WO2023120784A1

WO2023120784A1 - Adversarial learning system for electric vehicle charging infrastructure power distribution

Info

Publication number: WO2023120784A1
Application number: PCT/KR2021/019835
Authority: WO
Inventors: 홍충선; 쉬라줌 무니르엠디.; 김민석
Original assignee: 경희대학교 산학협력단
Priority date: 2021-12-21
Filing date: 2021-12-24
Publication date: 2023-06-29
Also published as: KR20230094608A

Abstract

A disclosed electric vehicle charging infrastructure power distribution system according to one embodiment of the present invention may comprise: a charging information collection unit which is provided at each of a plurality of charging stations and which collects charging-related information from an electric vehicle that is to be charged with power at the charging station; a central control unit which receives the charging-related information from a plurality of charging information collection units and which determines a power assignment amount of each charging station by using an artificial intelligence model on the basis of the charging-related information collected at each charging station; and an energy distribution unit for distributing, on the basis of the determined power assignment amount, power amounts assigned to the plurality of charging stations.

Description

Adversarial Learning System for Electric Vehicle Charging Infrastructure Power Distribution -Journal of the Korea Convergence Society Korea Science

The present invention relates to an electric vehicle charging infrastructure power distribution system capable of efficiently distributing power to a plurality of charging stations through adversarial machine learning.

The present invention was supported by the group research support of the Ministry of Science and ICT (Task number: 1711135177, task number: 2020R1A4A1018607, research task name: Meta Federated Learning-based mobile edge computing system core structure development, task management institution: National Research Foundation of Korea, task performing agency : Kyunghee University Industry-University Cooperation Foundation, research period: 2021.06.01.~2022.02.28.) and Ministry of Science and ICT (Ministry of Science and ICT) Information Communication Broadcasting Innovation Talent Fostering (Task number: 1711139517, task number: 2021-0-02068, research task name : AI innovation hub development, task management institution: National Institute of Information and Communications Technology Evaluation and Planning, task executing institution: Korea University, research period: 2021.07.01. ~ 2021.12.31.) Meanwhile, there is no property interest of the Korean government in any aspect of the present invention.

In general, electric vehicles are classified into two types: manned electric vehicles and fully autonomous electric vehicles. In addition, in the electric vehicle charging infrastructure, the electric vehicle is connected to the system through wireless communication technology. Because the charging requirements and actual behavior of EVs are different, it is essential to consider the behaviors of both manned EVs and fully autonomous EVs in order to capture the charging behavior of EVs.

Fully self-driving electric vehicles can sense their environment, move safely with little or no human input, and have powerful computing power and large data storage capable of making autonomous decisions on their own. Therefore, the fully autonomous electric vehicle estimates the exact energy demand for the EVSE at startup for the Electric Vehicle Supply Equipment (EVSE) selected by the Distribution System Operator (DSO) for self-analysis and charging. can request

On the other hand, manned electric vehicles involve humans in communicating information such as traffic information wirelessly with other vehicles, infrastructure, and devices. Therefore, there is no guarantee that human interaction is required to request energy demand, and manned electric vehicles require the exact amount of energy and charging time for vehicle charging. Additionally, it is entirely up to the driver whether or not to charge at the EVSE chosen by the DSO for vehicle charging. Due to this, there is a problem that a discrepancy may occur between the planned EVSE of the DSO and the EVSE actually used for charging.

Therefore, when determining the EV charging schedule, it is necessary for the DSO to reasonably consider the charging behavior of manned EVs so that the DSO can maximize the use of available power. In other words, in order to maximize the overall power use of all EVSEs, a power distribution system that can allocate energy, charging time, energy transfer rate, and other factors for the charging behavior of all EVSEs to EVSEs and maximize the overall power utilization of EVSEs is required. need.

The present invention is to provide an electric vehicle charging infrastructure power distribution system capable of maximizing the overall electricity utilization by increasing the use of each EVSE energy as a whole.

In addition, the present invention is to provide an electric vehicle charging infrastructure power distribution system that can more actively and efficiently overcome the irrational electric vehicle charging scheduling problem by processing an irrational charging session request of a manned electric vehicle driven by a human in a data information method. it is for

In addition, the present invention is to provide an electric vehicle charging infrastructure power distribution system capable of improving the charging speed in the overall charging infrastructure and greatly increasing the charging session fulfillment rate compared to the conventional electric vehicle charging system.

An electric vehicle charging infrastructure power distribution system according to an aspect of the disclosed invention includes a charging information collection unit provided at a plurality of charging stations and configured to collect charging-related information from an electric vehicle to be charged at the charging station; A central control unit configured to receive the charging-related information from the plurality of charging information collection units and to determine a power allocation amount of each charging station by using an artificial intelligence model based on the charging-related information collected for each charging station; and an energy distribution unit configured to distribute the amount of power allocated to the plurality of charging stations based on the determined power allocation amount.

