EV intelligent real-time scheduling method and system based on vehicle-station-network information interaction

Abstract

This invention relates to an intelligent real-time scheduling method and system for electric vehicles (EVs) based on vehicle-station-grid information interaction. The scheduling method includes the following: acquiring charging and discharging information related to electric vehicle users (EVUs); obtaining charging station selection indicators, including: the shortest path distance from the EVU's starting point to each charging station, power consumption, travel time, maximum waiting time, and utilization rate; obtaining the grid's original load information, real-time electricity price information, and real-time carbon emission factor information for the next 24 hours after the current moment, as charging and discharging optimization indicators; constructing an indicator system using charging station selection indicators and charging and discharging optimization indicators; assuming that charging and discharging cannot occur simultaneously at the same charging station, constructing the objective function and constraints of an intelligent real-time scheduling model for electric vehicles; solving the intelligent real-time scheduling model for electric vehicles based on the different objective preferences of EVUs and the grid, obtaining multiple recommended schemes with different preferences. This achieves dual optimization of EVU's selection of charging stations and EVU participation in charging and discharging scheduling.

Description

Technical Field

[0001] This invention relates to the fields of carbon emissions and intelligent scheduling of electric vehicles, specifically a method and system for intelligent real-time scheduling of electric vehicles (EVs) based on vehicle-station-network information interaction. Background Technology

[0002] With the rapid increase in global greenhouse gas emissions, climate change has become one of the major challenges facing the world today. To address this challenge, countries worldwide are vigorously developing electric vehicles (EVs). Compared to traditional gasoline-powered vehicles, EVs utilize electricity to replace traditional fossil fuels, significantly improving energy conversion efficiency and helping to reduce greenhouse gas emissions, improve air quality, and reduce noise pollution. Simultaneously, EVs can also serve as distributed energy storage facilities, participating in power balance regulation and providing ancillary services. Considering these advantages, the number of EVs has grown rapidly, reaching 6.4 million by the end of 2021. However, with the continuous growth of EV ownership, the indiscriminate selection of charging stations by a large number of EVs impacts the charging economy, convenience, and service quality for electric vehicle users (EVUs); their charging behavior is not properly controlled, and disorderly grid connection affects their safe operation, optimized control, and decarbonization. Therefore, appropriate scheduling methods must be adopted to guide EVUs to select suitable charging stations and conduct orderly charging and discharging, reducing negative impacts on EVUs and the power grid.

[0003] Current intelligent real-time scheduling methods for EVs rely solely on a single indicator to schedule the charging and discharging of EVs. For example, they use time-of-use pricing to incentivize EVs to charge during periods of lower electricity prices to reduce charging costs and discharge during periods of higher prices to generate revenue; or they use marginal carbon emission factors to guide EVs to charge when the values ​​are low to reduce the carbon emissions of the power grid. However, this only benefits either the EV or the power grid and is not optimal for the entire system.

[0004] This application, while meeting the charging and discharging needs of EVUs, establishes an indicator system to guide EVUs in selecting charging stations and performing charging and discharging operations, and achieves this real-time scheduling through deep reinforcement learning. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the technical problem this invention aims to solve is to provide an intelligent real-time scheduling method and system for electric vehicles (EVs) based on vehicle-station-network information interaction. This method and system can address the randomness of large-scale EVs selecting charging stations and connecting to the power grid. It utilizes evaluation indicators of EV charging convenience, economy, and charging service quality, as well as evaluation indicators of power grid safety and environmental friendliness, to establish an intelligent real-time scheduling model for electric vehicles while meeting EV charging and discharging needs. This model is then solved using a deep reinforcement learning algorithm to obtain multiple recommended schemes with different preferences.

[0006] The technical solution of the present invention to solve the aforementioned technical problem is:

[0007] In a first aspect, the present invention provides an intelligent real-time scheduling method for EVs based on vehicle-station-network information interaction, the scheduling method comprising the following:

[0008] Obtain the charging / discharging requests of electric vehicle users (EVUs), and obtain the EV's latitude and longitude, the time T0 of the charging / discharging request, the battery state of charge SCO0, and the EV's battery capacity E. cap EV charging and discharging power P ch / dis Expected State of Charge (SOC) when the user leaves the charging station end EVU's estimated time to leave the charging station (T) l Average speed of electric vehicles information;

[0009] The criteria for selecting charging stations include: the shortest path distance D from the EVU's starting point to each charging station. min Power consumption E c and travel time T min Maximum waiting time T q Utilization rate U j ;

[0010] The original load value L of the power grid, the real-time electricity price REP, and the real-time carbon emission factor REF are predicted for the next 24 hours from the current time. The original load information, real-time electricity price information, and real-time carbon emission factor information of the power grid for the next 24 hours after the current time are obtained. The predicted original load, real-time electricity price, and real-time carbon emission factor are used as charging and discharging optimization indicators.

[0011] An indicator system is constructed using charging station selection indicators and charging / discharging optimization indicators. This indicator system is used to evaluate the convenience, service quality, economy, environmental friendliness, and safety of EVU throughout the entire process from selecting a charging station to the end of charging / discharging.

[0012] Assuming that charging and discharging cannot occur simultaneously at the same charging station, the objective function f1-f8 for constructing the intelligent real-time scheduling model for electric vehicles is given by formulas (18)-(25):

[0013] f1 = T min (18)

[0014] f2 = E c (19)

[0015]

[0016]

[0017] f5 = T q(twenty two)

[0018] f6 = U j (twenty three)

[0019]

[0020]

[0021] Where T is the set of time periods, and REP t This represents the real-time electricity price within time period t, which is obtained through forecasting. This indicates that the i-th EV is charged during time period t, otherwise it is not charged. This indicates that the i-th EV discharges within time period t, otherwise it does not discharge; Δt is the length of a unit time period; η is the charging and discharging efficiency of the charging pile. The charging and discharging power of the i-th EV; P max P is the maximum value of the original load of the power grid plus the charging load. min REF represents the minimum value of the original grid load plus the charging load; t This represents the real-time carbon emission factor within time period t, which is obtained through prediction.

