, a nonvolatile rewritable semiconductor memory such as a flash memory, a hard disk drive (HDD), etc. The memory 12 stores a control program executed by the processor 11. The memory 12 also stores various information related to the processing executed by the processor 11.

[0059] For example, the memory 12 stores identification information of the area where the control device 10 is installed, the full charge capacity of the storage battery 40, and the rated output of the fuel cell 30 and the storage battery 40.

[0060] When the facility 1 is a factory or a plant, the memory 12 stores information indicating whether each day is an operating day or not for each day. When the facility 1 is a factory or a plant, the memory 12 may store information indicating a production plan for the facility 1. The production plan includes, for each day, the number of production lines in each of a plurality of time periods, the number of workers working on each production line, the number of air conditioning equipment and lighting to be operated on each production line, and the like.

[0061] The communication unit 13 is a communication circuit for communicating between the control device 10 and an external device connected via a network (not shown), such as the Internet.

[0062] The operation unit 14 accepts various operations by the user on the control device 10. The operation unit 14 includes, for example, a display that displays various information and a touch panel device that accepts various operations.

[0063] The sensors 15 include sensors that measure the power supplied to the load 50 from each of the fuel cell 30, the storage battery 40, and the power grid 90, a sensor that measures the remaining power in the storage battery 40, a sensor that measures solar power generation, and a sensor that measures the power charged to the storage battery 40 from the power grid 90. The sensors 15 output various measurement values ​​to the processor 11.

[0064] The processor 11 is configured by, for example, a central processing unit. The processor 11 executes a control program stored in the memory 12 to function as a first acquisition unit 111, a second acquisition unit 112, a planning unit 113, and an output control unit 114. However, this is just one example, and the first acquisition unit 111, the second acquisition unit 112, the planning unit 113, and the output control unit 114 may be realized by dedicated electric circuits such as ASICs.

[0065] The first acquisition unit 111 periodically (for example, every 30 minutes or every hour) controls the communication unit 13 to communicate with a weather information providing server (not shown) to acquire weather information for the area in which the control device 10 is installed from the weather information providing server. When communicating with the weather information providing server, the first acquisition unit 111 transmits identification information for the area in which the control device 10 is installed to the weather information providing server.

[0066] The local weather information includes at least one of the local weather (sunny, rainy, cloudy, etc.), temperature, probability of precipitation, amount of solar radiation, wind speed, wind direction, and a sky image of the local area. The first acquisition unit 111 stores the acquired weather information in the memory 12 as the current weather information for the local area where the control device 10 is installed.

[0067] Furthermore, the first acquisition unit 111 periodically (for example, every five minutes) acquires a measurement value of the power supplied to the load 50 from the sensor 15. The first acquisition unit 111 regards the measurement value as an actual value of the power demand by the load 50, associates the current date and time with the actual value, and stores the associated value in the memory 12.

[0068] Furthermore, the first acquisition unit 111 periodically (for example, every 30 minutes or every hour) predicts the current power demand of the load 50. The first acquisition unit 111 associates the current date and time with the predicted value of the current power demand of the load 50 and stores them in the memory 12.

[0069] Specifically, the memory 12 stores a trained model (hereinafter referred to as the first prediction model) that has been machine-learned to determine the relationship between the measurement date and time of the power supplied to the load 50 by the sensor 15, the measurement value of the power supplied to the load 50 at the measurement date and time, and information corresponding to the measurement date and time (hereinafter referred to as feature information).

[0070] The feature information corresponding to the measurement date and time includes at least one of the following: the day of the week corresponding to the measurement date and time, information indicating whether the measurement date and time is an operating day for facility 1, information indicating the temperature and production plan for the measurement date and time in facility 1, and information indicating whether a production plan has been set for the measurement date and time in facility 1. Having a production plan set for the measurement date and time in facility 1 means that a production plan for the measurement date and time is stored in memory 12.

[0071] When a date and time and feature information corresponding to the date and time are input, the first prediction model outputs a predicted value of the power demand by the load 50 at the date and time. The first acquisition unit 111 refers to the information stored in the memory 12 and inputs the current date and time and feature information corresponding to the current date and time to the first prediction model, thereby predicting the predicted value of the power demand output by the first prediction model as the power demand by the load 50 at the current time.

[0072] The first prediction model may be a model obtained by machine learning of the relationship between the measurement time of the power supplied to the load 50 by the sensor 15 and the measurement value of the power supplied to the load 50 at the measurement time. In this case, when a time is input, the first prediction model outputs a predicted value of the power demand by the load 50 at the input time. Therefore, the first acquisition unit 111 may input the current time to the first prediction model, so that the predicted value of the power demand output by the first prediction model is predicted as the power demand by the load 50 at the current time.

[0073] Furthermore, the first acquisition unit 111 may predict, as the current power demand of the load 50, an average value of past actual values ​​of power demand of the load 50 associated with the same time as the current time. Here, the same time as the current time refers to a time within a predetermined error range (for example, 5 minutes) from the current time.

[0074] Furthermore, the first acquisition unit 111 periodically (for example, every 30 minutes or every hour) acquires a lower limit value as a confidence interval of the predicted value of the power demand of the load 50 (hereinafter, referred to as the lower limit predicted value of the power demand). The confidence interval of the predicted value of the power demand of the load 50 is a range of power demand within which it is certain that the predicted value is the actual value of the power demand of the load 50. Details of how the first acquisition unit 111 acquires the lower limit predicted value of the power demand will be described later.

[0075] The second acquisition unit 112 periodically (for example, every 5 minutes) acquires the measurement value of the power generated by the solar cell 20 from the sensor 15. The second acquisition unit 112 regards the acquired measurement value as the actual value of solar power generation by the solar cell 20, and stores in the memory 12 the current date and time, weather information for the area where the control device 10 is currently installed, and the actual value in association with each other, by referring to the information stored in the memory 12.