In addition, a machine learning unit configured to learn the artificial intelligence model through a machine learning method by setting the charging related information as an input variable and setting the power allocation amount of the plurality of charging stations as an output variable may be further included.

In addition, the charging information collection unit: receives requested energy information and requested charging speed information from the electric vehicle; And it can be configured to collect the actual energy information and actual charging speed information charged by the electric vehicle.

In addition, the central control unit: calculates a requested charging time of the electric vehicle based on the requested energy information and the requested charging speed information; calculating an actual charging time of the electric vehicle based on the actual energy information and the actual charging speed information; And it may be configured to calculate a charging time error for each charging activity based on the requested charging time and the actual charging time.

In addition, the machine learning unit sets the charging time error as an input variable and sets the power allocation amount of the plurality of charging stations as an output variable so as to deliver the maximum amount of electricity to the plurality of electric vehicles. It can be configured to learn an intelligence model.

The vehicle may further include a self-driving electric vehicle classification unit configured to classify the electric vehicle that has delivered the charging-related information as a non-self-driving electric vehicle driven directly by a human when the charging time error is greater than or equal to a reference time error.

In addition, the machine learning unit: sets the requested energy information, requested charging speed information, requested charging time, and charging time error of the plurality of non-autonomous electric vehicles as input variables, and transmits the maximum amount of power to the plurality of electric vehicles. It may be configured to learn the artificial intelligence model through a machine learning method by setting the power allocation of the plurality of charging stations as an output variable to do so.

In addition, the central control unit: based on the location information of the plurality of self-driving electric vehicles, the location information of the plurality of self-driving electric vehicles, the current time information, and the location information of the plurality of charging stations, each using an artificial intelligence model. It may be configured to determine the power allocation of the charging station of the.

In addition, the artificial intelligence model includes an adversarial learning neural network that receives risk adversarial information as an input, and the machine learning unit includes: the requested energy information and the requested charging speed information of the plurality of self-driving electric vehicles. , generate the danger adversarial information based on the actual energy information and the actual charging rate information; And it may be configured to perform adversarial machine learning on the artificial intelligence model so that the maximum amount of power is delivered to the plurality of electric vehicles through past power distribution histories stored in the adversarial learning neural network.

In addition, the artificial intelligence model includes a recurrent neural network that receives the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information as inputs, and the machine learning unit includes the It may be configured to learn the artificial intelligence model so that the maximum amount of power is delivered to the plurality of electric vehicles through past power distribution histories stored in the recurrent neural network.

An electric vehicle charging infrastructure power distribution method according to an aspect of the disclosed invention includes the steps of collecting charging-related information from an electric vehicle to be charged at a charging station by a charging information collection unit provided at each of a plurality of charging stations; Receiving, by a central control unit, the charging-related information from a plurality of charging information collection units, and determining a power allocation of each charging station using an artificial intelligence model based on the charging-related information collected for each charging station. ; and distributing, by an energy distribution unit, an amount of power allocated to the plurality of charging stations based on the determined power allocation amount.

The method may further include, by a machine learning unit, learning the artificial intelligence model through a machine learning method by setting the charging related information as an input variable and setting the power allocation amount of the plurality of charging stations as an output variable. .

The collecting of the charging-related information may include receiving, by the charging information collection unit, requested energy information and requested charging speed information from the electric vehicle; and collecting actual energy information and actual charging speed information charged by the electric vehicle by the charging information collection unit.

In addition, calculating, by the central control unit, a requested charging time of the electric vehicle based on the requested energy information and the requested charging speed information; calculating, by the central control unit, an actual charging time of the electric vehicle based on the actual energy information and the actual charging speed information; and calculating, by the central controller, a charging time error for each charging activity based on the requested charging time and the actual charging time.

In addition, in the step of learning the artificial intelligence model, the machine learning unit sets the charging time error as an input variable and determines the power allocation of the plurality of charging stations to deliver the maximum amount of power to the plurality of electric vehicles. It may include setting as an output variable and learning the artificial intelligence model through a machine learning method.

The method may further include classifying, by a self-driving electric vehicle classification unit, an electric vehicle that has delivered the charging-related information as a non-self-driving electric vehicle directly driven by a person if the charging time error is greater than or equal to a reference time error. .

In addition, in the step of learning the artificial intelligence model, the machine learning unit sets requested energy information, requested charging speed information, requested charging time, and charging time error of the plurality of self-driving electric vehicles as input variables, , learning the artificial intelligence model through a machine learning method by setting the power allocation amount of the plurality of charging stations as an output variable so as to deliver the maximum amount of power to the plurality of electric vehicles.

In addition, the artificial intelligence model includes an adversarial learning neural network that receives risk hostile information as an input, and the step of learning the artificial intelligence model is, by the machine learning unit, the request of the plurality of self-driving electric vehicles. generating the risk hostile information based on energy information, the requested charging speed information, the actual energy information, and the actual charging speed information; and performing adversarial machine learning on the artificial intelligence model by the machine learning unit so that a maximum amount of power is delivered to the plurality of electric vehicles through past power distribution histories stored in the adversarial learning neural network.