[0022] The constraints of the intelligent real-time scheduling model for electric vehicles are given by formulas (26) and (27):

[0023] |SOC t -SOC end |≤δ (26)

[0024] SOC min +E c ≤SOC0 (27)

[0025] Among them, SOC min The minimum threshold for the state of charge (SOC) of an EV battery. t The battery charge status of an electric vehicle user when picking up the vehicle;

[0026] If the charging / discharging time ΔT of the EV is less than the estimated time T for the EV to leave the charging station. l The time T for the EVU to arrive at the charging station a The difference, and the initial battery state of the EVU arriving at the charging station when it needs to discharge. If it is greater than 55%, the charging and discharging start time of the EV can be optimized, and the optimization can be carried out according to the intelligent real-time scheduling model of electric vehicles.

[0027] Based on the different target preferences of EVU and power grid, indicators in the indicator system are selected to solve the intelligent real-time scheduling model of electric vehicles and obtain multiple recommended schemes with different preferences.

[0028] Furthermore, the intelligent real-time scheduling model for electric vehicles is equivalent to a Markov decision process (MDP). Deep reinforcement learning is used to train and solve the intelligent real-time scheduling model for electric vehicles to obtain multiple recommendation schemes with different preferences. The recommendation schemes include charging station locations, charging / discharging start times, and charging / discharging amounts. The deep reinforcement learning is the Deep Deterministic Policy Gradient (DDPG) algorithm.

[0029] Secondly, the present invention provides an EV intelligent real-time dispatch system based on vehicle-station-network information interaction, the system comprising an information collection module, a charging station selection index calculation module, an optimization judgment module, an information interaction module, a prediction module, a system model establishment module, and a system decision module;

[0030] The information collection module is used to obtain charging and discharging information related to electric vehicle users' EVUs;

[0031] The charging station selection index calculation module is used to calculate charging station selection indexes based on information obtained by the information collection module. The charging station selection indexes include: the shortest path distance D from the EVU's starting point to each charging station. min Power consumption E c and travel time T min Maximum waiting time T q Utilization rate U j ;

[0032] The optimization judgment module is used to judge the EV's optimized charging and discharging operation based on the EV's power status information and EVU time flexibility, and put the EV into the charging and discharging plan if the conditions are met.

[0033] The information interaction module is used to interact with the energy system and the power grid to obtain the required historical data information. The historical data information includes the power generation and unit power generation cost of each type of power plant per unit time period obtained from the energy system, and the original load value of each unit time period within a day obtained from the power grid.

[0034] The prediction module is used to predict the current grid load, real-time electricity price, and real-time carbon emission factor for the next 24 hours. Based on historical data obtained from the information interaction module, it predicts and obtains charging and discharging optimization indicators, including the predicted grid load L. t Real-time electricity price (REP) t and real-time carbon emission factor REF t ;

[0035] The system model building module is used to build an intelligent real-time scheduling model for electric vehicles.

[0036] The system decision module is used to solve the intelligent real-time scheduling model of electric vehicles based on the different target preferences of EVU and power grid, and obtain multiple recommended schemes with different preferences.

[0037] Furthermore, OD matrices are established based on the latitude and longitude of the EV and all charging stations within the region. The shortest path distance D from the EV's starting point to each charging station is then calculated using Dijkstra's algorithm. min The power consumption E under the shortest path distance is calculated according to formulas (1) and (2). c and travel time T min ;

[0038]

[0039]

[0040] Among them, E 100 E represents the electricity consumption of an electric vehicle per 100 kilometers under standard operating conditions. cap For the battery capacity of EV, The average speed of the electric vehicle;

[0041] According to queuing theory M / G / K, the maximum waiting time T for each charging station is calculated using formula (3). q :

[0042]

[0043] Where λ represents the average arrival rate of EVU following a Poisson process, μ represents the expected service time, σ represents the variance of the service time, k is the number of charging piles in the station, and ρ = λμ.

[0044] Utilization rate of each charging station in the area U j Expressed by formula (4):

[0045]

[0046] N m This represents the number of times charging station m provides service in a day, where n is the number of charging stations in a certain area.

[0047] Furthermore, the specific process of the optimization judgment module is as follows:

[0048] The initial state of charge (SOC) of the EV upon arrival at the charging station is calculated using formulas (5)-(7). start The time T for the EV to arrive at the charging station a EV charging and discharging time ΔT:

[0049] SOC start =SOC0-Ec (5)

[0050] T a =T0+T min (6)

[0051]

[0052] Where η is the charging and discharging efficiency of the charging pile; SOC0 is the battery state of charge; T0 is the time of the charging and discharging request; and SOC... end E represents the expected state of battery charge when the user leaves the charging station. cap For the battery capacity of EV, P ch / dis The charging and discharging power of the EV, SOC0, T0, SOC end E cap P ch / dis All data was obtained through the information collection module; power consumption E c and travel time T min Obtained through the charging station selection indicator calculation module;

[0053] If the charging / discharging time ΔT of the EV is less than the estimated time T for the EV to leave the charging station. l The time T for the EVU to arrive at the charging station a The difference, and the initial battery state of the EVU arriving at the charging station when it needs to discharge. Greater than 55%, that is

[0054] ΔT<T l -T a (8)

[0055]

[0056] The charging and discharging start time of the EV can be optimized by delaying charging and discharging and adding it to the charging and discharging plan; otherwise, if the EV is charged and discharged immediately, the EV will be removed from the charging and discharging plan.

