Power management device

Abstract

To improve accuracy of behavior prediction used when a charge / discharge plan is changed even when a behavior mode of an electric vehicle is greatly changed.SOLUTION: Action history of an electric vehicle is accumulated in a database of a server 300. When there is a DR request, an extraction unit 303 extracts the action history from a database 302 for each vehicle ID. A determination unit 304 determines whether or not current behavior history, for example behavior history within 24 hours from time of extraction among extracted behavior history, has greatly changed from the past behavior history. When the current behavior history is changed from the past behavior history by an allowable value or more, it is determined that the behavior history is greatly changed, and a weight determination unit 305 sets weight of the behavior history after the present time to be large with respect to the past behavior history. A learning model 306 estimates behavior prediction of an electric vehicle 100 by using weighting set by a weight determination unit 350.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a power management apparatus. [Background technology]

[0002] Japanese Patent Laid-Open Publication No. 2022-113460 (Patent Document 1) discloses a power adjustment device for a virtual power plant (VPP). This power adjustment device predicts the behavior of electric vehicles participating in the VPP from information including the past behavior history of the electric vehicles, and creates a charge / discharge plan based on the predicted behavior of the electric vehicles. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2022-113460 Summary of the Invention [Problem to be solved by the invention]

[0004] The power adjustment device in Patent Document 1 predicts vehicle behavior using a behavior prediction model obtained through machine learning, including deep learning, and predicts future behavior from information including past behavior history. Therefore, there is a concern that the accuracy of behavior prediction will decrease when the behavior of the electric vehicle changes significantly. If the accuracy of behavior prediction decreases, it will be impossible to create an appropriate charging and discharging plan.

[0005] An object of the present disclosure is to improve the accuracy of behavior prediction used when creating a charge / discharge plan even when the behavior of an electric vehicle changes significantly. [Means for solving the problem]

[0006] A power management device disclosed herein predicts the behavior of an electric vehicle from information including the behavior history of the electric vehicle, and creates a charge / discharge plan for the electric vehicle based on the behavior prediction. The behavior prediction is performed using a learning model, and the power management device includes a determination unit that determines whether the current behavior history of the electric vehicle has changed from past behavior histories by more than an allowable value, and if the current behavior history has changed from past behavior histories by more than the allowable value, the power management device increases the weight of the current and subsequent behavior histories so that the current and subsequent behavior histories are more significantly reflected in the behavior prediction than past behavior histories.

[0007] According to this configuration, when the behavior of the electric vehicle changes significantly and the current behavior history of the electric vehicle changes from the past behavior history by more than an allowable value, the weight of the behavior history from the current time onwards is increased so that the behavior history from the current time onwards is more significantly reflected in the behavior prediction compared to the past behavior history, and behavior prediction is performed using the learning model. Therefore, even when the behavior of the electric vehicle changes significantly, the accuracy of the behavior prediction used when creating a charge / discharge plan can be improved. Note that the learning model may be a behavior prediction model obtained by machine learning, including deep learning. [Effects of the Invention]

[0008] According to the present disclosure, even when the behavior of an electric vehicle changes significantly, the accuracy of behavior prediction used when creating a charge / discharge plan can be improved. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a diagram showing a schematic configuration of a power management system according to an embodiment of the present invention; [Figure 2] FIG. 2 is a functional block diagram configured in a server in the present embodiment. [Figure 3] 10 is a flowchart illustrating an example of a charge / discharge plan creation process that is processed by a server. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference numerals, and description thereof will not be repeated.

[0011] FIG. 1 is a diagram showing a schematic configuration of a power management system according to this embodiment. The power management system performs power management for a power grid PG. The power management system includes a vehicle group 1, an EVSE group 2, and a server 300. The vehicle group 1 includes a plurality of electric vehicles 100. The EVSE group 2 includes a plurality of EVSEs 200. The EVSEs 200 are electric vehicle service equipment. In this embodiment, the EVSEs 200 are installed in homes or businesses, and are capable of charging the electric vehicles 100 with power supplied from the power grid PG, and supplying power stored in the batteries of the electric vehicles 100 to the home or the power grid PG.

[0012] The server 300 is, for example, a server owned by a VPP operator, and is configured to aggregate multiple distributed energy resources (DERs) to realize a VPP. For example, the server 300 causes the electric vehicles 100 electrically connected to the EVSE 200 to perform power management of the power grid PG in response to a request from the power operator. The server 300 may be a computer including a processor 310, a RAM (Random Access Memory) 320, and a storage device 330. The server 300 is configured to enable communication between each EVSE 200 in the EVSE group 2 and each electric vehicle 100 in the vehicle group 1 via a network NW. The server 300 corresponds to an example of a "power management device" in the present disclosure.