In addition, the artificial intelligence model includes a recurrent neural network receiving the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information as inputs, and learning the artificial intelligence model comprises: The method may include learning, by a machine learning unit, the artificial intelligence model so that a maximum amount of power is delivered to the plurality of electric vehicles through past power distribution histories stored in the recurrent neural network.

A non-transitory recording medium according to an aspect of the disclosed invention may be a computer-readable non-transitory recording medium on which a program for executing a method for distributing electric vehicle charging infrastructure power is recorded.

According to one aspect of the disclosed invention, it is possible to maximize overall electricity utilization by increasing the use of each EVSE energy as a whole.

In addition, according to an embodiment of the present invention, it is possible to more actively and efficiently overcome the irrational electric vehicle charging scheduling problem by processing an irrational charging session request of a manned electric vehicle driven by a human using a data information method.

In addition, according to an embodiment of the present invention, the charging speed in the entire charging infrastructure can be improved and the charging session fulfillment rate can be greatly increased compared to the conventional electric vehicle charging system.

1 is a control block diagram of an electric vehicle charging infrastructure power distribution system according to an embodiment.

2 is a diagram illustrating an electric vehicle charging infrastructure power distribution system according to an embodiment.

Figure 3 is a diagram showing the overall framework of risk adversarial learning (Risk Adversarial Learning) according to an embodiment.

4 is a flowchart of a method of distributing power to each charging station according to an embodiment.

5 is a flowchart of a method for learning an artificial intelligence model according to an embodiment.

6 is a graph showing experimental results of an electric vehicle charging infrastructure power distribution method according to an embodiment.

7 is another graph showing experimental results of an electric vehicle charging infrastructure power distribution method according to an embodiment.

Like reference numbers designate like elements throughout the specification. This specification does not describe all elements of the embodiments, and general content or overlapping content between the embodiments in the technical field to which the disclosed invention belongs is omitted. The term '~unit' used in the specification may be implemented in software or hardware, and according to embodiments, a plurality of '~units' may be implemented as one component, or one '~unit' may constitute a plurality of components. It is also possible to include elements.

In addition, when a certain component is said to "include", this means that it may further include other components without excluding other components unless otherwise stated.

'~ unit' used in this specification is a unit that processes at least one function or operation, and may mean, for example, software, an FPGA, or a hardware component. Functions provided by '~unit' may be performed separately by a plurality of components or may be integrated with other additional components. '~unit' in this specification is not necessarily limited to software or hardware, and may be configured to be in an addressable storage medium or configured to reproduce one or more processors.

Expressions in the singular number include plural expressions unless the context clearly dictates otherwise.

In each step, the identification code is used for convenience of description, and the identification code does not explain the order of each step, and each step may be performed in a different order from the specified order unless a specific order is clearly described in context. there is.

Hereinafter, the working principle and embodiments of the disclosed invention will be described with reference to the accompanying drawings.

Referring to FIG. 1 , an electric vehicle charging infrastructure power distribution system 1 including a plurality of charging stations 100 and a central agent system 200 may be configured.

The charging station 100 may include at least one electric vehicle supply equipment (EVSE) that supplies electric energy to the electric vehicle 300 . Each charging station 100 may be distributed and disposed in several regions.

The charging station 100 may include a charging information collection unit 110, an autonomous driving electric vehicle classification unit 120, and a charging station control unit 130.

The central agent system 200 may be provided in a server managed by a distribution system operator (DSO), but the central agent system 200 does not necessarily have to be provided in the server.

Meanwhile, the central agent system 200 may communicate with a plurality of charging stations 100 through wired and wireless communication through a communication module.

The central agent system 200 learns the artificial intelligence model 231 based on past electric vehicle charging requests, charging cases, and power distribution history when there were such charging cases, and uses the learned artificial intelligence model 231. Therefore, it is possible to determine the power allocation for each charging station based on the current information.

The central agent system 200 may include a central control unit 210, a machine learning unit 220, and a memory 230.

The charging information collection unit 110 is provided in each of the plurality of charging stations 100 and may collect charging-related information from the electric vehicle 300 that wants to charge power at the corresponding charging station 100 .

Charging-related information may include requested energy information and requested charging speed information transmitted from the side of the electric vehicle 300 that wants to be charged. In addition, the charging-related information may be information on actual energy and actual charging speed charged by the corresponding electric vehicle 300 . That is, the charging-related information may be information about the amount of power requested by the electric vehicle 300 to be charged, the requested charging speed, the amount of power actually charged, and the actual charging speed.

That is, the charging information collection unit 110 may receive the requested energy information and the requested charging speed information from the electric vehicle 300, and collect the actual energy information and actual charging speed information charged by the electric vehicle 300. .

The central control unit 210 may receive charging-related information from a plurality of charging information collection units 110 respectively provided in the plurality of charging stations 100 .

The central control unit 210 may determine the power allocation amount of each charging station 100 using the artificial intelligence model 231 based on the charging-related information collected for each charging station 100 .