[0057] Furthermore, the prediction module utilizes a feature extraction network LSTM to obtain future trend information from historical data. The input of the feature extraction network LSTM is L, REP, and REF of the past 24 hours, and the output is L, REP, and REF of the next 24 hours after the current moment.

[0058] Furthermore, the system decision module constructs the intelligent real-time scheduling model of electric vehicles as a Markov decision process (MDP), and selects the deep deterministic policy gradient (DDPG) algorithm to solve the MDP-based intelligent real-time scheduling model of electric vehicles.

[0059] For the Deep Deterministic Policy Gradient (DDPG) algorithm, the input is: input = (t, T)min E c ,T q U j ,L t ,REP t ,REF t Output: U j This represents the utilization rate of charging station j. and Let represent the start time of charging and the start time of discharging of the i-th EV, respectively. Let represent the charging capacity and discharging capacity of the i-th EV, respectively;

[0060] Using historical data as the environmental state, the network of the Deep Deterministic Policy Gradient (DDPG) algorithm is trained offline.

[0061] Then, the parameters of the trained DDPG algorithm are fixed, and the intelligent real-time scheduling model for electric vehicles is solved. When a scheduling task occurs in each cycle, the trained DDPG algorithm is used to select a scheduling action based on the current system state. The agent takes action to enter the next environmental state and receives a reward. Then, the original grid load, real-time grid electricity price, and real-time grid carbon emission factor for time period t+1 are collected as new samples to obtain the decision for that time period. The agent continuously learns based on the current reward to obtain the optimization scheme corresponding to maximizing the reward, including the optimal schemes for EVU, grid, and system.

[0062] The intelligent agent is any EV that is being scheduled.

[0063] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0064] (1) Existing electric vehicle scheduling methods only consider the charging and discharging scheduling after the electric vehicle arrives at the charging station, without considering the process of the EVU arriving at the charging station from the starting point. This application is the first to combine the entire process of the EVU selecting a charging station and carrying out orderly charging and discharging in a model, which optimizes both the EVU's selection of charging stations and the EVU's participation in the charging and discharging scheduling.

[0065] (2) In the existing electric vehicle scheduling methods, only a single indicator is used to guide EVU to select charging stations and schedule charging and discharging. This may affect the participation of other stakeholders. This application establishes an indicator system to reflect the entire process of EVU selecting charging stations and charging and discharging operations. It also proposes for the first time to evaluate the service quality of charging stations by the utilization rate of charging stations, so that EVU can select suitable charging stations and improve charging satisfaction.

[0066] (3) This invention proposes an intelligent real-time scheduling system for electric vehicles. The system is equipped with an information collection module, a charging station selection index calculation module, an optimization judgment module, an information interaction module, a prediction module, a system model establishment module, and a system decision module. Each module performs corresponding operations in sequence. The system proposes to optimize the charging and discharging operation based on the initial power status of the electric vehicle and the charging and discharging time, so as to achieve positive guidance for the EVU, save energy, and reduce the negative impact on the EVU and the power grid. Attached Figure Description

[0067] Figure 1 This is a flowchart of the EV intelligent real-time scheduling method based on vehicle-station-network information interaction according to the present invention.

[0068] Figure 2 This is a schematic diagram of the EV intelligent real-time dispatching system based on vehicle-station-network information interaction according to the present invention.

[0069] In the diagram, there are: information collection module 1, charging station selection index calculation module 2, optimization judgment module 3, information interaction module 4, prediction module 5, system model building module 6, and system decision-making module 7. Detailed Implementation

[0070] Specific embodiments of the present invention are given below. These specific embodiments are only used to further illustrate the present invention and do not limit the scope of protection of this application.

[0071] This invention considers carbon emissions and vehicle-station-network information interaction in a real-time intelligent EV scheduling method (see [link]). Figure 1 ):

[0072] Obtain the charging and discharging requests of electric vehicle users (EVUs), and obtain the EV's latitude and longitude, the time of the charging / discharging request (T0), the battery state of charge (SOC0), and the EV's battery capacity (E). cap ), EV charging and discharging power (P) ch / dis ), the expected state of charge (SOC) when the user leaves the charging station end ), EVU's estimated time to leave the charging station (T) l The average speed of electric vehicles information;

[0073] The criteria for selecting charging stations include: the shortest path distance D from the EVU's starting point to each charging station. min Power consumption E c and travel time T min Maximum waiting time T q Utilization rate U j ;

[0074] The original load value L of the power grid, the real-time electricity price REP, and the real-time carbon emission factor REF are predicted for the next 24 hours from the current time. The original load information, real-time electricity price information, and real-time carbon emission factor information of the power grid for the next 24 hours after the current time are obtained. The predicted original load, real-time electricity price, and real-time carbon emission factor are used as charging and discharging optimization indicators.

[0075] An indicator system is constructed using charging station selection criteria and charging / discharging optimization criteria. This system evaluates the convenience, service quality, economy, environmental friendliness, and safety of the entire process of an EV (Electronic Vehicle) from selecting a charging station to the end of charging / discharging. This indicator system serves as the network input for the system's decision-making module.

