A multi-time scale electric vehicle load forecasting method based on gravity model

Abstract

The present invention is a multi-time scale electric vehicle load forecasting method based on a gravity model. First, based on the EV ownership obtained by deduction, and then by combining the improved gravity model with time characteristics, a user multi-time scale transfer probability data set is obtained to describe the transfer of electric vehicle users in different regions. Secondly, the randomness of multiple power levels is considered, and the selected power level type is judged by analyzing variables such as charge state, residence time and power consumption. The randomness of the charging destination is characterized by analyzing the user's charging station selection time and queuing cost, and a random charging behavior model is constructed; finally, based on vehicle travel distribution, charging decision-making and ownership development, Monte Carlo is used to simulate vehicle behavior and predict the multi-time scale charging load of each administrative region. This method predicts the multi-time scale charging load of EVs in each region by considering the development of EV ownership, the dynamic spatiotemporal distribution of travel behavior in each region, and the random charging behavior of differentiated individual attributes and attitude attributes.

Description

Technical Field

[0001] The present invention relates to the technical field of electric vehicle load prediction in a power system, and in particular to a method for multi-time scale electric vehicle load prediction based on an improved gravity model. Background Art

[0002] With the rapid development of the electric vehicle (EV) industry, the number of electric vehicles in 2023 has reached 15.52 million, and the impact of their charging load on the power grid cannot be ignored. EV charging load prediction is the basis for the long-term development of charging facility planning and distribution network upgrade and transformation. However, most current studies are short-term charging load predictions, and rarely analyze the differences in charging loads in different day types and seasons in future years. It is impossible to show the EV charging load characteristics after a long time, and it is impossible to predict the spatiotemporal distribution of charging loads at different time scales. It is obviously far from enough to analyze the charging loads at different time periods within the day. Therefore, in order to advance the planning and construction of EV charging facilities and power grids in the future, it is urgent to combine short-term and medium- and long-term characteristics to realize multi-time scale EV charging load predictions such as different day types and different seasons in future years.

[0003] 5. First, based on the EV ownership obtained by deduction, and then by combining the improved gravity model with time characteristics, the user multi-time scale transfer probability data set is obtained to describe the transfer of electric vehicle users in different regions. Secondly, considering the randomness of multiple power levels, the selected power level type is determined by analyzing variables such as charge state, residence time and power consumption, and the randomness of the charging destination is characterized by analyzing the user's charging station selection time and queuing cost, and a random charging behavior model is constructed; finally, based on vehicle travel distribution, charging decisions and ownership development, Monte Carlo is used to simulate vehicle behavior and predict the multi-time scale charging load of each administrative region. This method predicts the multi-time scale charging load of EVs in each region by considering the development of EV ownership, the dynamic spatiotemporal distribution of travel behavior in each region, and the random charging behavior considering multiple power levels and station selection costs.

[0004] In summary, this paper considers the interaction between EV ownership and vehicle travel and charging behavior, and proposes a multi-time scale EV charging load forecasting method. Summary of the invention

[0005] In order to realize the multi-time scale EV charging load forecasting of different day types and different seasons in future years, the present invention provides a multi-time scale electric vehicle load forecasting method based on an improved gravity model.

[0006] The charging load of electric vehicles is predicted at multiple time scales by combining an improved gravity model with a stochastic charging behavior model.

[0007] The present invention provides a multi-time scale electric vehicle load forecasting method based on an improved gravity model, which is divided into the following steps:

[0008] Step 1: Obtain the initial number of electric vehicles and basic information on user travel.

[0009] The travel volume of each region in historical years is counted to determine the travel volume ratio of each administrative area; the travel volume ratio of each region is used to replace the EV ownership ratio of each region, and the EV ownership of each region is determined based on the EV ownership deduction method considering the individual decision-making behavior of consumers, as shown in formula (1).

[0010] O i (t) = Z(t)·r i (1)

[0011] Among them, Oi(t) is the number of EVs in region i in forecast year t. The number of people who purchase EVs in each period is determined according to the consumer state decision model based on Markov chain, and then the EV ownership is obtained. Z(t) is the total number of EVs in forecast year t; ri is the travel volume ratio of region i.

