Method for generating electric vehicle charging demand based on community clustering and road network simulation

Abstract

A method for generating electric vehicle charging requirements based on community clustering and road network simulation comprises the following steps that firstly, electric vehicle sample information is generated through a Monte Carlo sampling method, and the electric vehicle sample information comprises travel origin and destination, vehicle battery parameters and the like; thirdly, generating a road network model, representing road network nodes by using IEEE33 power distribution network nodes, and establishing a road network-power distribution network coupling model; then, the travel path of each electric vehicle is determined according to a Dijkstra algorithm, the electric quantity change is calculated, and corresponding charging operation is executed according to the user charging logic to generate a charging demand; and finally, carrying out clustering operation on the charging demands by using a community detection algorithm based on breadth-first search and clustering evaluation, accumulating the charging demands in the region to a clustering center, carrying out iterative calculation by using a Davies-Bouldin index to determine an optimal clustering number, and obtaining simplified spatial-temporal distribution of the charging demands of the electric vehicle. According to the method, the practicability of the charging demand can be guaranteed by combining the traffic network information, and the data processing process is simplified.

Description

Technical Field

[0001] The present invention relates to the field of new energy vehicle technology, in particular to the field of electric vehicle charging demand prediction technology, and specifically to a method for generating electric vehicle charging demand based on community clustering and road network simulation. Background Art

[0002] The methods for predicting electric vehicle charging demand are mainly divided into probability statistics-based methods and deep learning-based methods. The former mainly starts from EV user behavior, travel data, and EV power consumption model, establishes a corresponding travel model, and finally obtains the spatiotemporal distribution of EV charging demand by using the probability sampling method in statistics. The latter method based on deep learning mainly relies on historical charging data to train the network to predict EV charging demand. Existing studies have used methods such as recurrent neural networks and reinforcement learning to predict EV charging demand. However, due to the limited historical charging data of existing EVs, the probability statistics-based EV load prediction method is still the mainstream method.

[0003] The paper "Electric Vehicle Charging Demand Prediction Model Based on Dynamic Traffic Information and Analysis of Its Impact on Distribution Network" assumes that electric vehicles and fuel vehicles have the same travel patterns. The Monte Carlo sampling method is used to establish models such as daily mileage, start and end travel time, and charging characteristics. The path planned by the real-time Dijkstra algorithm is used to predict the distribution of electric vehicle charging load.

[0004] The model established by the above-mentioned prior art can effectively predict the load distribution of electric vehicle charging demand, and at the same time take into account the driving dynamic characteristics and charging random characteristics of electric vehicles. However, when the number of traffic network nodes is large and the number of EVs is huge, the EV charging demand data is not only large in quantity, but also highly dispersed in time and space. If it is directly used in a calculation model such as charging station planning, the charging plan for each EV needs to be determined in the optimization model, which leads to too many optimization variables and makes the optimization model difficult to solve. Summary of the invention

[0005] The purpose of the present invention is to solve the technical problems in the prior art that the charging demand data is too complicated, and the dynamic characteristics of electric vehicle driving and the random characteristics of charging are ignored, which makes it difficult to obtain real charging demand data and accurately predict the charging load. A method for generating electric vehicle charging demand based on Monte Carlo sampling and community clustering is proposed.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A method for generating electric vehicle charging demand based on community clustering and road network simulation includes the following steps:

[0008] Step 1: Use the Monte Carlo sampling method to generate several electric vehicle (EV) sample information in the road network. The EV samples are divided into private car samples, taxi samples, and public car samples according to vehicle type;

[0009] Step 2: Establish a road network-distribution network coupling model, which includes a road network model and a distribution network model; directly use the distribution network nodes to represent the road network nodes in the road network model, add the road weight model and time period set to the basic structure of the road network model, and generate a road network topology mathematical model;

[0010] Step 3: Based on the starting and ending points of the electric vehicle sample information obtained in Step 1.1, the Dijkstra algorithm is used to determine the shortest travel path for each EV to the destination in the road network diagram generated in Step 2, and the shortest travel path is divided into several corresponding sections according to the nodes passed through. At the end of each section of the road, the energy consumption is calculated according to the electric vehicle energy consumption model, and finally the power level SOC of each electric vehicle sample is obtained;