Specifically, the central control unit 210 uses the artificial intelligence model 231 based on the location information of the plurality of electric vehicles 300, the current time information, and the location information of the plurality of charging stations 100 to charge each charging station 100. ) can determine the power allocation.

The energy distribution unit may distribute the amount of power allocated to the plurality of charging stations 100 based on the determined power allocation amount.

The machine learning unit 220 may generate the artificial intelligence model 231 through a machine learning method based on a large amount of charging-related information and past power allocation information collected in the past. Meanwhile, the artificial intelligence model 231 may be stored in the memory 230 provided in the central agent system 200, but the artificial intelligence model 231 does not necessarily have to be stored in the central agent system 200, and each charging station The same artificial intelligence models 231 may be stored in the system of (100).

The machine learning unit 220 may be configured to learn the artificial intelligence model 231 through a machine learning method by setting the charging related information as an input variable and the power allocation of the plurality of charging stations 100 as an output variable.

Machine learning may mean using a model composed of multiple parameters and optimizing the parameters with given data. Machine learning may include supervised learning, unsupervised learning, and reinforcement learning depending on the form of a learning problem. Supervised learning is to learn the mapping between inputs and outputs, and can be applied when input and output pairs are given as data. Unsupervised learning is applied when there are only inputs and no outputs, and regularities between inputs can be found.

However, machine learning according to an embodiment is not necessarily limited to the aforementioned learning method. For example, machine learning may include risk adversarial learning. Adversarial learning can be a learning method that trains neural networks on how to spot intentionally misleading data or behavior.

The machine learning unit 220 may learn the artificial intelligence model 231 in various ways. For example, the machine learning unit 220 features features extracted from requested energy information, requested charging speed information, actual energy information, actual charging speed information, and past power allocation history included in a plurality of past charging related information. ) can be learned with a deep learning-based learning method. In this case, a CNN (Convolutional Neural Networks) structure in which several convolutional layers are stacked may be used to learn a method of extracting features from past charging-related information and past power allocation history, but a machine learning unit may be used. The learning method of (220) is not necessarily limited to a method using a CNN structure.

For example, the learning method of the machine learning unit 220 may be a method through a machine learning algorithm including a recurrent neural network (RNN).

That is, the artificial intelligence model 231 may include a recurrent neural network that receives the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information as inputs. A recurrent neural network is a deep learning model for learning data that changes over time, such as time-series data, and may be an artificial neural network configured by connecting networks at a reference point in time and a next point in time.

The machine learning unit 220 may learn the artificial intelligence model 231 so that as much power as possible is delivered to the plurality of electric vehicles 300 for a predetermined reference period through past power distribution histories stored in the recurrent neural network.

The charging information collection unit 110, the self-driving electric vehicle classification unit 120, the charging station control unit 130, the central control unit 210, and the machine learning unit 220 are included in the electric vehicle charging infrastructure power distribution system 1. It may include any one processor among a plurality of processors.

In addition, the electric vehicle charging infrastructure power distribution method according to the embodiments of the present invention described so far and the embodiments to be described in the future may be implemented in the form of a program that can be driven by a processor.

Here, the program may include program commands, data files, and data structures alone or in combination. The program may be designed and manufactured using machine language codes or high-level language codes. The program may be specially designed to implement the above-described code correction method, or may be implemented using various functions or definitions that are known and usable to those skilled in the art in the field of computer software. A program for implementing the above information display method may be recorded on a non-transitory recording medium readable by a processor. At this time, the non-transitory recording medium may be the memory 230 .

The memory 230 may store a program for performing the above-described operation and an operation to be described later, and the memory 230 may execute the stored program. When there are a plurality of processors and memories 230, they may be integrated into one chip or may be provided in physically separate locations. The memory 230 may include a volatile memory such as static random access memory (S-RAM) or dynamic random access memory (D-lab) for temporarily storing data. In addition, the memory 230 may include a ROM (Read Only Memory), an Erasable Programmable Read Only Memory (EPROM), and an Electrically Erasable Programmable Read Only Memory (EEPROM) for long-term storage of control programs and control data. It may contain non-volatile memory.

The processor may include various logic circuits and arithmetic circuits, process data according to programs provided from the memory 230, and generate control signals according to processing results.

Referring to FIG. 2 , the central control unit 210 provided in the central agent system 200 may be a DSO controller managed by a distribution system operator.

The illustrated electric vehicle charging infrastructure power distribution system 1 is applied to an electric vehicle (EV) 300 including both an electric vehicle 300 driven by a human and an electric vehicle 300 capable of fully autonomous driving without human intervention. Electric charging facilities may be included.

The DSO controller may be connected to an energy source to energize each electric vehicle supply equipment (EVSE) with a fixed power capacity.

The electric vehicle 300 may request power charging through V2X communication at an average charging speed of each time zone. At this time, each electric vehicle 300 may request a certain amount of energy and usable charging time from any one electric vehicle supply equipment.