[0076] The objective function for constructing an intelligent real-time scheduling model for electric vehicles is given by formulas (18)-(25):

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] The constraints of the intelligent real-time scheduling model for electric vehicles are given by formulas (26) and (27):

[0084] |SOC t -SOC end |≤δ (26)

[0085] SOC min +E c ≤SOC0 (27)

[0086] SOC min This is the minimum threshold for the state of charge of an EV battery, typically set to 10%.

[0087] To determine whether the charging and discharging behavior of the EVU needs optimization, if the charging and discharging time ΔT of the EV is less than the expected time T for the EVU to leave the charging station. l The time T for the EVU to arrive at the charging station a The difference, and the initial battery state of the EVU arriving at the charging station when it needs to discharge. If the value is greater than 55%, the charging and discharging start time of the EV can be optimized. An optimization scheme is given according to the intelligent real-time scheduling model of electric vehicles.

[0088] Based on the different target preferences of EVU and power grid, the intelligent real-time scheduling model of electric vehicles is solved to obtain multiple recommended schemes with different preferences.

[0089] A real-time intelligent scheduling model for electric vehicles is trained and solved using deep reinforcement learning to obtain multiple recommendation schemes with different preferences (including charging station location, charging / discharging start time, and charging / discharging capacity). The real-time intelligent scheduling model for electric vehicles is equivalent to a Markov decision process (MDP), and the deep reinforcement learning is the Deep Deterministic Policy Gradient (DDPG) algorithm.

[0090] The structure of the EV Intelligent Real-Time Dispatch System (EVIRTSS) based on vehicle-station-network information interaction of this invention is as follows: Figure 2 As shown, it includes an information collection module 1, a charging station selection index calculation module 2, an optimization judgment module 3, an information interaction module 4, a prediction module 5, a system model building module 6, and a system decision-making module 7.

[0091] The information collection module is used to acquire charging and discharging related information of electric vehicle users (EVUs), specifically including: the latitude and longitude of the EV, the time of the charging / discharging request (T0), the battery state of charge (SOC0), and the EV's battery capacity (E). cap ), EV charging and discharging power (P) ch / dis The expected state of charge (SOC) of the battery when the user leaves the charging station. end ), the estimated time for the EVU to leave the charging station (T l The average speed of electric vehicles

[0092] The charging station selection index calculation module is used to calculate charging station selection indexes based on information obtained by the information collection module. The charging station selection indexes include: the shortest path distance D from the EVU's starting point to each charging station. min Power consumption E c and travel time T min Maximum waiting time T q Utilization rate U j ;

[0093] (1) Calculate the shortest path distance, power consumption, and travel time: Establish OD matrices based on the latitude and longitude of the EV and all charging stations in the area, and calculate the shortest path distance D from the starting point of the EV to each charging station using Dijkstra's algorithm. min The power consumption E under the shortest path distance is calculated according to formulas (1) and (2). c and travel time T min .

[0094]

[0095]

[0096] E 100 E represents the electricity consumption of an electric vehicle per 100 kilometers under standard operating conditions. cap For the battery capacity of EV, This represents the average speed of the electric vehicle.

[0097] (2) The maximum waiting time is one of the important indicators for EVU to select a charging station for charging and discharging. According to the queuing theory M / G / K, the maximum waiting time T of each charging station is calculated according to formula (3). q :

[0098]

[0099] Where λ represents the average arrival rate of EVU following a Poisson process, μ represents the expected service time, σ represents the variance of the service time, k is the number of charging piles in the station, and ρ = λμ.

[0100] (3) Utilization rate can characterize the charging service quality of a charging station. For example, a charging station with good service quality can demonstrate this through free parking, enthusiastic service personnel, and charging discount activities. The utilization rate of a charging station is also one of the important indicators for EVUs to select charging stations. Therefore, the utilization rate U of each charging station in the region... j Expressed by formula (4):

[0101]

[0102] N m Let m represent the number of times charging station m provides service in a day, and n represent the number of charging stations in a certain area. m The larger the value, the better the charging service of the charging station, the more popular it is, and the higher its utilization rate.

[0103] The optimization judgment module is used to judge the EV's optimized charging and discharging operation based on the EV's power status information and EVU time flexibility, and to include the EV in the charging and discharging plan if the conditions are met.

[0104] The initial state of charge (SOC) of the EV upon arrival at the charging station is calculated using formulas (5)-(7). start The time T for the EV to arrive at the charging station a EV charging and discharging time ΔT:

[0105] SOC start =SOC0-E c (5)

[0106] T a =T0+Tmin (6)

[0107]

[0108] Where η is the charging and discharging efficiency of the charging pile, which is a known value; SOC0 is the battery state of charge; T0 is the time of the charging and discharging request; and SOC... end E represents the expected state of battery charge when the user leaves the charging station. cap For the battery capacity of EV, P ch / dis The charging and discharging power of the EV is obtained through the information collection module; the power consumption E c and travel time T min The results are obtained through the charging station selection indicator calculation module.

[0109] If the charging / discharging time ΔT of the EV is less than the estimated time T for the EV to leave the charging station. l The time T for the EVU to arrive at the charging station a The difference, and the initial battery state of the EVU arriving at the charging station when it needs to discharge. Greater than 55%, that is

[0110] ΔT<T l -T a (8)

[0111]

[0112] The charging and discharging start time of EVs can be optimized by delaying charging and discharging and adding them to the charging and discharging plan. For EVs in the charging and discharging plan, the intelligent real-time scheduling model for electric vehicles needs to be used for optimization to provide recommended solutions with different preferences. Conversely, if an EV does not meet the criteria, it should be charged and discharged immediately, and the vehicle should be removed from the charging and discharging plan. The charging and discharging plan refers to a set of EVs that can be optimized. For example, if EV1 meets the criteria, the scheduling system records EV1 and uses time control to benefit both the EV and the power grid. Otherwise, no charging and discharging control is applied.