[0011] Step 4: According to the estimated power level SOC2 of each electric vehicle arriving at the next node obtained in Step 3.2, determine whether the user needs to charge. If charging is required, fast charging or slow charging is performed according to the vehicle type of the user, so that the charging load of each user in a specific time period is superimposed on the corresponding road network node in the time period to obtain the spatiotemporal distribution of the 24-hour charging demand of 1,000 electric vehicle users in the entire road network;

[0012] Step 5: The 24-hour EV charging demand spatiotemporal distribution diagram obtained by Step 4.5; use the adaptive community detection clustering algorithm based on breadth-first search (BFS) and clustering evaluation to perform clustering operations on the charging demand of each node, accumulate the charging demand of all nodes within a specific range to the cluster center node, and use the DBI (Davies-Bouldin) indicator to iteratively calculate to determine the cluster number σ to reduce the error caused by manual selection. Finally, after clustering, the simplified 24-hour EV charging demand spatiotemporal distribution diagram is obtained.

[0013] Step 1 includes the following steps:

[0014] Step 1.1: Randomly assign the starting and ending points of each electric vehicle user;

[0015] Step 1.2: Generate battery parameters C for each vehicle that follow the gamma distribution according to the vehicle model bat , battery parameters C bat Specifically:

[0016]

[0017] In the formula: are the battery capacity parameters of three types of EVs; α t and β t are gamma distribution parameters; Γ(α t ) is the normalization constant of the gamma distribution. See Table 1 for the specific values above.

[0018] Step1.3: Generate the average vehicle speed, travel return time, and initial vehicle battery level SOC0 (State of Charge) that follow a normal distribution.

[0019] In Step2, the generation steps of the road network topology mathematical model are as follows:

[0020]

[0021] In the formula: G is the overall model of the traffic road network; V represents the set of all nodes of graph G; E represents the set of all road segments of graph G; W is the set of road weights, that is, the road impedance; K represents the set of divided time periods, that is, the whole day is divided into 24 time periods; the road weight W represents the travel cost of the road.

[0022] In Step2, when obtaining the road weight model, the following steps are adopted:

[0023] Step2.1: Road impedance model:

[0024]

[0025] In the formula: the saturation Q is the road traffic flow, C is the road capacity; t0 is the zero-flow travel time; α, β are impedance influence factors, where both the road traffic flow and the road capacity follow a normal distribution, represents the road impedance when the saturation S is 0 < S ≤ 1.0, represents the road impedance when the saturation S is 1.0 ≤ S ≤ 2.0. The relevant data is shown in Table 3 below.

[0026] Step2.2: Node impedance model:

[0027]

[0028] In the formula: c is the traffic light signal cycle; γ is the green signal ratio (i.e., the ratio of the green light time to the whole signal cycle); q is the vehicle arrival rate per minute on the road, represents the node impedance when the saturation S is 0 < S ≤ 1.0, represents the node impedance when the saturation S is S > 0.6. The relevant data is shown in Table 3 below.

[0029] Step2.3: The road weight model can be obtained by combining Step2.1 and Step2.2:

[0030]

[0031] Where: represents the road impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the road impedance when the saturation S satisfies 1.0 ≤ S ≤ 2.0, represents the node impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the calculation result of the node impedance model when the saturation S > 0.6. See Figure 5 ;

[0032] In Step3, it specifically includes the following steps;

[0033] Step3.1 Electric vehicle energy consumption model

[0034]

[0035] Where: is the road weight at the road v where the vehicle is located at time k generated in Step2.3, ij V car is the average vehicle speed.

[0036] Step3.2 When the vehicle arrives at each distribution network node, calculate the remaining battery charge (SOC) of the vehicle according to the power consumption:

[0037]

[0038] Where: SOC0 represents the initial battery charge level (State of Charge), SOC1 and SOC2 represent the current battery charge level and the expected battery charge level when arriving at the next node respectively; C bat represents the battery capacity; X1 represents the cumulative energy consumption generated by all the roads the vehicle has traveled so far, and X2 represents the expected energy consumption when the vehicle travels to the next distribution network node.