Referring to FIG. 3 , a framework of a Risk Adversarial Learning System for a non-autonomous electric vehicle driven by a person and a charging infrastructure for an autonomous electric vehicle can be confirmed. The framework of the risk adversarial learning system is described below.

First, the electric vehicle 300 may request charging of the electric vehicle supply equipment (EVSE) (201A-201N). The charging request of each electric vehicle 300 may consist of energy demand and session duration, but may be configured differently according to the required number of request parameters.

The EVs Charging Data 202A-202N of the electric vehicle 300 may include all types of information necessary for the charging schedule of the electric vehicle 300. For example, the charging-related information may include energy demand of the electric vehicle 300, session duration request, charging start time, charging end time, electric vehicle disconnection time, and actual delivered energy.

Each electric vehicle supply equipment can estimate its own compensation based on the current charging request of the electric vehicle 300 and the adversarial risk of the Centralized Risk Adversarial Agent (RAA) that analyzes the inaccurate laxity. there is.

That is, the machine learning unit 220 is not only present in the central agent system 200, but may also be provided in the electric vehicle supply equipment of each charging station 100, and each electric vehicle supply equipment is reward-based hostile risk (Reward). Based on Adversarial Risk) (203A-203N).

The learning agent (EVSE-LA) of each electric vehicle supply equipment can learn and discretize its own EV session scheduling policy in response to the unreasonable risk of adversarial agents. While the learning agent for each electric vehicle supply equipment has its own Markov Decision Process (MDP), the decision of each EV session schedule (e.g. schedule indicator, expected energy demand, delivery rate, and charging duration) is dependent on the risk of inaccuracies in the CBaR-Tail distribution. It can be affected by laxity. This compensation can be used for optimal scheduling decisions and charging policy decisions by the electric vehicle supply equipment. In addition, the central risk adversarial agent can use all rewards for analysis on adversarial laxity.

Each Electric Vehicle Supply Equipment (EVSE) may determine the EVSE Agent Charging Policy (204A-204N) according to the advice of the Central Risk Adversarial Agent (RAA). For example, if a recurrent neural network (RNN) having the same configuration is used, it is possible to collect characteristics dependent on a specific time of an uncertain operation in response to a charging request of the electric vehicle 300 . Additionally, power allocation policies can be established by determining reservations or queues, expected energy demand, energy delivery rate, and charging duration.

The charging infrastructure power distribution system according to an embodiment may perform characterization of charging behaviors (205). Among the electric vehicles 300, the fully autonomous driving electric vehicle (AV) may estimate and request an accurate amount of energy and charging time for the electric vehicle supply system (EVSE) through self-analysis. Therefore, since the difference between the amount of energy required and the actually charged energy is minimized, the energy and charging time allocated to the fully self-driving electric vehicle can meet the exact demand. In addition, since the difference between the requested charging time and the actual charging time can be minimized, the electric vehicle supply device can maximize the utilization of energy resources, so the charging schedule and power distribution to the fully autonomous electric vehicle can be rationalized.

On the other hand, non-autonomous electric vehicles driven by humans have a problem in that there is no guarantee that an accurate amount of energy and charging time are required from an electric vehicle supply device due to manual interaction of individuals.

A rational charging session scheduling decision for the electric vehicle 300 is related not only to the ratio of the actual charging time to the total connection time, but also to the ratio between the amount of energy delivered by the electric vehicle supply and the requested energy of the electric vehicle 300. . Therefore, estimating the inaccuracy (looseness) of the DSO controller, i.e., the central control unit 210, may be one of the methods suitable for characterizing the charging operation.

[Equation 1]

Referring to [Equation 1], the laxity of each EV request is the requested energy information (Requested energy), the requested charging rate information (Demand rate), the actual energy information (Delivered energy), and the actual charging rate information. (Charging rate).

The central control unit 210 may calculate the requested charging time of the electric vehicle 300 based on the requested energy information and the requested charging speed information. Specifically, the central control unit 210 may calculate the requested charging time by dividing the value of the requested energy information by the value of the requested charging speed information.

The central control unit 210 may calculate the actual charging time of the electric vehicle 300 based on the actual energy information and the actual charging speed information. Specifically, the actual charging time may be calculated by dividing the value of the actual energy information by the value of the actual charging speed information.

The central control unit 210 may calculate a charging time error for each charging activity based on the actual charging time of the requested charging time. Specifically, the central control unit 210 may determine a charging time error, which is a value obtained by subtracting the actual charging time from the requested charging time, as the laxity of each EV request.

The machine learning unit 220 sets the charging time error as an input variable and sets the power allocation amount of the plurality of charging stations 100 as an output variable to deliver the maximum amount of power to the plurality of electric vehicles 300, thereby performing a machine learning method. Through this, the artificial intelligence model 231 can be learned.