[0113] The information interaction module is used to exchange information with the energy system and the power grid to obtain the necessary historical data. The historical data includes the power generation and unit power generation cost of each type of power plant every 15 minutes obtained from the energy system, and the original load values ​​for each time period (15 minutes) within a day obtained from the power grid. Based on the historical data, charging and discharging optimization indicators are obtained, which include the original load information of the power grid, real-time electricity price information, and real-time carbon emission factor information.

[0114] Grid load information is a crucial indicator of grid stability. The grid aims for EVUs to charge during off-peak hours and discharge during peak hours to achieve peak shaving and valley filling, thereby improving grid stability. Real-time electricity price information is a key indicator of EVU charging economics. EVUs aim to charge when electricity prices are low to reduce charging costs and discharge when prices are high to maximize discharge revenue, thus improving charging economics. Real-time carbon emission factor information is a crucial indicator of grid environmental friendliness. The grid aims for EVUs to charge during off-peak hours based on real-time carbon emission factors to reduce grid carbon emissions and improve grid environmental friendliness.

[0115] (1) Power grid original load curve

[0116] EVIRTSS's information interaction module obtains the original load values ​​for each time period (15 minutes) of the day from the power grid and plots them as the original load curve of the power grid. The load values ​​for each time period of the day are different, and there are generally two peak times in the morning and evening. If a large number of electric vehicles charge during peak times, it will have a great impact on the security of the power grid.

[0117] (2) Real-time electricity price calculation

[0118] First, real-time data acquisition and processing of the energy system are performed to obtain the power generation and unit power generation cost of each type of power plant every 15 minutes. The average power generation cost C of the g types of power plants in time period t is calculated by formula (10). t .

[0119] Secondly, the transmission and distribution costs and line loss rates are obtained from the power grid, and the real-time electricity price REP for time period t is calculated using equation (11). t .

[0120]

[0121]

[0122] C t This represents the average power generation cost of all g types of power plants during time period t. This represents the unit power generation cost of the f-th type of power plant. Represents the power generation of the f-th type of power plant, REP t C represents the real-time electricity price during time period t. T ΔL represents the transmission and distribution cost, and ΔL represents the line loss rate.

[0123] (3) Real-time carbon emission factor calculation

[0124] The calculation of REF should only consider power plants that act as marginal generators. Power plants that do not respond to changes in demand are not considered marginal generators. Therefore, most literature only considers conventional power plants. Thus, the power generation of marginal power plants such as thermal power plants, hydropower plants, and nuclear power plants is collected every 15 minutes from the energy system in real time. The total power generation of g-type marginal power plants is calculated using Equation (12), the total emissions of g-type marginal power plants are calculated using Equation (13), and the carbon emissions of the k-th type marginal power plant during time period t are calculated using Equation (14).

[0125]

[0126]

[0127]

[0128] Among them G t E represents the total power generation of a class g power plant during power generation period t. t This represents the total emissions generated by a class g power plant during power generation period t. β f γ represents the average carbon emission factor of the f-th type of power plant. f This represents the power generation efficiency of the f-th type of power plant. Let f be the carbon emissions of the class f power plant during period t.

[0129] When EVs arrive at charging stations, their charging demand changes. Marginal power plants increase their power generation to meet this additional demand. Therefore, REF is calculated using equation (15). t .

[0130]

[0131] ΔE t =E t -E t-1 (16)

[0132] ΔG t =G t -G t-1 (17)

[0133] REF t ΔE represents the real-time carbon emission factor during time period t. t and ΔG t These represent the changes in carbon emissions and electricity generation between two adjacent time periods, respectively.

[0134] The prediction module is used to predict the current grid load, real-time electricity price (REP), and real-time carbon emission factor (REF) for the next 24 hours.

[0135] To meet the economic requirements of EV charging and discharging, as well as the safety and environmental protection of the power grid, it is necessary to know the L, REP, and REF values ​​for the next 24 hours from the current moment. Therefore, it is necessary to predict the future L, REP, and REF values ​​using historical L, REP, and REF data. Since L, REP, and REF are quasi-periodic and natural time series, the Feature Extraction Network (LSTM) can obtain their future trend characteristics from historical data. Specifically, the LSTM takes the L, REP, and REF values ​​of the past 24 hours as input and outputs the L, REP, and REF values ​​for the next 24 hours after the current moment. This predicted output information is then input into the system decision-making module to coordinate the charging and discharging operations of the electric vehicle.

[0136] The system model building module is used to build an intelligent real-time scheduling model for electric vehicles.

[0137] This application aims to achieve different goals based on the different preferences of EVUs and the power grid. On the user side, the focus is on charging convenience, charging service quality, and charging economy; on the power grid side, the focus is on safety and environmental protection.

[0138] Each indicator corresponds to a different evaluation objective. For the user side, the convenience of going to the charging station is reflected by minimizing the travel time and power consumption of the EVU from the charging demand point to the charging station. The objective function is expressed by equations (18) and (19):

[0139] f1 = T min (18)

[0140] f2 = E c (19)

[0141] After arriving at the charging station, EVUs hope to charge when the electricity price is low to reduce charging costs and discharge when the electricity price is high to increase discharge revenue, thereby improving the charging and discharging economy. Therefore, the real-time electricity price is an important indicator for evaluating the charging and discharging economy. The charging and discharging objective functions are represented by equations (20) and (21), respectively.