[0076] Furthermore, the second acquisition unit 112 periodically (for example, every 30 minutes or every hour) predicts the current solar power generation by the solar cell 20. The second acquisition unit 112 refers to the information stored in the memory 12, and stores in the memory 12 the current date and time, weather information for the area where the control device 10 is currently installed, and the predicted value of solar power generation by the solar cell 20 at the current time, in association with each other.

[0077] Specifically, the memory 12 stores a trained model (hereinafter referred to as the second prediction model) that has been machine-learned to describe the relationship between the time when the sensor 15 measures the power generated by the solar cell 20, the weather information corresponding to the measurement time, and the measured value of solar power generation at the measurement time.

[0078] When a time and weather information corresponding to that time are input, the second prediction model outputs a predicted value of solar power generation by the solar cells 20 at that time. The second acquisition unit 112 refers to the information stored in the memory 12 and inputs the current time and weather information for the area where the control device 10 is currently installed to the second prediction model, thereby predicting the predicted value of solar power generation output by the second prediction model as solar power generation by the solar cells 20 at the current time.

[0079] Furthermore, the second acquisition unit 112 may predict the average value of the actual solar power generation by the solar cell 20 associated with weather information including the same weather as the weather information for the area where the control device 10 is currently installed as the solar power generation by the solar cell 20 at the current time.

[0080] Furthermore, the second acquisition unit 112 periodically (for example, every 30 minutes or every hour) acquires an upper limit value (hereinafter, an upper limit predicted value of solar power generation) as a confidence interval of the predicted value of solar power generation by the solar cell 20. The confidence interval of the predicted value of solar power generation by the solar cell 20 is a range of solar power generation within which it is possible to be confident that the predicted value is an actual value of solar power generation by the solar cell 20. Details of a method for the second acquisition unit 112 to acquire the upper limit predicted value of solar power generation will be described later.

[0081] The planner 113 periodically (for example, every 30 minutes or every hour) creates a plan for supplying power to the load 50 from each of the solar cell 20, the fuel cell 30, the storage battery 40, and the power grid 90. The method for creating the plan by the planner 113 will be described in detail later.

[0082] The output control unit 114 controls the power output from each of the solar cell 20 , the fuel cell 30 , the storage battery 40 and the power grid 90 in accordance with the plan formulated by the planner 113 .

[0083] The above is the configuration of the control device 10. Next, the processing performed by the control device 10 will be described. Fig. 2 is a flowchart showing an example of the processing performed by the control device 10. The processor 11 periodically (for example, every 30 minutes or every hour) executes the processing shown in Fig. 2.

[0084] In step S11 , the first acquisition unit 111 acquires a lower limit predicted value, which is the lower limit of the confidence interval of the predicted value of the power demand of the load 50 .

[0085] In step S12 , the second acquisition unit 112 acquires an upper limit predicted value, which is the upper limit of the confidence interval for the predicted value of the solar power generation by the solar cell 20 .

[0086] In step S13 , the planner 113 creates a plan for supplying power to the load 50 from each of the solar cell 20 , the fuel cell 30 , the storage battery 40 , and the power grid 90 .

[0087] In step S14, the output control unit 114 controls the power output from each of the solar cell 20, the fuel cell 30, the storage battery 40, and the power grid 90 in accordance with the plan established in step S13. Then, the processor 11 ends the processing shown in FIG.

[0088] Next, a detailed description will be given of the process in step S11 ( FIG. 2 ) in which the first acquisition unit 111 acquires a lower limit predicted value, which is the lower limit of the confidence interval of the predicted value of the power demand of the load 50. FIG. 3 is a flowchart illustrating an example of the process of acquiring a lower limit of the confidence interval of the predicted value of the power demand.

[0089] When step S11 (FIG. 2) starts, in step S21, the first acquisition unit 111 predicts the current power demand of the load 50 as described above.

[0090] In step S22, the first acquisition unit 111 acquires the time trend of the actual prediction error, which is the difference between the power demand of the load 50 predicted in the past and the actual value of the power demand of the load 50 at the time of the prediction.

[0091] Specifically, the first acquisition unit 111 refers to the memory 12 and acquires, for multiple past actual values ​​of electricity demand, the absolute value of the difference between the predicted value of electricity demand associated with the date and time (e.g., August 1st, 10:00) closest to the date and time (e.g., August 1st, 10:05) associated with each actual value, as the prediction error actual at the date and time (e.g., August 1st, 10:00) associated with each actual value.

[0092] In addition, the first acquisition unit 111 may acquire, as the prediction error actual for the date and time (e.g., August 1st, 10:00) associated with each actual value, the absolute value of the difference between the average value of the predicted power demand value associated with the closest past date and time (e.g., August 1st, 10:00) and the closest future date and time (e.g., August 1st, 10:30) associated with each actual value.

[0093] In step S23, the first acquisition unit 111 calculates the standard deviation of the actual prediction error results based on the temporal transition of the actual prediction error results acquired in step S22.

[0094] Specifically, the first acquisition unit 111 calculates, in chronological order, the average values ​​of the actual forecast errors at one or more dates and times (e.g., 10:00 on July 31st and 10:00 on August 1st) that indicate the same time (e.g., 10:00) for the multiple dates and times acquired in step S22. The first acquisition unit 111 uses a group of average values ​​of the calculated actual forecast errors as a population and calculates the standard deviation of the group of average values ​​as the standard deviation of the actual forecast errors. The one or more dates and times that indicate the same time refer to one or more dates and times that match within a predetermined error range.

[0095] In step S24, the first acquisition unit 111 calculates a lower limit predicted value of the power demand based on the current power demand by the load 50 predicted in step S21 and the standard deviation of the actual prediction error calculated in step S23. Then, the first acquisition unit 111 ends step S11 ( FIG. 2 ).