Meanwhile, as described above, the difference between the amount of energy required and the actual charged energy and the difference between the required charging time and the actual charging time can be minimized. On the other hand, in the case of non-autonomous vehicles, differences in energy and charging time inevitably occur to some extent, and such irrational and unpredictable charging-related information also needs to be reflected in the power distribution process. In summary, the electric vehicle charging infrastructure power distribution system (1) distinguishes between fully autonomous electric vehicles and non-autonomous driving electric vehicles, and sets only the charging-related information of the non-autonomous driving electric vehicles as input variables to perform machine learning. need something

The self-driving electric vehicle classifying unit 120 may classify the electric vehicle 300 that has delivered charging-related information as a non-self-driving electric vehicle directly driven by a person if the charging time error is equal to or greater than the reference time error. The reference time error may be a value of a charging time error that is a standard for distinguishing whether the electric vehicle 300 that has requested charging is a fully autonomous driving electric vehicle or a non-autonomous driving electric vehicle.

The machine learning unit 220 sets the requested energy information, requested charging speed information, requested charging time, and charging time error of the plurality of non-autonomous electric vehicles as input variables, and sets the maximum amount of power to the plurality of electric vehicles 300. The artificial intelligence model 231 may be learned through a machine learning method by setting the power allocation of the plurality of charging stations 100 as an output variable to deliver.

The central control unit 210 generates an artificial intelligence model 231 based on location information of a plurality of non-autonomous driving electric vehicles, location information of a plurality of self-driving electric vehicles, current time information, and location information of a plurality of charging stations 100. It is possible to determine the power allocation of each charging station 100 by using.

The adversarial laxity analysis process 206 of the central risk adversarial agent may be an adversarial machine learning process. The electric vehicle charging infrastructure power distribution system 1 may be a Risk Adversarial Multiple Learning Agent Learning System (RAMALS) in which the charging behavior of the electric vehicle 300 is learned in a data-based manner.

That is, the artificial intelligence model 231 may include an adversarial learning neural network that receives risk adversarial information as an input.

The machine learning unit 220 may generate risk hostile information based on requested energy information, requested charging speed information, actual energy information, and actual charging speed information of a plurality of non-autonomous electric vehicles.

The machine learning unit 220 may perform adversarial machine learning on the artificial intelligence model 231 so that the maximum amount of power is delivered to the plurality of electric vehicles 300 through the past power distribution history stored in the adversarial learning neural network. there is.

The central control unit 210 may analyze inaccurate laxity in the charging operation of the electric vehicle 300 by using Conditional Value at Risk (CVaR). While the decision of each electric vehicle charging session is quantified in the CVaR-Tail risk distribution of laxity, a strong correlation can be established between the energy demand and supply behavior of the charging system. Thus, an inaccurate laxity can cause each electric vehicle supply equipment to meet an efficient electric vehicle charging schedule.

Observed Memory 207 is a vector for each electric vehicle 300 of the electric vehicle supply equipment, and the behavior of the electric vehicles 300 in the past, compensation, past power distribution history, and recurrent neural network (RNN). can be configured. During the information acquisition period, information acquisition may be performed in the learning agent EVSE-LA of the electric vehicle supply equipment.

At least one component may be added or deleted corresponding to the performance of the components described above. In addition, it will be easily understood by those skilled in the art that the mutual positions of the components can be changed corresponding to the performance or structure of the system.

Referring to FIG. 4 , the charging information collection unit 110 provided in each of the plurality of charging stations 100 may collect charging-related information from an electric vehicle 300 that wants to charge power at the charging station 100 (1001 ). At this time, the charging information collection unit 110 may receive requested energy information and requested charging speed information from the electric vehicle 300, and may collect information on actual energy and actual charging speed information charged by the electric vehicle 300.

The machine learning unit 220 sets the charging related information as an input variable and sets the power allocation of the plurality of charging stations 100 as an output variable to learn the artificial intelligence model 231 through a machine learning method. can (1002). At this time, the machine learning unit 220 may learn the artificial intelligence model 231 so that the maximum amount of power is delivered to the plurality of electric vehicles 300 through the past power distribution history stored in the recurrent neural network.

The central control unit 210 receives charging-related information from the plurality of charging information collection units 110 and uses the artificial intelligence model 231 based on the charging-related information collected for each charging station 100 for each charging station. A power allocation amount of (100) may be determined (1003). At this time, the central control unit 210 sets the artificial intelligence model 231 based on the location information of the plurality of non-autonomous driving electric vehicles, the location information of the plurality of self-driving electric vehicles, the current time information, and the location information of the plurality of charging stations 100. ) can be used to determine the power allocation of each charging station 100.

The energy distribution unit may distribute the amount of power allocated to the plurality of charging stations 100 based on the determined power allocation amount (1004).

Referring to FIG. 5 , the central control unit 210 may calculate the requested charging time of the electric vehicle 300 based on the requested energy information and the requested charging speed information (2001).

The central control unit 210 may calculate the actual charging time of the electric vehicle 300 based on the actual energy information and the actual charging speed information (2002).

The central control unit 210 may calculate a charging time error for each charging activity based on the requested charging time and the actual charging time (2003).