[0013] Identification information (vehicle ID) of each electric vehicle 100 included in the vehicle group 1 is registered in advance in the server 300. The storage device 330 of the server 300 stores information (vehicle information) related to the electric vehicle 100, distinguishing them by vehicle ID. The vehicle information includes basic data such as the specifications of the electric vehicle 100 (for example, rated charging and discharging power) and behavior history data. The behavior history data is associated with date and time, and includes travel data (start position, return position, travel route, travel distance, SOC consumption, etc.), charge and discharge performance, etc.

[0014] The electric vehicle 100 is an electric vehicle (xEV) that includes a battery and an ECU (Electronic Control Unit) and is configured to be able to travel using power stored in the battery. The ECU of the electric vehicle 100 is capable of communicating with a server 300 via a network NW and transmits behavior history data to the server 300. The battery is charged and discharged by connecting a connector provided at the end of a charge / discharge cable of the EVSE 200 to an inlet of the electric vehicle 100. In this embodiment, connecting the connector of the EVSE 200 to the inlet of the electric vehicle 100 is referred to as plugging in, and opening the connection between the connector and the inlet is also referred to as plugging out.

[0015] 2 is a functional block diagram configured in the server 300 in this embodiment. These functions are formed by the processor 310, RAM 320, and storage device 330 working together. The receiving unit 301 receives vehicle information from the electric vehicle 100 (vehicle group 1). The received vehicle information is stored in the database 302. As a result, the behavior history of the electric vehicle is accumulated in the database 302. The receiving unit 301 also receives a demand response (DR) request from the power company.

[0016] When the receiving unit 301 receives a DR request signal, the extracting unit 303 extracts behavioral histories from the database 302 for each vehicle ID. The determining unit 304 determines whether the current behavioral history, for example, the behavioral history within 24 hours from the time of extraction, has changed significantly from the previous behavioral history. For example, if the plug-in location (the location where the vehicle was plugged in) in the current behavioral history is different from the previous plug-in location, the determining unit 304 determines that the current behavioral history has changed by more than a tolerance value from the previous behavioral history, and that the behavioral history has changed significantly. If the battery's State of Charge (SOC) in the current behavioral history differs by 30% or more from the average SOC value in the previous behavioral history, the determining unit 304 determines that the current behavioral history has changed by more than a tolerance value from the previous behavioral history, and that the behavioral history has changed significantly. If the departure time (plug-out time) in the current behavioral history differs by more than 3 hours from the departure time on that day in the previous behavioral history, the determining unit 304 determines that the current behavioral history has changed by more than a tolerance value from the previous behavioral history, and that the behavioral history has changed significantly. Alternatively, the travel route or travel distance may be used to determine whether the current behavior history has changed significantly from the past behavior history. The tolerance is set appropriately depending on each parameter.

[0017] When the determination unit 304 determines, for example, three times in a row, that the current behavior history has changed significantly from the past behavior history (changed by more than an allowable value), the weight determination unit 305 sets a higher weight for the current and subsequent behavior histories relative to the past behavior history. The weight is a weighting of a learning model (inference model). When the current behavior history has not changed significantly from the past behavior history, the weight (weighting) is not changed (the weights for the past and subsequent behavior histories are the same). Note that, when the current behavior history has changed significantly from the past behavior history, the weight determination unit 305 may confirm with the user of the electric vehicle 100 that the current behavior history will be repeated in the future, and if the current behavior history will be repeated in the future, set a higher weight for the current and subsequent behavior histories relative to the past behavior history.

[0018] The learning model 306 is a trained inference model and a behavior prediction model. The learning model 306 may be, for example, a behavior prediction model obtained by machine learning including deep learning. The learning model 306 estimates the behavior of the electric vehicle 100 based on the past behavior history extracted by the extraction unit 303. The behavior prediction is performed using weights set by the weight determination unit 305. The weights may be weightings from the input layer to the intermediate layer (hidden layer), for example. When the weight determination unit 305 sets the weights of the current and subsequent behavior histories to be larger than those of the past behavior history, the state in which the weights of the current and subsequent behavior histories are set to be larger continues until the next time the determination unit 304 determines that the current behavior history has changed significantly from the past behavior history (changed by more than an allowable value).

[0019] When the learning model 306 estimates the behavior predictions of all the electric vehicles 100 included in the vehicle group 1, the charge / discharge plan creation unit 307 creates a charge / discharge plan based on a DR request from the electric power company. The request unit 308 makes a charge / discharge request to each electric vehicle 100 based on the charge / discharge plan. The charge / discharge request may be, for example, a plug-in time, a charge / discharge power amount, a plug-out time, etc.