[0096] Specifically, in step S24, the first acquisition unit 111 calculates the current power demand DM due to the load 50 predicted in step S21 as shown in the following formula (1): ｃｅｎｔｅｒ The result of subtracting the product of the standard deviation of the actual prediction error calculated in step S23 and a predetermined coefficient (for example, 2) (standard deviation of the actual prediction error × predetermined coefficient) ΔDM from the above is the lower limit predicted value DM of the power demand by the load 50. ｍｉｎ It is calculated as follows.

[0097] Furthermore, the specified coefficient is not limited to a constant, and may be set to a larger value, for example, the stricter the regulations for determining whether reverse power flow, which is the supply of power to the power grid 90, is occurring in the facility 1.

[0098] Specifically, the processor 11 determines that reverse power flow is occurring when the time during which power equal to or greater than a predetermined threshold is continuously supplied to the power grid 90 is equal to or greater than a predetermined specified time. In this case, the smaller the predetermined threshold and the shorter the predetermined specified time, the stricter the rules for determining whether reverse power flow, which is the supply of power to the power grid 90, is occurring. Therefore, the smaller the predetermined threshold, the larger the predetermined coefficient may be set. Furthermore, the shorter the predetermined specified time, the larger the predetermined coefficient may be set.

[0099] Furthermore, the predetermined coefficient may be set to a larger value as the control delay time that occurs when the control device 10 controls the storage battery 40 to output power becomes longer.

[0100] Specifically, when processor 11 controls storage battery 40 to output specified power, it takes a considerable amount of time for processor 11 to create a control signal indicating the control instruction to storage battery 40 and to transmit the control signal to storage battery 40. In addition, storage battery 40 requires a considerable amount of start-up time before it can stably output the power specified by the control signal.

[0101] Thus, a considerable control delay occurs in the control by the control device 10 to cause the storage battery 40 to output power. Therefore, the time from when the processor 11 starts creating a control signal indicating the control instruction to the storage battery 40 until the storage battery 40 stably outputs the instructed power may be set as the control delay time that occurs in the control by the control device 10 to cause the storage battery 40 to output power, and may be determined based on experimental values. The longer the control delay time, the larger the predetermined coefficient may be set.

[0102] Next, a detailed description will be given of the process in step S12 ( FIG. 2 ) in which the second acquisition unit 112 acquires an upper limit predicted value, which is an upper limit value as a confidence interval of a predicted value of solar power generation by the solar cell 20. FIG. 4 is a flowchart illustrating an example of the process of acquiring an upper limit value as a confidence interval of a predicted value of solar power generation.

[0103] When step S12 (FIG. 2) starts, in step S31, the second acquisition unit 112 predicts the solar power generation by the solar cell 20 at the current time, as described above.

[0104] In step S32, the second acquisition unit 112 acquires the time trend of the prediction error actual value, which is the difference between the solar power generation by the solar cell 20 predicted in the past and the actual value of the solar power generation by the solar cell 20 at the time of the prediction.

[0105] Specifically, the second acquisition unit 112 refers to the memory 12 and acquires, for multiple past actual values ​​of solar power generation, the absolute value of the difference between each actual value and the predicted value of solar power generation associated with the date and time (e.g., August 1st, 10:00) closest to the date and time (e.g., August 1st, 10:05) associated with each actual value, as the prediction error actual at the date and time (e.g., August 1st, 10:00) associated with each actual value.

[0106] The second acquisition unit 112 may acquire, as the prediction error actual at the date and time (e.g., August 1st, 10:00) associated with each actual value, the absolute value of the difference between the average value of the predicted value of solar power generation associated with the closest past date and time (e.g., August 1st, 10:00) and the closest future date and time (e.g., August 1st, 10:30) associated with each actual value.

[0107] In step S33, the second acquisition unit 112 calculates the standard deviation of the actual prediction error results based on the temporal transition of the actual prediction error results acquired in step S32.

[0108] Specifically, the second obtainment unit 112 calculates, in chronological order, the average of the actual forecast errors at one or more dates and times (e.g., July 31, 10:00, August 1, 10:00) that indicate the same time (e.g., 10:00) for the multiple dates and times obtained in step S32. The second obtainment unit 112 uses a group of average values ​​of the calculated actual forecast errors as a population and calculates the standard deviation of the group of average values ​​as the standard deviation of the actual forecast errors.

[0109] In step S34, the second acquisition unit 112 calculates the upper limit predicted value of the solar power generation based on the solar power generation by the solar cell 20 at the current time predicted in step S31 and the standard deviation of the actual prediction error calculated in step S33, and then ends step S12 (Figure 2).

[0110] Specifically, in step S34, the second acquisition unit 112 calculates the solar power generation PV by the solar cell 20 at the current time, which is predicted in step S31, as shown in the following formula (2): ｃｅｎｔｅｒ and the product of the standard deviation of the actual prediction error calculated in step S33 and a predetermined coefficient (for example, 2) (standard deviation of the actual prediction error × predetermined coefficient) ΔPV, which is the upper limit predicted value PV of photovoltaic power generation. ｍａｘ It is calculated as follows.

[0111] Note that the predetermined coefficient used in step S34 is not limited to a constant, similar to the predetermined coefficient used in step S24 ( FIG. 3 ). Similar to the predetermined coefficient used in step S24 ( FIG. 3 ), the predetermined coefficient used in step S34 may be set to a larger value as the regulations for determining whether reverse power flow is occurring in facility 1 are stricter. Similarly to the predetermined coefficient used in step S24 ( FIG. 3 ), the predetermined coefficient used in step S34 may be set to a larger value as the control delay time occurring in control by control device 10 to cause storage battery 40 to output power is longer.