[0142]

[0143]

[0144] Where T is the set of time periods, a day is divided into 96 time periods, and the length of each time period Δt is 15 minutes, REP t This represents the real-time electricity price within time period t, obtained from the prediction module. It is assumed that the discharge price is also represented as REP. t ; This indicates that the i-th EV is charged during time period t, otherwise it is not charged. This indicates that the i-th EV discharges during time period t, otherwise it does not discharge; η is the charging and discharging efficiency of the charging pile. Let be the charging and discharging power of the i-th EV; assume that it is not possible to charge and discharge at the same charging station.

[0145] For the charging station, the maximum waiting time is an important indicator of the congestion level. EVUs arriving at the charging station should ideally have the shortest possible queuing time for charging and discharging. The objective function is expressed by equation (22):

[0146] f5 = T q (twenty two)

[0147] The utilization rate can reflect the popularity of the charging station. Before charging and discharging, the EVU will consider the service quality of the charging station. The higher the utilization rate, the better the service quality. Therefore, the utilization rate of the charging station is one of the important indicators for the EVU to select the charging station. The objective function is represented by equation (23).

[0148] f6 = U j (twenty three)

[0149] For the power grid side, the simultaneous connection of a large number of electric vehicles to the grid during peak load periods will result in peak-to-peak phenomena, leading to extreme grid instability and a high risk of safety accidents. Therefore, the power grid hopes that EVUs will charge during off-peak hours and discharge during peak hours to achieve peak shaving and valley filling, thereby improving grid security. Thus, load is an important indicator for evaluating grid security. The peak-valley difference of the load reflects grid stability, which is represented by the objective function (24):

[0150]

[0151] Where P max P is the maximum value of the original load of the power grid plus the charging load. min It is the minimum value of the original load of the power grid plus the charging load.

[0152] With the introduction of the "dual carbon" target, various industries are making their own contributions to emission reduction. The power grid hopes that EVUs will be charged during off-peak REF hours to reduce carbon emissions and improve the environmental friendliness of the power grid. Therefore, REF is an important indicator for evaluating the environmental friendliness of the power grid. The environmental friendliness of the power grid is reflected by the amount of carbon emissions generated, which is represented by the objective function (25):

[0153]

[0154] REF t This represents the real-time carbon emission factor (REF) within time period t, obtained from predictions by the prediction module. It represents the REF for each time period within a 24-hour day. t The values ​​may all be different.

[0155] The above objective function and equations comply with the following constraints:

[0156] To ensure that the charging and discharging needs of electric vehicles are met, the State of Charge (SOC) of the battery when the electric vehicle user picks up the vehicle is... t Compared to the expected battery state of charge (SOC) end similar.

[0157] |SOC t -SOC end |≤δ (26)

[0158] Where δ represents the allowable difference between the battery state when the EV user picks up the vehicle and the expected battery state.

[0159] To ensure that the EV battery has sufficient charge to reach the charging station, the EV's battery charge constraint is expressed as:

[0160] SOC min +E c ≤SOC0 (27)

[0161] SOC min This is the minimum threshold for the state of charge of an EV battery, typically set to 10%.

[0162] The system decision module is used to solve the intelligent real-time scheduling model of electric vehicles based on the different target preferences of EVU and power grid, and obtain multiple recommended schemes with different preferences to adapt to the uncertainty of EVU behavior and environmental changes;

[0163] Because EVs' selection of charging stations and their charging / discharging behavior are uncertain, the constraints in the intelligent real-time scheduling model for electric vehicles may be affected by the random behavior of EV users. Therefore, the EV's selection of charging stations and its charging / discharging behavior exhibits Markov properties, allowing the intelligent real-time scheduling model for electric vehicles to be constructed as a Markov Decision Process (MDP). This application employs the Deep Deterministic Policy Gradient (DDPG) algorithm. The optimal policy function is estimated using a deep neural network, and the MDP-based intelligent real-time scheduling model for electric vehicles is solved to adapt to the uncertainty of EV behavior and environmental changes.

[0164] For the DDPG algorithm, the network input is: input = (t, T) min E c ,T q U j ,L t ,REP t ,REF t Output: U jThis represents the utilization rate of charging station j, and its function is to select charging station j for charging and discharging operations. and Let represent the start time of charging and the start time of discharging of the i-th EV, respectively. Let represent the charging and discharging capacities of the i-th EV, respectively. Using historical data as the environmental state, the DDPG algorithm network is trained offline. Then, the parameters of the trained DDPG algorithm are fixed, and the intelligent real-time scheduling model for electric vehicles is solved. For the intelligent real-time scheduling system for electric vehicles, when a scheduling task occurs in each cycle, the trained DDPG algorithm selects a scheduling action based on the current system state. The agent (referring to any EV being scheduled) takes action to enter the next environmental state and receives a reward. Then, the system state information for time period t+1 (here, system state information refers to the original grid load, real-time electricity price, and real-time carbon emission factor) is collected as new samples to derive the decision for that time period. The agent continuously learns based on the current reward to obtain the optimal solution that maximizes the reward, including optimal solutions for EVU, grid, and system.