[0012] In the consumer state decision model based on Markov chain, individual consumers are divided into four states, namely non-cognitive state S1 (consumers lack knowledge of electric vehicles), cognitive state S2 (consumers have a certain understanding of electric vehicles), willingness to buy state S3 (consumers have a willingness to buy electric vehicles) and purchase state S4 (consumers buy electric vehicles). Consumers in different states are affected by consumers, EVs and society, and their states are transferred. The specific transfer process is as follows:

[0013] 1) When consumer i is in state S1:

[0014] If the publicity impact generated by the advertisement exceeds the cognitive threshold of consumer i, the state of consumer i changes from S1 to S2; otherwise, S1 remains unchanged.

[0015] 2) When consumer i is in state S2:

[0016] If consumer i's cognitive ability cannot keep up with the technological development of EV (represented by the slope of the comprehensive technical performance curve), cognitive deficiencies will occur and the consumer will retreat from S2 to S1 with a certain probability. Otherwise, the perceived total utility considering consumer preferences and social network utility will be judged to determine whether the consumer is willing to purchase EV. When the perceived total utility is greater than the consumer's purchase intention threshold, the consumer's state will change from S2 to S3. When the perceived total utility is less than the threshold, the consumer remains in state S2.

[0017] 3) When the consumer i state is S3:

[0018] If the current comprehensive technical performance does not meet consumers' perceived quality expectations, consumers may not have a strong willingness to buy and will return from S3 to S2.

[0019] 4) When consumer i is in state S4

[0020] If the age of the consumer's car is less than its service life, there is no need to change the car, and the consumer remains in S4; otherwise, a demand for changing the car arises, that is, the consumer returns from S4 to S3.

[0021] Finally, the number of consumers who decide to buy EVs in each period is counted and accumulated to obtain the final number of buyers Z(t), as shown in formula (2).

[0022]

[0023] Step 2: By combining the improved gravity model with temporal and spatial characteristics, a user multi-time scale transfer probability dataset is obtained to characterize user travel behavior.

[0024] EV charging load is affected by travel behavior. Considering different time seasons and traffic flow conditions, the gravity model is used to characterize the daily travel behavior of vehicles. Combined with the large amount of spatial and temporal information included in the travel chain model, the travel chain is roughly divided into 8 types according to the activity type, and the gravity model is used to conduct a specific analysis of various types of information in the travel chain.

[0025] Define the regional transition probability matrix A for a period of time t t =[a ij t ] N×N Describes the travel transfer probability between regions in 24 time periods of a day, where N is the total number of regions; a ij t is the travel transfer probability from region i to region j in period t, where period t includes a multi-year time scale, four seasons, two typical day types (weekdays and weekends), and 24 hours a day. The final generated a ij t is a multi-time scale transition probability dataset. ij t The size is related to regional characteristics, such as being proportional to the travel force and attraction between regions and inversely proportional to traffic impedance. It is characterized by the gravity model, as shown in formula (3).

[0026]

[0027] Among them, θ t is the time adjustment coefficient, which is used to describe the time differences between different seasons and different day types; θ fis the traffic flow adjustment coefficient, which is used to describe the difference in vehicle flow in the road network in different seasons and different types of days. The size of the parameter is firstly obtained by cleaning the EV historical data, and the information obtained is used to construct a decision matrix to define the attribute weight. The entropy technology method is used to use the discrete degree of the attribute value to achieve the initial solution of the size of the weight assignment parameter, and then it is optimized in the iterative process, so as to realize the dual constraints of the gravity model, and then more accurately obtain the transfer probability matrix data set. i t is the travel force of area i during period t; Q i t is the attraction of region j in period t. The travel force and attraction of a region are related to the characteristics of the region and can be obtained from traffic data; where f1(.) is the probability density function, μ is the mean, and σ is the standard deviation. Function F ij is the traffic impedance between area i and area j, which is determined by the distance l between the two places. ij Since the regional location data are mostly expressed in longitude and latitude, the distance between two longitude and latitude points can be calculated by the haversine equation, satisfying formula (5)

[0028]

[0029] Among them, lon1, lat1 and lon2, lat2 represent the longitude and latitude of location i and j respectively, and R is the radius of the earth (the radius of the earth is 6378km, R = 6378*1000m)

[0030] In addition, for the single-region single-time transition probability matrix A t =[a ij t ] N×N The sum of the transition probabilities from region i to other regions is 1, that is, the sum of the elements in the i-th row of the matrix should be 1, as shown in formula (6).