[0112] Next, details of the process performed by the planner 113 in step S13 (FIG. 2) to create a plan for supplying power from the solar cell 20, the fuel cell 30, the storage battery 40, and the power grid 90 to the load 50 will be described.

[0113] In step S13 (FIG. 2), the planner 113 calculates the lower limit predicted value DM ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ The output of the fuel cell 30 and the output of the storage battery 40 are planned so as to supply to the load 50 an electric power corresponding to the difference between the above.

[0114] Specifically, the upper limit predicted value of solar power generation PV ｍａｘ is the lower limit forecast value of electricity demand DM ｍｉｎIn this case, the planner 113 plans to stop the output of the fuel cell 30, and further, ｍａｘ and the lower limit predicted value DM ｍｉｎ A plan is made to charge the storage battery 40 by photovoltaic power generation corresponding to the difference between the above.

[0115] Upper limit predicted value of solar power generation PV ｍａｘ is the lower limit forecast value of electricity demand DM ｍｉｎ If the remaining power amount is greater than , the planner 113 further acquires the remaining power amount of the storage battery 40 measured by the sensor 15. The planner 113 determines whether the storage battery 40 is in a predetermined state close to a fully charged state depending on whether the remaining power amount is less than a predetermined threshold.

[0116] The planning unit 113 may determine whether the storage battery 40 is in a predetermined state close to a fully charged state using another method based on the remaining power amount of the storage battery 40 measured by the sensor 15. When the planning unit 113 determines that the storage battery 40 is in a predetermined state close to a fully charged state, the planning unit 113 further plans to suppress solar power generation.

[0117] In addition, the lower limit of the predicted power demand DM ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ The power corresponding to the difference between the fuel cell 30 and the storage battery 40 may be greater than the sum of the rated outputs of the fuel cell 30 and the storage battery 40. In this case, the planner 113 calculates the lower limit predicted value DM ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ The plan is to purchase from the power grid 90 the power corresponding to the difference between the difference between the total rated output of the fuel cell 30 and the storage battery 40 and the power (power shortage).

[0118] Furthermore, the planning unit 113 plans the output of the fuel cell 30 and the output of the storage battery 40 so that the frequency of output fluctuations of the fuel cell 30 is lower than the frequency of output fluctuations of the storage battery 40. The frequency of output fluctuations refers to the number of times per predetermined period that the difference between the previous output value and the current output value is equal to or greater than a predetermined value.

[0119] Step S13 ( FIG. 2 ) can be implemented, for example, as follows: The planner 113 acquires the remaining power amount of the storage battery 40 measured by the sensor 15, and determines whether the storage battery 40 is in a predetermined state close to a fully charged state based on whether the remaining power amount is less than a predetermined threshold.

[0120] When the planning unit 113 determines that the storage battery 40 is in a predetermined state close to a fully charged state, the planning unit 113 calculates the lower limit predicted value DM of the power demand by a known mathematical optimization method. ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ The upper limit value PV of photovoltaic power generation is set to satisfy the constraints shown in the following formula (3), including the above, and the objective functions shown in 1) to 4) below. ｌｉｍｉｔ The output FC of the fuel cell 30, the output SB of the storage battery 40, and the amount of power Grid purchased from the power grid 90 are calculated.

[0121] The objective function is: 1) to set the output FC of the fuel cell 30 to a value equal to or greater than 0 so that the number of times that the output FC changes by more than a predetermined value from the previously calculated output FC is as small as possible; and 2) to set the upper limit value PV of the photovoltaic power generation. ｌｉｍｉｔ is set to a value that is as large as possible and is equal to or greater than 0. Furthermore, the objective function includes 3) setting the purchased power Grid from the power grid 90 to a value that is as small as possible and is equal to or greater than 0, and 4) setting the absolute value of the output SB of the storage battery 40 to a value that is equal to or less than the full charge capacity of the storage battery 40.

[0122] When the planning unit 113 determines that the storage battery 40 is not in a predetermined state close to a fully charged state, the upper limit value PV of the photovoltaic power generation is calculated from the above formula (3). ｌｉｍｉｔ The output FC of the fuel cell 30, the output SB of the storage battery 40, and the purchased power Grid from the power grid 90 are calculated so as to satisfy the constraints shown in the following equation (4) with the omitted values ​​and the objective functions shown in 1), 3), and 4) above.

[0123] In this case, the output SB of the storage battery 40 is calculated to satisfy the control conditions shown in the above formula (3) or (4) and also to satisfy the objective function shown in 4). On the other hand, the output FC of the fuel cell 30 is calculated to satisfy the control conditions shown in the above formula (3) or (4) and also to satisfy the objective function shown in 1). Therefore, the output FC of the fuel cell 30 is calculated to be lower than the output fluctuation of the output SB of the storage battery 40.

[0124] The planning unit 113 determines whether the solar power generation by the solar cell 20 measured by the sensor 15 at the current time is greater than the calculated upper limit value PV of the solar power generation. ｌｉｍｉｔ If the upper limit value PV is larger than the upper limit value PV, the power output from the solar cell 20 is ｌｉｍｉｔ Therefore, it is possible to reduce the possibility of reverse power flow occurring due to continued solar power generation even after the storage battery 40 is fully charged.

[0125] If the calculated output FC of the fuel cell 30 is 0, the planner 113 plans to stop the output of the fuel cell 30. If the calculated output FC of the fuel cell 30 is a positive value, the planner 113 plans to supply power corresponding to the value of the output FC from the fuel cell 30 to the load 50.

[0126] When the calculated output SB of the storage battery 40 is a positive value, the planning unit 113 plans to supply power corresponding to the value of the output SB from the storage battery 40 to the load 50. On the other hand, when the calculated output SB of the storage battery 40 is a negative value, the planning unit 113 plans to charge the storage battery 40 with power generated by the solar cell 20 corresponding to the absolute value of the output SB.