[0165] Considering T from the perspective of EVU optimization min or E c Smaller path, U j Higher charging stations or T q Smaller charging stations and charging times with lower REP values ​​or discharging times with higher REP values ​​are optimized based on the different preference criteria set by the EVU, providing multiple recommended schemes for users to choose from. For example, if only the EVU's economy is considered, it can be guided to charge when the REP value is low to reduce charging costs and discharge when the REP value is high to increase discharge benefits. If only the EVU's charging and discharging convenience is considered, the shorter route can be selected to minimize T min Arrive at the charging station; if you only consider the service quality of the charging station, simply select U. j Higher or T q Smaller charging stations; if the EVU simultaneously considers charging and discharging economy, charging and discharging convenience, and the service quality of the charging station, then T needs to be included. min E c ,T q U j and REP t Equal weights are assigned to obtain the recommended solution. L is considered from the perspective of grid optimization. t and REF t Charge at lower times, in L t Discharge occurs at higher times. Different charge / discharge optimization metrics have varying impacts on the power grid. If only grid security is considered, it is only necessary to guide the EVU to select L...t Charging operations are performed in shorter time periods, L t Discharging operations are performed over longer periods to achieve peak shaving and valley filling on the grid side. If only the environmental friendliness of the grid is considered, it is only necessary to guide the EVU to select off-peak charging times during REF periods to reduce carbon emissions. If both grid security and environmental friendliness need to be considered simultaneously, then L needs to be... t and REF t To obtain a recommended solution, all indicators should be assigned equal weights. Considering the interests of both the EVU and the power grid from a system-optimal perspective, this requires assigning equal weights to all the above indicators. While this may prevent any single objective from reaching its optimal level, it provides a compromise solution that achieves the system-optimal recommended solution.

[0166] This invention presents an intelligent real-time scheduling method for electric vehicles (EVs) based on vehicle-station-network information interaction: 1. By establishing an index system to evaluate charging station selection and the EV charging / discharging process, guidance is provided for intelligent real-time EV scheduling; 2. Based on different evaluation indicators, different objectives for EVs and the power grid are achieved, and an intelligent real-time scheduling model for electric vehicles is established; 3. The intelligent real-time scheduling model for electric vehicles is constructed as a Markov process, and deep reinforcement learning is used to train and solve the Markov decision process model to obtain multiple recommended solutions with different preferences. Through the above methods, the agent will obtain the location of the charging station, the dynamic start time of charging / discharging, and the charging / discharging capacity, thereby solving the problems of where the EV is charged (referring to the charging station number), when it is charged / discharged, and the charging / discharging capacity.

[0167] Any aspects not covered in this invention are applicable to existing technologies.

Claims

1. A real-time intelligent EV scheduling method based on vehicle-station-network information interaction, characterized in that, The scheduling method includes the following: Obtain the charging / discharging requests of electric vehicle users (EVUs), and obtain the EV's latitude and longitude, the time T0 of the charging / discharging request, the battery state of charge (SOC0), and the EV's battery capacity (E). cap EV charging and discharging power P ch / dis Expected State of Charge (SOC) when the user leaves the charging station end EVU's estimated time to leave the charging station (T) l Average speed of electric vehicles information; The criteria for selecting charging stations include: the shortest path distance D from the EVU's starting point to each charging station. min Power consumption E c and travel time T min Maximum waiting time T q Utilization rate U j ; The original load value L of the power grid, the real-time electricity price REP, and the real-time carbon emission factor REF are predicted for the next 24 hours from the current time. The original load information, real-time electricity price information, and real-time carbon emission factor information of the power grid for the next 24 hours after the current time are obtained. The predicted original load, real-time electricity price, and real-time carbon emission factor are used as charging and discharging optimization indicators. The indicator system consists of charging station selection indicators and charging / discharging optimization indicators; Assuming that charging and discharging cannot occur simultaneously at the same charging station, the objective function f1-f8 for constructing the intelligent real-time scheduling model for electric vehicles is given by formulas (18)-(25): f1=T min (18) f2=E c (19) f5=T q (22) f6=U j (23) Where T is the set of time periods, and REP t This represents the real-time electricity price within time period t, which is obtained through forecasting. This indicates that the i-th EV is charged during time period t, otherwise it is not charged. This indicates that the i-th EV discharges within time period t, otherwise it does not discharge; Δt is the length of a unit time period; η is the charging and discharging efficiency of the charging pile. The charging and discharging power of the i-th EV; P max P is the maximum value of the original load of the power grid plus the charging load. min REF represents the minimum value of the original grid load plus the charging load; t This represents the real-time carbon emission factor within time period t, which is obtained through prediction. The constraints of the intelligent real-time scheduling model for electric vehicles are given by formulas (26) and (27): |SOC t -SOC end |≤δ (26) SOC min +E c ≤SOC0 (27) Among them, SOC min The minimum threshold for the state of charge (SOC) of an EV battery. t The state of battery charge is the state of charge when the EV user picks up the vehicle; δ represents the allowable difference between the state of battery charge when the EV user leaves the vehicle and the desired state of battery charge. If the charging / discharging time ΔT of the EV is less than the estimated time Tl for the EVU to leave the charging station and the estimated time T for the EVU to arrive at the charging station... a The difference, and the initial battery state of the EVU arriving at the charging station when it needs to discharge. If it is greater than 55%, the charging and discharging start time of the EV can be optimized, and the optimization can be carried out according to the intelligent real-time scheduling model of electric vehicles. Based on the different target preferences of EVU and power grid, indicators in the indicator system are selected to solve the intelligent real-time scheduling model of electric vehicles and obtain multiple recommended schemes with different preferences.

2. The EV intelligent real-time scheduling method based on vehicle-station-network information interaction according to claim 1, characterized in that, The intelligent real-time scheduling model for electric vehicles is equivalent to a Markov decision process (MDP). Deep reinforcement learning is used to train and solve the intelligent real-time scheduling model for electric vehicles to obtain multiple recommendation schemes with different preferences. The recommendation schemes include charging station locations, charging / discharging start times, and charging / discharging amounts. The deep reinforcement learning is the Deep Deterministic Policy Gradient (DDPG) algorithm.