[0031]

[0032] Step 3: Construct a random charging behavior model considering multiple charging power levels and the randomness of destinations.

[0033] Considering the heterogeneous charging needs of EV users, at the level of charging infrastructure, following the logical framework of increasing charging efficiency, we select slow charging facilities (SCF), fast charging facilities (FCF) and ultra-fast charging facilities (UCF) as three representative types of electric vehicle charging piles for users to randomly choose. k , its charging power can be quantitatively described by formula (7).

[0034]

[0035] Among them, P EVk Indicates EV k The charging power, P SCF , P FCF , P UCF Respectively represent the rated charging power of slow charging pile, fast charging pile and super fast charging pile, T k dur For EV k The expected length of stay, Cap k For EV k Battery capacity, SOC k For EV k state of charge.

[0036] For each stop at the activity location in the travel chain, assuming that the charging facilities at the stop are sufficient, the owner needs to decide whether to charge (i.e., charging decision) at the stop and calculate the charging load. If the EV decides to stop and charge at a certain place, then it is assumed that the charging facilities are connected at the moment of arrival at the stop, and charging is continued until the ideal power is reached or the maximum stay time is reached, then the charging time T ch As shown in formula (8).

[0037]

[0038] Among them, SOC exp is the ideal battery state after charging, assuming it is 95% of the total battery capacity, τ s is the charging efficiency, T stay is the residence time. The charging efficiency is affected by temperature, and different seasons have different charging efficiencies.

[0039] According to the charging habits of the car owner, the charging decision is divided into two situations: deterministic charging and random charging. Deterministic charging: Generally speaking, when the EV's range is not enough to complete the next trip, it will choose to charge at the stop, as shown in formula (9):

[0040] SOC k -ε s d k+1 <SOC alert (9)

[0041] Among them, SOC k is the remaining power after the kth trip of the travel chain; ε s is the power consumption per unit mileage; d k+1 is the k+1th trip mileage; SOC alert The remaining power after the trip can be calculated by sampling each mileage using Monte Carlo simulation, and then calculated based on the current power and power consumption per unit distance. The power consumption per unit distance is affected by weather temperature and has different power consumption in different seasons.

[0042] However, when the power can support the next trip, the random charging behavior of the destination must be considered. While considering observable factors such as individual attributes and charging attributes, the present invention takes into account the impact of destination randomness on charging behavior. In view of this, the Destination-Stochastic charging behavior model (D-SCB) considering the randomness of the destination is used, and the station selection cost of the destination is integrated. When a certain activity in the travel chain stops, a judgment is made to estimate the charging psychological probability P. CM (Charging mental probability), when the charging mental threshold P CM When the value is greater than a certain threshold (assuming the threshold is 0.5), the user decides to charge at the place where he / she stays, otherwise he / she decides not to charge.

[0043] Charging psychological probability P CM The calculation of is very important, and the internal mechanism and external environment of the charging decision process are deeply analyzed. The charging behavior satisfies the principle of utility maximization, that is, the option with the greatest utility (i.e., charging / not charging) is selected after comprehensive consideration of various factors. The station selection time and queuing time of EV users are described as the charging probability p cs and p qt, where the station selection time represents the time it takes for the EV user to travel from the end of the activity to the nearest charging station, and v satisfies the negative exponential function distribution; the queuing time represents the time that the EV user may wait in line for the required charging power charging pile after arriving at the charging station. The user's expectation of the available charging time is described as a probability, and the probability of the electric vehicle accepting charging is shown in the formula.