[0127] If the calculated power grid 90 purchase amount Grid is zero, the planner 113 makes a plan to stop purchasing power from the power grid 90. If the calculated power grid Grid purchase amount Grid is a positive value, the planner 113 makes a plan to purchase power corresponding to the value of the power grid 90 purchase amount Grid from the power grid 90 and supply the power to the load 50.

[0128] FIG. 5 shows the lower limit predicted value DM of the power demand when the planning unit 113 calculates the lower limit predicted value DM of the power demand. ｍｉｎis the upper limit predicted value of solar power generation PV ｍａｘ 10 is a diagram showing an example of a plan created when it is determined that the storage battery 40 is not in a predetermined state close to a fully charged state, and the charge level is greater than the predetermined value.

[0129] The horizontal axis of Fig. 5 represents time, and the vertical axis of Fig. 5 represents power values. Graph G10 shows an example of the transition over time of the actual value of power demand by the load 50. Graph G11 shows an example of the transition over time of the predicted value of power demand by the load 50. Graph G12 shows an example of the transition over time of the lower limit predicted value DM of power demand. ｍｉｎ Graph G20 shows an example of the time transition of the actual value of solar power generation. Graph G21 shows an example of the time transition of the predicted value of solar power generation. Graph G22 shows an example of the upper limit predicted value PV of solar power generation. ｍａｘ 1 shows an example of the time progression of

[0130] For example, when the process shown in FIG. 2 is performed at time t, the lower limit predicted value DM of the power demand is ｍｉｎ is the upper limit predicted value of solar power generation PV ｍａｘ Since it is larger than the lower limit predicted value DM of the power demand corresponding to the left side of the formula (4), ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ Therefore, in step S13 (FIG. 2), the planner 113 calculates at least one of the output FC of the fuel cell 30 and the output SB of the storage battery 40 as a positive value, and plans to supply power from at least one of these to the load 50.

[0131] Therefore, the actual value of the power demand is the lower limit predicted value DM ｍｉｎ If the actual value of the power demand and the lower limit forecast value DM ｍｉｎ The grid power corresponding to the difference P1(t) between the actual value of the photovoltaic power generation and the upper limit predicted value PV ｍａｘ If it falls below this, the actual value of solar power generation and the upper limit predicted value PV ｍａｘ The grid power corresponding to the difference P2(t) between the above is purchased from the grid 90.

[0132] This reduces the lower limit of the predicted power demand DM ｍｉｎand the upper limit predicted value of solar power generation PV ｍａｘ It is possible to plan to supply the minimum amount of power predicted to be required to be supplied to the load 50, which corresponds to the difference P0(t) between the fuel cell 30 and the storage battery 40, to the load 50 from at least one of the fuel cell 30 and the storage battery 40. Therefore, the minimum amount of power required by the load 50 can be met by the power output from the fuel cell 30 and the storage battery 40 according to the plan.

[0133] Furthermore, since it is not planned to output power from the storage battery 40 that exceeds the minimum power required by the load 50, it is possible to avoid charging the storage battery 40 with solar power generation that exceeds the minimum power required by the load 50. This prevents the outputs of the solar cell 20 and the storage battery 40 from exceeding the power demand of the load 50. As a result, it is possible to realize power supply from the solar cell 20, the fuel cell 30, and the storage battery 40 to the load 50 without causing reverse power flow.

[0134] On the other hand, the upper limit predicted value of solar power generation PV ｍａｘ is the lower limit forecast value of electricity demand DM ｍｉｎ and the storage battery 40 is not in a predetermined state close to a fully charged state, the power corresponding to the left side of equation (4) will be a negative value. Therefore, the planner 113 calculates the output FC of the fuel cell 30 as 0 and the output SB of the storage battery 40 as a negative value, stops the output of the fuel cell 30, and plans to charge the storage battery 40 with the power generated by the solar cell 20 corresponding to the absolute value of the output SB. Therefore, it is possible to reduce the possibility of reverse power flow occurring due to solar power generation.

[0135] Furthermore, since there is a considerable control delay in the power output of the fuel cell 30 and the storage battery 40, it is difficult to obtain the rated output of the fuel cell 30 and the storage battery 40 at the desired timing. ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘIf the power corresponding to the difference between the rated output of the fuel cell 30 and the storage battery 40 is greater than the sum of the rated outputs of the fuel cell 30 and the storage battery 40, the purchased power Grid from the power grid 90 will have a positive value. Therefore, the planner 113 plans to purchase power corresponding to the value of the purchased power Grid from the power grid 90 and supply that power to the load 50. Therefore, power that cannot be supplied to the load 50 by the solar cell 20, the fuel cell 30, and the storage battery 40 alone can be purchased from the power grid 90 and appropriately supplied to the load 50.

[0136] In step S11 ( FIG. 2 ), the first acquisition unit 111 may calculate a different lower limit predicted value of the power demand depending on whether the current time is an operating day or not. In this case, the lower limit predicted value of the power demand can be calculated more accurately depending on whether the current time is an operating day or not.

[0137] Specifically, if the current time is an operating day, in step S22 (FIG. 3), the first acquisition unit 111 may acquire a temporal change in the actual prediction error, which is the difference between the power demand by the load 50 predicted on a past operating day and the actual value of the power demand by the load 50 at the time of the prediction. On the other hand, if the current time is a non-operating day that is not an operating day, the first acquisition unit 111 may acquire a temporal change in the actual prediction error, which is the difference between the power demand by the load 50 predicted on a past non-operating day and the actual value of the power demand by the load 50 at the time of the prediction.

[0138] Similarly, in step S12 ( FIG. 2 ), the second acquisition unit 112 may calculate a different upper limit predicted value for solar power generation depending on the current probability of precipitation. In this case, the upper limit predicted value for solar power generation depending on the current probability of precipitation can be calculated with higher accuracy.