If the charging time error is greater than or equal to the reference time error, the electric vehicle classification unit may classify the electric vehicle 300 that has delivered charging-related information as a non-autonomous electric vehicle driven by a person (2004). In this case, if the charging time error is less than the reference time error, the electric vehicle classification unit may classify the electric vehicle 300 that has transmitted the charging-related information as a completely self-driving electric vehicle.

The machine learning unit 220 sets the requested energy information, requested charging speed information, requested charging time, and charging time error of the plurality of non-autonomous electric vehicles as input variables, and sets the maximum amount of power to the plurality of electric vehicles 300. The artificial intelligence model 231 may be learned through a machine learning method by setting the power allocation of the plurality of charging stations 100 as an output variable to deliver (2005).

In order to verify the performance of the electric vehicle charging infrastructure power distribution method according to an embodiment of the present invention, a power distribution simulation experiment was conducted.

6 is a graph showing experimental results of an electric vehicle charging infrastructure power distribution method according to an embodiment, and FIG. 7 is another graph showing experimental results of an electric vehicle charging infrastructure power distribution method according to an embodiment.

Referring to FIG. 6 , it can be confirmed that the electric vehicle charging rate is improved by the performance of the proposed risk adversarial multi-agent learning system (RAMALS), that is, the electric vehicle charging infrastructure power distribution system 1 .

Compared with the conventional A3C-based framework, A2C-based model and ACN data, the proposed electric vehicle charging infrastructure power distribution system (1) will improve the electric vehicle charging rate of Caltech EVSE site by about 71.4%, 40% and 46.6%, respectively. can confirm that it can.

This improvement in charging speed occurs due to rational scheduling by the electric vehicle charging infrastructure power distribution system (1), and may be due to improving the charging capacity of the EVSE site with data-based scheduling that significantly reduces the idle charging time of the electric vehicle supply equipment. there is.

Referring to FIG. 7 , a graph in which performance is compared based on the total active EV charging time of the JPL EVSE site from October 16, 2019 to December 31, 2019 can be seen.

Specifically, referring to the graph, it can be seen that the total active charging time of 52 EVSE included in the JPL EVSE site can be confirmed, and the active charging time can be improved from 8 hours to 40 hours compared to the conventional power distribution method for the JPL EVSE site can confirm.

Since the electric vehicle charging infrastructure power distribution system (1) can handle the uncertain electric vehicle charging demand during scheduling of each electric vehicle supply equipment using data-informed rational decisions from prior knowledge, the proposed electric vehicle charging infrastructure It can be seen that the performance of the power distribution system 1 is higher than that of the conventional A3C-based framework and A2C-based system. Specifically, it can be confirmed that the proposed electric vehicle charging infrastructure power distribution system 1, that is, RAMALS, can significantly improve (ie, about 28.6%) the active charging time of each electric vehicle supply equipment.

As above, the disclosed embodiments have been described with reference to the accompanying drawings. Those skilled in the art to which the present invention pertains will understand that the present invention can be implemented in a form different from the disclosed embodiments without changing the technical spirit or essential features of the present invention. The disclosed embodiments are illustrative and should not be construed as limiting.

Claims

a charging information collection unit provided at each of a plurality of charging stations and configured to collect charging-related information from an electric vehicle to be charged at the charging station;

A central control unit configured to receive the charging-related information from the plurality of charging information collection units and to determine a power allocation amount of each charging station by using an artificial intelligence model based on the charging-related information collected for each charging station; and

An electric vehicle charging infrastructure power distribution system comprising an energy distribution unit configured to distribute the amount of power allocated to the plurality of charging stations based on the determined power allocation amount.
According to claim 1,

The electric vehicle charging infrastructure power distribution system further comprising a machine learning unit configured to learn the artificial intelligence model through a machine learning method by setting the charging related information as an input variable and setting the power allocation amount of the plurality of charging stations as an output variable. .
According to claim 2,

The charging information collection unit:

Receiving requested energy information and requested charging speed information from the electric vehicle; and

An electric vehicle charging infrastructure power distribution system configured to collect actual energy information and actual charging speed information charged by the electric vehicle.
According to claim 3,

The central control unit:

calculating a requested charging time of the electric vehicle based on the requested energy information and the requested charging speed information;

calculating an actual charging time of the electric vehicle based on the actual energy information and the actual charging speed information; and

An electric vehicle charging infrastructure power distribution system configured to calculate a charging time error for each charging activity based on the requested charging time and the actual charging time.
According to claim 4,

The machine learning unit,

Configured to learn the artificial intelligence model through a machine learning method by setting the charging time error as an input variable and setting the power allocation amount of the plurality of charging stations as an output variable to deliver the maximum amount of power to the plurality of electric vehicles , electric vehicle charging infrastructure power distribution system.
According to claim 4,

An electric vehicle charging infrastructure power distribution system further comprising an autonomous driving electric vehicle classification unit configured to classify an electric vehicle that has delivered the charging-related information as a non-autonomous driving electric vehicle directly driven by a person if the charging time error is greater than or equal to a reference time error. .
According to claim 6,