[0044]

[0045] Where u is the average service rate; 1 / u is the average service time. When the number of electric vehicles is less than the number of charging piles in the charging station, the total service rate of the system is k; when the number of electric vehicles exceeds the number of charging piles in the system, the total service rate of the system is C; η is the parameter obtained after considering observable factors such as individual attributes and charging attributes, which can be obtained using the hybrid choice model HCM.

[0046] Step 4: Generate multi-time scale load forecast results through solving.

[0047] According to the above steps, the model is simulated and solved. If all regions have been simulated, the charging load curves of different seasons and typical days in each region in the tth year are counted to generate multi-time scale load forecast results.

[0048] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: combining the gravity model characterizing the travel chain and the random charging behavior model to realize the EV charging load prediction on multiple time scales such as different day types and different seasons in future years. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a flow chart provided by the present invention.

[0050] Figure 2 It is the travel chain information provided by the present invention.

[0051] Figure 3 This is the type of travel chain provided by the present invention.

[0052] Figure 4 This is the charging decision process provided by the present invention.

[0053] Figure 5 This is the EV charging load prediction result in different seasons provided by the present invention. DETAILED DESCRIPTION

[0054] The specific embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings;

[0055] The present invention provides a multi-time scale electric vehicle load forecasting method based on an improved gravity model. The framework diagram of the present invention is as follows: Figure 1 As shown, it is divided into the following steps:

[0056] Step 1: Obtain the initial number of electric vehicles and basic information on user travel.

[0057] The travel volume of each region in historical years is counted to determine the travel volume ratio of each administrative area; the travel volume ratio of each region is used to replace the EV ownership ratio of each region, and the EV ownership of each region is determined based on the EV ownership deduction method considering the individual decision-making behavior of consumers, as shown in formula (1).

[0058] O i (t) = Z(t)·r i (11)

[0059] Among them, Oi(t) is the number of EVs in region i in forecast year t. The number of people who purchase EVs in each period is determined according to the consumer state decision model based on Markov chain, and then the EV ownership is obtained. Z(t) is the total number of EVs in forecast year t; ri is the travel volume ratio of region i.

[0060] In the consumer state decision model based on Markov chain, individual consumers are divided into four states, namely non-cognitive state S1 (consumers lack knowledge of electric vehicles), cognitive state S2 (consumers have a certain understanding of electric vehicles), willingness to buy state S3 (consumers have a willingness to buy electric vehicles) and purchase state S4 (consumers buy electric vehicles). Consumers in different states are affected by consumers, EVs and society, and their states are transferred. The specific transfer process is as follows:

[0061] (1) When consumer i is in state S1:

[0062] If the publicity impact generated by the advertisement exceeds the cognitive threshold of consumer i, the state of consumer i changes from S1 to S2; otherwise, S1 remains unchanged.

[0063] (2) When consumer i is in state S2:

[0064] If consumer i's cognitive ability cannot keep up with the technological development of EV (represented by the slope of the comprehensive technical performance curve), cognitive deficiencies will occur and the consumer will retreat from S2 to S1 with a certain probability. Otherwise, the perceived total utility considering consumer preferences and social network utility will be judged to determine whether the consumer is willing to purchase EV. When the perceived total utility is greater than the consumer's purchase intention threshold, the consumer's state will change from S2 to S3. When the perceived total utility is less than the threshold, the consumer remains in state S2.

[0065] (3) When consumer i is in state S3:

[0066] If the current comprehensive technical performance does not meet consumers' perceived quality expectations, consumers may not have a strong willingness to buy and will return from S3 to S2.

[0067] (4) When consumer i is in state S4

[0068] If the age of the consumer's car is less than its service life, there is no need to change the car, and the consumer remains in S4; otherwise, a demand for changing the car arises, that is, the consumer returns from S4 to S3.

[0069] Finally, the number of consumers who decide to buy EVs in each period is counted and accumulated to obtain the final number of buyers Z(t), as shown in formula (12).

[0070]

[0071] Step 2: By combining the improved gravity model with temporal and spatial characteristics, a user multi-time scale transfer probability dataset is obtained to characterize user travel behavior.