[0139] Specifically, if the current probability of precipitation is less than 10%, in step S32 (Figure 4), the second acquisition unit 112 may acquire the time trend of the actual prediction error, which is the difference between the power demand from the load 50 predicted on a date and time in the past when the probability of precipitation was less than 10% and the actual value of the power demand from the load 50 at the time of the prediction.

[0140] Similarly, the current precipitation probability may be classified in increments of 10%, and in step S32 (FIG. 4), the second acquisition unit 112 may acquire the time trend of the actual prediction error, which is the difference between the power demand from the load 50 predicted on a date and time with a past precipitation probability of the same classification (e.g., 10% or more but less than 20%) as the current precipitation probability (e.g., 15%) and the actual value of the power demand from the load 50 at the time of the prediction.

[0141] (Embodiment 2) In embodiment 1, an example was described in which in step S11 (FIG. 2), the first acquisition unit 111 performs the processing shown in FIG. 3 to calculate the standard deviation of the actual prediction error, which is the difference between the previously predicted power demand and the actual value of the power demand at the time of the prediction, and acquires a lower limit predicted value of the power demand.

[0142] In the second embodiment, an example will be described in which the first acquisition unit 111 performs the process shown in Fig. 6 in step S11 (Fig. 2) to acquire a lower limit prediction value of the power demand using the trained model. Fig. 6 is a flowchart showing another example of the process of acquiring a lower limit as a confidence interval of the prediction value of the power demand.

[0143] In embodiment 2, the memory 12 stores a learned model (hereinafter referred to as the first estimation model) that has been machine-learned to determine the relationship between the difference between the power demand due to the load 50 previously predicted by the first acquisition unit 111 and the actual value of the power demand at the time of the prediction, and one or more predetermined features corresponding to the time of the prediction.

[0144] Hereinafter, the difference between the power demand due to the load 50 previously predicted by the first acquisition unit 111 and the actual value of the power demand at the time of the prediction will be referred to as a first actual prediction error (actual prediction error). When the one or more feature quantities corresponding to the current time are input, the first estimation model outputs an estimated value of the first actual prediction error at the current time.

[0145] The one or more predetermined feature amounts corresponding to the prediction time include at least one of the day of the week corresponding to the prediction time, information indicating whether the prediction time is an operating day of the facility 1, information indicating the temperature and production plan at the prediction time in the facility 1, and information indicating whether a production plan at the prediction time has been set in the facility 1. A production plan at the prediction time in the facility 1 has been set means that a production plan at the prediction time is stored in the memory 12.

[0146] Similarly, the one or more feature amounts corresponding to the current time include at least one of the day of the week corresponding to the current time, information indicating whether the current time is an operating day for facility 1, information indicating the current temperature and production plan for facility 1, and information indicating whether a current production plan has been set for facility 1. The fact that a current production plan has been set for facility 1 means that a current production plan is stored in memory 12.

[0147] Specifically, for multiple past predicted values ​​of electricity demand at facility 1, the absolute value of the difference between each predicted value and the actual value of electricity demand associated with the date and time (e.g., August 1st, 10:00) closest to the date and time (e.g., August 1st, 10:00) associated with each predicted value is calculated as the first prediction error actual at the date and time (e.g., August 1st, 10:00) associated with each predicted value.

[0148] Furthermore, the first estimation model may be generated by the processor 11 in the control device 10, or may be generated in an information processing device other than the control device 10 using multiple past predicted and actual values ​​of electricity demand in the facility 1 stored in the memory 12 of the control device 10.

[0149] As shown in FIG. 6, when step S11 (FIG. 2) starts, in step S41, the first acquisition unit 111 predicts the current power demand of the load 50, similar to step S21 (FIG. 3).

[0150] In step S42, the first acquisition unit 111 refers to the memory 12 and acquires the one or more feature amounts corresponding to the current time.

[0151] In step S43, the first acquisition unit 111 estimates a first actual prediction error at the current time by inputting the one or more feature amounts corresponding to the current time acquired in step S42 into a first estimation model. Specifically, the first acquisition unit 111 inputs the one or more feature amounts corresponding to the current time acquired in step S42 into the first estimation model, thereby acquiring an estimated value of the first actual prediction error output by the first estimation model.

[0152] In step S44, the first acquisition unit 111 calculates the lower limit predicted value of the power demand by subtracting the product of the estimated value of the first prediction error actual result estimated in step S43 and a predetermined coefficient (e.g., 2) from the power demand due to the load 50 at the current time predicted in step S41 (= estimated value of the first prediction error actual result × predetermined coefficient).

[0153] According to the second embodiment, the first actual forecast error at the current time can be appropriately estimated by inputting into the first estimation model at least one of the day of the week corresponding to the current time, information indicating whether the current time is an operating day for facility 1, information indicating the current temperature and production plan for facility 1, and information indicating whether a production plan has been set for facility 1 at the current time. Therefore, the first actual forecast error at the current time can be used to appropriately calculate the lower limit of the confidence interval for the forecast value of power demand.

[0154] (Embodiment 3) In the first embodiment, an example has been described in which, in step S12 (FIG. 2), the second acquisition unit 112 performs the processing shown in FIG. 4 to calculate the standard deviation of the actual prediction error, which is the difference between the previously predicted solar power generation and the actual value of solar power generation at the time of the prediction, and acquires the upper limit predicted value of solar power generation.

[0155] In the third embodiment, an example will be described in which the second acquisition unit 112 performs the process shown in Fig. 7 in step S12 (Fig. 2) to acquire an upper limit predicted value of photovoltaic power generation using a trained model. Fig. 7 is a flowchart showing another example of the process of acquiring an upper limit value as a confidence interval of the predicted value of photovoltaic power generation.