The machine learning unit:

The power of the plurality of charging stations is set so that the requested energy information, the requested charging speed information, the requested charging time, and the charging time error of the plurality of self-driving electric vehicles are set as input variables, and the maximum amount of power is delivered to the plurality of electric vehicles. An electric vehicle charging infrastructure power distribution system configured to learn the artificial intelligence model through a machine learning method by setting a quota as an output variable.
According to claim 7,

The central control unit:

Determining the power allocation of each charging station using an artificial intelligence model based on the location information of the plurality of self-driving electric vehicles, the location information of the plurality of self-driving electric vehicles, the current time information, and the location information of the plurality of charging stations. An electric vehicle charging infrastructure power distribution system configured to do so.
According to claim 8,

The artificial intelligence model,

Including an adversarial learning neural network that receives risk adversarial information as an input,

The machine learning unit:

generating the risk hostile information based on the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information of the plurality of non-autonomous electric vehicles; and

An electric vehicle charging infrastructure power distribution system configured to perform adversarial machine learning on the artificial intelligence model so that the maximum amount of power is delivered to the plurality of electric vehicles through the past power distribution history stored in the adversarial learning neural network.
According to claim 3,

The artificial intelligence model,

A recurrent neural network receiving the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information as inputs,

The machine learning unit,

The electric vehicle charging infrastructure power distribution system configured to learn the artificial intelligence model so that the maximum amount of power is delivered to the plurality of electric vehicles through the past power distribution history stored in the recurrent neural network.
Collecting charging-related information from an electric vehicle to be charged at the charging station by a charging information collection unit provided at each of the plurality of charging stations;

Receiving, by a central control unit, the charging-related information from a plurality of charging information collection units, and determining a power allocation of each charging station using an artificial intelligence model based on the charging-related information collected for each charging station. ; and

and distributing, by an energy distributor, an amount of power allocated to the plurality of charging stations based on the determined power allocation amount.
According to claim 11,

The electric vehicle charging infrastructure further comprising learning the artificial intelligence model through a machine learning method by a machine learning unit by setting the charging related information as an input variable and setting the power allocation of the plurality of charging stations as an output variable. Power distribution method.
According to claim 12,

The step of collecting the charging-related information,

receiving requested energy information and requested charging speed information from the electric vehicle by the charging information collection unit; and

The electric vehicle charging infrastructure power distribution method comprising the step of collecting actual energy information and actual charging speed information charged by the electric vehicle by the charging information collection unit.
According to claim 13,

calculating, by the central control unit, a requested charging time of the electric vehicle based on the requested energy information and the requested charging speed information;

calculating, by the central control unit, an actual charging time of the electric vehicle based on the actual energy information and the actual charging speed information; and

The electric vehicle charging infrastructure power distribution method further comprising calculating, by the central control unit, a charging time error for each charging activity based on the requested charging time and the actual charging time.
According to claim 14,

The step of learning the artificial intelligence model,

The machine learning unit sets the charging time error as an input variable and sets the power allocation amount of the plurality of charging stations as an output variable so as to deliver the maximum amount of electricity to the plurality of electric vehicles. An electric vehicle charging infrastructure power distribution method comprising the step of learning an intelligent model.
According to claim 14,

Classifying, by a self-driving electric vehicle classification unit, an electric vehicle that has delivered the charging-related information as a non-autonomous driving electric vehicle directly driven by a person if the charging time error is greater than or equal to a reference time error; Power distribution method.
According to claim 16,

The step of learning the artificial intelligence model,

The machine learning unit sets requested energy information, requested charging speed information, requested charging time, and charging time error of the plurality of self-driving electric vehicles as input variables, and transmits the maximum amount of power to the plurality of electric vehicles. And setting the power allocation of the plurality of charging stations as an output variable to learn the artificial intelligence model through a machine learning method.
According to claim 17,

The artificial intelligence model,

It includes an adversarial learning neural network that receives risk adversarial information as an input,

The step of learning the artificial intelligence model,

generating, by the machine learning unit, the risk hostile information based on the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information of the plurality of self-driving electric vehicles; and

and adversarial machine learning of the artificial intelligence model so that the maximum amount of power is delivered to the plurality of electric vehicles through the past power distribution history stored in the adversarial learning neural network, by the machine learning unit. Power distribution method.
According to claim 13,

The artificial intelligence model,

A recurrent neural network receiving the requested energy information, the requested charging speed information, the actual energy information, and the actual charging speed information as inputs;

The step of learning the artificial intelligence model,

Learning, by the machine learning unit, the artificial intelligence model so that the maximum amount of power is delivered to the plurality of electric vehicles through past power distribution histories stored in the recurrent neural network, an electric vehicle charging infrastructure power distribution method comprising: .
A computer-readable non-transitory recording medium on which a program for executing the electric vehicle charging infrastructure power distribution method of claim 11 is recorded.