[0072] EV charging load is affected by travel behavior. Considering different seasons and traffic flow conditions, the gravity model is used to characterize the daily travel behavior of vehicles. Combined with the large amount of spatial and temporal information included in the travel chain model, Figure 2 According to the activity type, the travel chain can be roughly divided into 8 types: Figure 3 As shown in the figure, various types of information in the travel chain are analyzed in detail through the gravity model.

[0073] Define the regional transition probability matrix A for a period of time t t =[a ij t ] N×N Describes the travel transfer probability between regions in 24 time periods of a day, where N is the total number of regions; a ij t is the travel transfer probability from region i to region j in period t, where period t includes a multi-year time scale, four seasons, two typical day types (weekdays and weekends), and 24 hours a day. The final generated a ij t is a multi-time scale transition probability dataset. ij t The size is related to regional characteristics, such as being proportional to the travel force and attraction between regions and inversely proportional to traffic impedance. It is characterized by the gravity model, as shown in formula (13).

[0074]

[0075] Among them, θ t is the time adjustment coefficient, which is used to describe the time differences between different seasons and different day types; θf is the traffic flow adjustment coefficient, which is used to describe the difference in vehicle flow in the road network in different seasons and different types of days. The size of the parameter is firstly obtained by cleaning the EV historical data, and the information obtained is used to construct a decision matrix to define the attribute weight. The entropy technology method is used to use the discrete degree of the attribute value to achieve the initial solution of the size of the weight assignment parameter, and then it is optimized in the iterative process, so as to realize the dual constraints of the gravity model, and then more accurately obtain the transfer probability matrix data set. i t is the travel force of area i during period t; Q i t is the attraction of region j in period t. The travel force and attraction of a region are related to the characteristics of the region and can be obtained from traffic data; where f1(.) is the probability density function, μ is the mean, and σ is the standard deviation. Function F ij is the traffic impedance between area i and area j, which is determined by the distance l between the two places. ij Since the regional location data are mostly expressed in longitude and latitude, the distance between two longitude and latitude points can be calculated by the haversine equation, satisfying formula (15)

[0076]

[0077] Among them, lon1, lat1 and lon2, lat2 represent the longitude and latitude of location i and j respectively, and R is the radius of the earth (the radius of the earth is 6378km, R = 6378*1000m)

[0078] In addition, for the single-region single-time transition probability matrix A t =[a ij t ] N×N The sum of the transition probabilities from region i to other regions is 1, that is, the sum of the elements in the i-th row of the matrix should be 1, as shown in formula (16).

[0079]

[0080] Step 3: Construct a random charging behavior model considering multiple charging power levels and the randomness of destinations.

[0081] Considering the heterogeneous charging needs of EV users, at the level of charging infrastructure, following the logical framework of increasing charging efficiency, we select slow charging facilities (SCF), fast charging facilities (FCF) and ultra-fast charging facilities (UCF) as three representative types of electric vehicle charging piles for users to randomly choose. k , its charging power can be quantitatively described by formula (17).

[0082]

[0083] Among them, P EVk Indicates EV k The charging power, P SCF , P FCF , P UCF Respectively represent the rated charging power of slow charging pile, fast charging pile and super fast charging pile, T k dur For EV k The expected length of stay, Cap k For EV k Battery capacity, SOC k For EV k state of charge.

[0084] For each stop at the activity location in the travel chain, assuming that the charging facilities at the stop are sufficient, the owner needs to decide whether to charge (i.e., charging decision) at the stop and calculate the charging load. If the EV decides to stop and charge at a certain place, then it is assumed that the charging facilities are connected at the moment of arrival at the stop, and charging is continued until the ideal power is reached or the maximum stay time is reached, then the charging time T ch As shown in formula (18).

[0085]

[0086] Among them, SOC exp is the ideal battery state after charging, assuming it is 95% of the total battery capacity, τ s is the charging efficiency, T stay is the residence time. The charging efficiency is affected by temperature, and different seasons have different charging efficiencies.