[0156] In embodiment 3, the memory 12 stores a trained model (hereinafter referred to as the second estimation model) that is machine-learned to determine the relationship between the difference between the solar power generation by the solar cell 20 previously predicted by the second acquisition unit 112 and the actual value of the solar power generation at the time of the prediction, and one or more predetermined features corresponding to the time of the prediction.

[0157] Hereinafter, the difference between the solar power generation by the solar cell 20 previously predicted by the second acquisition unit 112 and the actual value of the solar power generation at the time of the prediction will be referred to as a second prediction error actual value (prediction error actual value). When the one or more feature quantities corresponding to the current time are input, the second estimation model outputs an estimated value of the second prediction error actual value at the current time.

[0158] The one or more predetermined feature quantities corresponding to the prediction time include at least one of the probability of precipitation, the amount of solar radiation, the wind speed, the wind direction, and a sky image of the area corresponding to the prediction time. Similarly, the one or more feature quantities corresponding to the current time include at least one of the probability of precipitation, the amount of solar radiation, the wind speed, the wind direction, and a sky image of the area corresponding to the current time.

[0159] Specifically, for a plurality of past predicted values ​​of photovoltaic power generation at facility 1, the absolute value of the difference between each predicted value and the actual photovoltaic power generation value associated with the date and time (e.g., 10:00 on August 1) closest to the date and time (e.g., 10:00 on August 1) associated with each predicted value is calculated as the second actual prediction error at the date and time (e.g., 10:00 on August 1) associated with each predicted value. The second estimation model is generated by machine learning the relationship between the second actual prediction error at the date and time associated with each predicted value and one or more predetermined feature quantities corresponding to the date and time for a plurality of past predicted values ​​of photovoltaic power generation at facility 1.

[0160] Furthermore, the second estimation model may be generated by the processor 11 in the control device 10, or may be generated in an information processing device other than the control device 10 using multiple past predicted and actual values ​​of solar power generation in the facility 1 stored in the memory 12 of the control device 10.

[0161] As shown in FIG. 7, when step S12 (FIG. 2) starts, in step S51, the second acquisition unit 112 predicts the solar power generation by the solar cell 20 at the current time, similarly to step S31 (FIG. 4).

[0162] In step S52, the second acquisition unit 112 refers to the memory 12 and acquires the one or more feature amounts corresponding to the current time.

[0163] In step S53, the second acquisition unit 112 estimates a second actual prediction error at the current time by inputting the one or more feature amounts corresponding to the current time acquired in step S52 into a second estimation model. Specifically, the second acquisition unit 112 inputs the one or more feature amounts corresponding to the current time acquired in step S52 into the second estimation model, thereby acquiring an estimated value of the second actual prediction error output by the second estimation model.

[0164] In step S54, the second acquisition unit 112 calculates the upper limit predicted value of solar power generation as the sum of the solar power generation by the solar cell 20 at the current time predicted in step S51 and the product of the estimated value of the second prediction error actual result estimated in step S53 and a predetermined coefficient (e.g., 2) (= estimated value of the second prediction error actual result × predetermined coefficient).

[0165] According to the third embodiment, by inputting at least one of the precipitation probability, solar radiation, wind speed, wind direction, and all-sky image corresponding to the current time into the second estimation model, the second forecast error actual value at the current time can be appropriately estimated. Therefore, the second forecast error actual value at the current time can be used to appropriately calculate the upper limit value as the confidence interval of the forecast value of solar power generation.

[0166] In the first to third embodiments, the control system 100 (FIG. 1) may not include the fuel cell 30. Accordingly, in step S13 (FIG. 2), the planner 113 calculates the lower limit predicted value DM ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ The output of the storage battery 40 may be planned to supply the load 50 with power corresponding to the difference between the current and the load 50 .

[0167] Specifically, when the planning unit 113 determines that the storage battery 40 is in a predetermined state close to a fully charged state, the upper limit value PV of the photovoltaic power generation is calculated so as to satisfy the constraint condition obtained by deleting the term of the output FC of the fuel cell 30 from the constraint condition shown in the above equation (3) and the objective functions shown in 2) to 4). ｌｉｍｉｔ , the output SB of the storage battery 40, and the power Grid purchased from the power grid 90. If the planning unit 113 determines that the storage battery 40 is not in a predetermined state close to a fully charged state, the upper limit value PV of the photovoltaic power generation is calculated so as to satisfy the constraint condition obtained by deleting the term of the output FC of the fuel cell 30 from the constraint condition shown in the above formula (4) and the objective functions shown in 2) to 4) above. ｌｉｍｉｔ , the output SB of the storage battery 40, and the amount of power Grid purchased from the power grid 90 may be calculated.

[0168] Alternatively, in the first to third embodiments, the control system 100 (FIG. 1) may not include the storage battery 40. Accordingly, in step S13 (FIG. 2), the planner 113 calculates the lower limit predicted value DM ｍｉｎ and the upper limit predicted value of solar power generation PV ｍａｘ The output of the fuel cell 30 may be planned to supply to the load 50 an electric power corresponding to the difference between the above.

[0169] Specifically, when the control system 100 (FIG. 1) does not include the storage battery 40, surplus power cannot be charged to the storage battery 40, just as when the storage battery 40 is in a predetermined state close to a fully charged state. For this reason, the planning unit 113 calculates the upper limit value PV of photovoltaic power generation so as to satisfy the constraint condition obtained by deleting the term for the output SB of the storage battery 40 from the constraint condition shown in the above formula (3) and the objective functions shown in 1) to 3). ｌｉｍｉｔ , the output FC of the fuel cell 30, and the amount of power Grid purchased from the power grid 90 may be calculated.

[0170] According to the present disclosure, it is possible to realize power supply from solar cells, fuel cells, and storage batteries to a load without generating reverse power flow, which is useful for achieving self-sufficiency in power in facilities such as factories.