3. An intelligent real-time dispatching system for EVs based on vehicle-station-network information interaction, characterized in that, The system executes the method of claim 1 or 2, including an information collection module, a charging station selection index calculation module, an optimization judgment module, an information interaction module, a prediction module, a system model building module, and a system decision-making module; The information collection module is used to acquire charging and discharging related information of the EVU; The charging station selection index calculation module is used to calculate charging station selection indexes based on information obtained by the information collection module. The charging station selection indexes include: the shortest path distance D from the EVU's starting point to each charging station. min Power consumption E c and travel time T min Maximum waiting time T q Utilization rate U j ; The optimization judgment module is used to judge the EV's optimized charging and discharging operation based on the EV's power status information and EVU time flexibility, and put the EV into the charging and discharging plan if the conditions are met. The information interaction module is used to interact with the energy system and the power grid to obtain the required historical data information. The historical data information includes the power generation and unit power generation cost of each type of power plant per unit time period obtained from the energy system, and the original load value of each unit time period within a day obtained from the power grid. The prediction module is used to predict the current grid load, real-time electricity price, and real-time carbon emission factor for the next 24 hours. Based on historical data obtained from the information interaction module, it predicts and obtains charging and discharging optimization indicators, including the predicted grid load L. t Real-time electricity price (REP) t and real-time carbon emission factor REF t ; The system model building module is used to build an intelligent real-time scheduling model for electric vehicles. The system decision module is used to solve the intelligent real-time scheduling model of electric vehicles based on the different target preferences of EVU and power grid, and obtain multiple recommended schemes with different preferences.

4. The EV intelligent real-time dispatching system based on vehicle-station-network information interaction according to claim 3, characterized in that: Based on the latitude and longitude of the EV and all charging stations in the area, OD matrices are established. The shortest path distance D from the starting point of the EV to each charging station is calculated using Dijkstra's algorithm. min The power consumption E under the shortest path distance is calculated according to formulas (1) and (2). c and travel time T min ; Among them, E 100 E represents the electricity consumption of an electric vehicle per 100 kilometers under standard operating conditions. cap For the battery capacity of EV, The average speed of the electric vehicle; According to queuing theory M / G / K, the maximum waiting time T for each charging station is calculated using formula (3). q : Where λ represents the average arrival rate of EVU following a Poisson process, μ represents the expected service time, σ represents the variance of the service time, k is the number of charging piles in the station, and ρ = λμ. Utilization rate of each charging station in the region U j Expressed using formula (4): N m This represents the number of times charging station m provides service in a day, where n is the number of charging stations in a certain area.

5. The EV intelligent real-time dispatching system based on vehicle-station-network information interaction according to claim 3, characterized in that, The specific process of the optimization judgment module is as follows: The initial state of charge (SOC) of the EV upon arrival at the charging station is calculated using formulas (5)-(7). start The time T for the EV to arrive at the charging station a EV charging and discharging time ΔT: DIRTY start =SOC0-E c (5) T a =T0+T min (6) Where η is the charging and discharging efficiency of the charging pile; SOC0 is the battery state of charge; T0 is the time of the charging and discharging request; and SOC... end E represents the expected state of battery charge when the user leaves the charging station. cap For the battery capacity of EV, P ch / dis The charging and discharging power of the EV, SOC0, T0, SOC end E cap P ch / dis All data was obtained through the information collection module; power consumption E c and travel time T min Obtained through the charging station selection indicator calculation module; If the charging / discharging time ΔT of the EV is less than the estimated time T for the EV to leave the charging station. l The time T for the EVU to arrive at the charging station a The difference, and the initial battery state of the EVU arriving at the charging station when it needs to discharge. Greater than 55%, that is ΔT<T l -T a (8) The charging and discharging start time of the EV can be optimized by delaying charging and discharging and adding it to the charging and discharging plan; otherwise, if the EV is charged and discharged immediately, the EV will be removed from the charging and discharging plan.

6. The EV intelligent real-time dispatching system based on vehicle-station-network information interaction according to claim 3, characterized in that, The prediction module utilizes a feature extraction network (LSTM) to obtain future trend information from historical data. The input to the LSTM network is the L data from the past 24 hours. t ,REP t ,REF t The output is L in the next 24 hours after the current time. t ,REP t ,REF t .

7. The EV intelligent real-time dispatching system based on vehicle-station-network information interaction according to claim 3, characterized in that, The system decision module constructs the intelligent real-time scheduling model of electric vehicles into a Markov decision process (MDP), and selects the deep deterministic policy gradient (DDPG) algorithm to solve the intelligent real-time scheduling model of electric vehicles based on the MDP. For the Deep Deterministic Policy Gradient (DDPG) algorithm, the input is: input = (t, T) min E c ,T q U j ,L t ,REP t ,REF t Output: U j This represents the utilization rate of charging station j. and Let represent the start time of charging and the start time of discharging of the i-th EV, respectively. Let represent the charging capacity and discharging capacity of the i-th EV, respectively; Using historical data as the environmental state, the network of the Deep Deterministic Policy Gradient (DDPG) algorithm is trained offline. Then, the parameters of the trained DDPG algorithm are fixed, and the intelligent real-time scheduling model for electric vehicles is solved. When a scheduling task occurs in each cycle, the trained DDPG algorithm is used to select a scheduling action based on the current system state. The agent takes action to enter the next environmental state and receives a reward. Then, the original grid load, real-time grid electricity price, and real-time grid carbon emission factor for time period t+1 are collected as new samples to obtain the decision for that time period. The agent continuously learns based on the current reward to obtain the optimization scheme corresponding to maximizing the reward, including the optimal schemes for EVU, grid, and system. The intelligent agent is any EV that is being scheduled.

Images