[0087] According to the charging habits of the car owner, the charging decision is divided into two cases: deterministic charging and random charging. Deterministic charging: Generally, when the EV range is not enough to complete the next trip, it will choose to charge at the stop, as shown in formula (19):

[0088] SOC k -ε s d k+1 <SOC alert (19)

[0089] Among them, SOC k is the remaining power after the kth trip of the travel chain; ε s is the power consumption per unit mileage; d k+1 is the k+1th trip mileage; SOC alert The remaining power after the trip can be calculated by sampling each mileage using Monte Carlo simulation, and then calculated based on the current power and power consumption per unit distance. The power consumption per unit distance is affected by weather temperature and has different power consumption in different seasons.

[0090] However, when the power can support the next trip, the random charging behavior of the destination must be considered. While considering observable factors such as individual attributes and charging attributes, the present invention takes into account the impact of destination randomness on charging behavior. In view of this, the Destination-Stochastic charging behavior model (D-SCB) considering the randomness of the destination is used, and the station selection cost of the destination is integrated. When a certain activity in the travel chain stops, a judgment is made to estimate the charging psychological probability P. CM (Charging mental probability), when the charging mental threshold P CM When the value is greater than a certain threshold (assuming the threshold is 0.5), the user decides to charge at the stopover location, otherwise he decides not to charge. The specific decision-making process is as follows: Figure 4 shown.

[0091] Charging psychological probability P CM The calculation of is very important, and the internal mechanism and external environment of the charging decision process are deeply analyzed. The charging behavior satisfies the principle of utility maximization, that is, the option with the greatest utility (i.e., charging / not charging) is selected after comprehensive consideration of various factors. The station selection time and queuing time of EV users are described as the charging probability p cs and p qt, where the station selection time refers to the time it takes for the EV user to travel to the nearest charging station from the end of the activity, and v satisfies the negative exponential function distribution; the queuing time refers to the time that the EV user may wait in line for the required charging power charging pile after arriving at the charging station. The user's expectation of the available charging time is described as a probability, and the probability of the electric vehicle accepting charging is shown in formula (20).

[0092]

[0093] Where u is the average service rate; 1 / u is the average service time. When the number of electric vehicles is less than the number of charging piles in the charging station, the total service rate of the system is k; when the number of electric vehicles exceeds the number of charging piles in the system, the total service rate of the system is C; η is the parameter obtained after considering observable factors such as individual attributes and charging attributes, which can be obtained using the hybrid choice model HCM.

[0094] Step 4: Generate multi-time scale load forecast results through solving.

[0095] According to the above steps, the model is simulated and solved. If all regions have been simulated, the charging load curves of different seasons and typical days in each region in the tth year are counted to generate multi-time scale load forecast results.

[0096] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: combining the gravity model characterizing the travel chain and the random charging behavior model to realize the EV charging load prediction on multiple time scales such as different day types and different seasons in future years.

Claims

1. A multi-time scale electric vehicle load forecasting method based on a gravity model, characterized by: The following steps are involved: Step 1: Obtain the initial number of electric vehicles and basic information about user travel; Step 2: By combining the improved gravity model with temporal and spatial characteristics, a user multi-time scale transfer probability dataset is obtained to characterize the user's travel behavior; Step 3: Consider the multiple power levels of charging and the randomness of the destination to build a random charging behavior model; Step 4: Generate multi-time scale load forecast results through iterative solution.

2. The gravity model improved by combining the time characteristics as claimed in claim 1 is used to obtain a user multi-time scale transfer probability data set to describe the transfer of electric vehicle users in different areas, characterized in that: By adopting the improved gravity model combining the time and space characteristics, a user multi-time dimension transfer probability data set is obtained to characterize the user travel behavior.

3. As claimed in claim 1, the randomness of multiple power levels is first considered, and the selected power level type is determined by analyzing variables such as charge state, residence time and power consumption, and the randomness of the charging destination is characterized by analyzing the user's charging station selection time and queuing cost, and a random charging behavior model is constructed, characterized by: Considering the randomness of multi-power level charging types and charging destinations, a random charging behavior model is constructed.

4. The method for generating a multi-time scale load forecast result according to claim 1, wherein: Considering the progress of planning and construction of EV charging facilities and power grids in the future, and combining short-term and medium-term characteristics, multi-time scale EV charging load forecasting for different day types and seasons in future years is achieved.

Images