Claims

1. A control method in a control device that controls the supply of power to a load, comprising: obtaining a lower limit value as a confidence interval for a predicted value of power demand by the load; obtaining an upper limit value as a confidence interval for a predicted value of solar power generation by a solar cell; planning the output of at least one of a fuel cell and a storage battery that supplies power to the load corresponding to the difference between the lower limit value and the upper limit value; and controlling the output of the at least one of the batteries in accordance with the plan.

2. The control method of claim 1, wherein, when the upper limit value is greater than the lower limit value, the plan includes a plan to stop output of the fuel cell, and further includes a plan to charge the storage battery by the solar power generation corresponding to the difference between the upper limit value and the lower limit value, and the control further includes controlling the output of the solar cell in accordance with the plan.

3. The control method of claim 2, wherein the plan further comprises obtaining the remaining power amount of the storage battery, determining whether the storage battery is in a predetermined state close to a fully charged state based on the remaining power amount, and if it is determined that the storage battery is in the predetermined state, further planning to suppress the solar power generation, and the control further comprises controlling the output of the solar cell in accordance with the plan if it is determined that the storage battery is in the predetermined state.

4. A control method as claimed in any one of claims 1 to 3, wherein, when the power corresponding to the difference between the lower limit value and the upper limit value is greater than the sum of the rated outputs of the fuel cell and the storage battery, the plan further plans to purchase from the power grid the shortage power, which is the power corresponding to the difference between the difference between the lower limit value and the upper limit value and the sum, and the control further supplies the shortage power purchased from the power grid to the load in accordance with the plan.

5. A control method as described in claim 1, wherein the planning includes planning the output of the fuel cell and the output of the storage battery so that the frequency of output fluctuations of the fuel cell is lower than the frequency of output fluctuations of the storage battery.

6. The control method of claim 1, wherein in obtaining the lower limit value, the current power demand is predicted based on past actual values ​​of the power demand, a time transition of actual prediction error which is the difference between the previously predicted power demand and the actual value of the power demand at the time of the prediction is obtained, a standard deviation of the actual prediction error is calculated based on the time transition of the actual prediction error, and a result obtained by subtracting the product of the standard deviation of the actual prediction error and a predetermined coefficient from the current power demand is obtained as the lower limit value.

7. The control method of claim 1, wherein in obtaining the lower limit value, the power demand at the current time is predicted based on past actual values ​​of the power demand, the one or more features corresponding to the current time are input into a trained model that has been machine-learned to determine the relationship between a prediction error actual value, which is the difference between the previously predicted power demand and the actual value of the power demand at the time of the prediction, and the one or more predetermined features corresponding to the time of the prediction, thereby estimating the prediction error actual value at the current time, and the product of the estimated value of the prediction error actual value at the current time and a predetermined coefficient is obtained as the lower limit value.

8. The control method described in claim 7, wherein the one or more features corresponding to the current time and the predicted time include at least one of the days of the week corresponding to the current time and the predicted time, information indicating whether the current time and the predicted time are operating days of the facility in which the load is installed, information indicating the temperature and production plan at the facility at the current time and the predicted time, and information indicating whether the production plan at the facility at the current time and the predicted time has been set.

9. The control method according to claim 1, wherein in obtaining the upper limit value, the solar power generation at the current time is predicted based on past actual values ​​of the solar power generation, a time transition of a prediction error actual value which is a difference between the previously predicted solar power generation and the actual value of the solar power generation at the time of the prediction is obtained, a standard deviation of the prediction error actual value is calculated based on the time transition of the prediction error actual value, and the upper limit value is obtained as a sum of the solar power generation at the current time and a product of the standard deviation of the prediction error actual value and a predetermined coefficient.

10. The control method according to claim 1, wherein in obtaining the upper limit value, the solar power generation at the current time is predicted based on past actual values ​​of the solar power generation, the one or more features corresponding to the current time are input into a trained model that has been machine-learned to determine the relationship between a prediction error actual value, which is the difference between the previously predicted solar power generation and the actual value of the solar power generation at the time of the prediction, and one or more predetermined features corresponding to the time of the prediction, thereby estimating the prediction error actual value at the current time, and the sum of the solar power generation at the current time and the product of the estimated value of the prediction error actual value at the current time and a predetermined coefficient is obtained as the upper limit value.

11. The control method described in claim 10, wherein the one or more features corresponding to the current time and the predicted time include at least one of the probability of precipitation, solar radiation, wind speed, wind direction, and an all-sky image corresponding to the current time and the predicted time.

12. A control method as claimed in any one of claims 6 to 11, wherein the specified coefficient is set to a larger value as the regulations for determining whether or not power is being supplied to the power grid in the facility in which the load is installed are stricter.

13. A control method as claimed in any one of claims 6 to 11, wherein the predetermined coefficient is set to be larger as a control delay time occurring in the control by the control device to output power to the storage battery becomes longer.

14. A control device that controls the supply of power to a load, comprising: a first acquisition unit that acquires a lower limit value as a confidence interval of a predicted value of power demand by the load; a second acquisition unit that acquires an upper limit value as a confidence interval of a predicted value of photovoltaic power generation by a solar cell; a planning unit that plans the output of at least one of a fuel cell and a storage battery that supplies power to the load corresponding to the difference between the lower limit value and the upper limit value; and an output control unit that controls the output of the at least one of the batteries in accordance with the plan.

15. A program for a control device that controls the supply of power to a load, the program causing the control device to execute the following processes: acquire a lower limit value as a confidence interval for a predicted value of power demand by the load; acquire an upper limit value as a confidence interval for a predicted value of solar power generation by a solar cell; plan the output of at least one of a fuel cell and a storage battery that supplies power to the load corresponding to the difference between the lower limit value and the upper limit value; and control the output of the at least one of the batteries in accordance with the plan.

Images