[0020] FIG. 3 is a flowchart showing an example of a charge / discharge plan creation process processed by the server 300. This flowchart starts when a DR request is made by the electric power company. In step (hereinafter, step is abbreviated as "S") 10, a behavioral history is extracted from the database 302 for each vehicle ID. In S11, it is determined whether the current behavioral history, for example, the behavioral history within 24 hours from the time of extraction, has changed significantly from the past behavioral history among the extracted behavioral history. If the current behavioral history has changed by more than an allowable value from the past behavioral history and it is determined that the behavioral history has changed significantly, a positive determination is made and the process proceeds to S12. If the current behavioral history has not changed by more than an allowable value from the past behavioral history, a negative determination is made and the process proceeds to S14.

[0021] In S12, it is determined whether the significant change in the current behavior history from the past behavior history (change of more than an allowable value) is a continuous change. For example, if a positive determination is made three times in a row in S11, a positive determination is made and the process proceeds to S13. If a negative determination is made, the process proceeds to S14. Note that the user of electric vehicle 100 may be asked to confirm that the current behavior history will be repeated in the future, and when it is confirmed that the current behavior history will be repeated in the future, a positive determination may be made and the process proceeds to S13.

[0022] In S13, a larger weight is set for the current and subsequent behavioral histories relative to the past behavioral histories, and the process proceeds to S14. In S14, the behavioral prediction of the electric vehicle 100 is estimated using the learning model 306. When S13 is processed, the behavioral prediction is performed using the weight set in S13. When S13 is not processed, the weight (weighting) is not changed (the weights for the past and subsequent behavioral histories are the same). When a larger weight is set for the current and subsequent behavioral histories relative to the past behavioral histories in S13, the state in which the weight for the current and subsequent behavioral histories is set large continues until the next time in S13, it is determined that the current behavioral history has changed significantly from the past behavioral history (changed by more than an allowable value).

[0023] In S15, it is determined whether or not the estimation of the predicted behavior of all the electric vehicles 100 included in the vehicle group 1 has been completed. When the estimation of the predicted behavior of all the electric vehicles 100 has been completed, in S16, a charge / discharge plan is created based on the DR request from the electric power company, and then the process proceeds to S17. In S17, a charge / discharge request is made to each electric vehicle 100 based on the charge / discharge plan created in S16.

[0024] According to this embodiment, when the behavioral mode of electric vehicle 100 changes significantly and the current behavior history of electric vehicle 100 changes from the past behavior history by more than an allowable value, the weight of the behavior history from the current time onwards is increased so that the behavior history from the current time onwards is more significantly reflected in the behavior prediction compared to the past behavior history from the current time onwards, and behavior prediction is performed using learning model 306. Even when the behavioral mode of electric vehicle 100 changes significantly, the accuracy of the behavior prediction used when creating a charge / discharge plan can be improved.

[0025] In the above embodiment, when the current behavior history of electric vehicle 100 differs from past behavior history by more than an allowable value, the weight of the behavior history from the current time onwards is increased so that the behavior history from the current time onwards is more significantly reflected in the behavior prediction compared to past behavior history from the current time onwards, and behavior prediction is performed using learning model 306. Even if the current behavior history of electric vehicle 100 does not differ from past behavior history by more than an allowable value, a weight (weighting) may be set so that past behavior history having a temperature similar to the current temperature is more significantly reflected when behavior prediction is performed using the learning model.

[0026] Furthermore, if the current behavior history of the electric vehicle 100 has changed from the past behavior history by more than an allowable value, behavior prediction may be performed from the learning model 306 using only the behavior history from this time onwards (without using the past behavior history).

[0027] The embodiments disclosed herein should be considered to be illustrative in all respects and not restrictive. The scope of the present disclosure is defined by the claims, not by the description of the above embodiments, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0028] 1 Vehicle group, 2 EVSE group, 100 Electric vehicle, 200 EVSE, 300 Server, 301 Receiving unit, 302 Database, 303 Extraction unit, 304 Judgment unit, 305 Weight determination unit, 306 Learning model, 307 Charging and discharging plan creation unit, 308 Request unit, 310 Processor, 320 RAM, 330 Storage device.

Claims

[Claim 1] 1. A power management device that predicts a behavior of an electric vehicle from information including a behavior history of the electric vehicle, and creates a charge / discharge plan for the electric vehicle based on the behavior prediction, The behavior prediction is performed using a learning model, a determination unit that determines whether a current behavior history of the electric vehicle has changed by an allowable value or more from a past behavior history; If the current behavior history has changed from past behavior history by more than an allowable value, the power management device increases the weight of the behavior history from this time onwards so that the behavior history from this time onwards is more significantly reflected in the behavior prediction compared to past behavior history from this time onwards, and performs the behavior prediction using the learning model.

Images