[0039] In Step4, it specifically includes the following steps:

[0040] Step4.1 When the user arrives at each distribution network node, make a judgment. When the SOC2 of a private car < 0 or the SOC2 of a bus and a taxi < 0.25, the user needs to charge.

[0041] Step4.2 Compare the expected battery charge levels to determine whether charging is needed. If the remaining battery charge SOC2 of a private car is less than 0, slow charging is required, while if the SOC2 of a bus and a taxi < 0.25, fast charging is required. The charging time calculation formula is as follows:

[0042] Fast charging:

[0043]

[0044] Slow charging:

[0045]

[0046] Where: Take the slow charging power P slow =7KW, fast charging power P fast =60KW,C bat Indicates the battery capacity. SOC1 and SOC2 respectively represent the current power level and the expected power level to reach the next node.

[0047] Step 4.3 Obtain the time period from the start to the end of charging for each user in the distribution network based on the user's charging time. The calculation formula is as follows:

[0048] 1) Private car charging starts start and end time t end :

[0049]

[0050] t start =t floor

[0051] t end =t floor +T slow

[0052] Where: V car is the average speed of the vehicle; L ture is the accumulated mileage of the current node vehicle, t floor The cumulative time the vehicle travels at the current node.

[0053] 2) Charging start and end time for public buses and taxis:

[0054]

[0055] t start =t floor

[0056] t end =t floor +T fast

[0057] Where: V car is the average speed of the vehicle; L ture is the accumulated mileage of the current node vehicle, t floorThe cumulative time the vehicle travels at the current node.

[0058] Step 4.4 Obtain the charging load of each user (i.e., each electric vehicle) in the distribution network according to the user's charging time, i.e., the user's charging load in v i The charging demand at the node at time k, the charging load calculation formula is as follows:

[0059] Private car charging load:

[0060] P i,k =T slow *P slow

[0061] Charging load for public vehicles and taxis:

[0062] P i,k =T fast *P fast

[0063] Where k is the time at which charging starts. start and end time t end And the travel time of the vehicle is determined, the slow charging power P slow =7KW, fast charging power P fast =60KW.

[0064] Step 4.5 accumulates the charging demand of each electric vehicle in the same period to the corresponding node (i.e. the charging node) to obtain the overall 24-hour EV charging demand spatiotemporal distribution diagram of the system.

[0065] In Step 5, the following steps are specifically included:

[0066] Step 5.1. Randomly select a node from the unassigned nodes as a seed node, and use breadth-first search (BFS) to expand the community starting from the seed node until the predetermined maximum number of nodes is reached.

[0067] Step 5.2, reassign nodes in multiple iterations to ensure that each node is assigned to the most appropriate community, and update the community cluster center node by calculating the sum of the distances from the node in the community to all nodes (road weight is used here instead of node distance), and in this process, the charging demand corresponding to the nodes contained in the community is included in the cluster center.

[0068] Step 5.3: Use DBI (Davies-Bouldin) indicator I DBI Automatically determine the number of clusters q to reduce the error caused by manual selection. This indicator is defined as:

[0069]

[0070] Where: σ is the number of clusters, as shown in Table 3, when σ = 8, I DBI Minimum, C i and C j are the average distances from i and j to the corresponding cluster centers (here the road weight is used to represent the distance), D i,j is the distance from the center of cluster i to the center of cluster j; the smaller the Davies-Bouldin index is, the better the clustering performance is.

[0071] The present invention has the following technical effects:

[0072] 1) The present invention can use Monte Carlo sampling to generate electric vehicle user-related parameters such as battery parameters, initial battery power, average speed, etc. to ensure that the data has a certain degree of randomness;

[0073] 2) In the present invention, each electric vehicle is used to travel according to the start-end OD path determined by the real-time Dijkstra algorithm, and the EV energy parameters are updated in real time to obtain the charging characteristic parameters;

[0074] 3) The present invention can not only ensure that the generated electric vehicle samples are more consistent with the actual operation of electric vehicles, but also ensure the interaction between each sample and the traffic network, and use clustering algorithms to reduce the complexity of charging requirements and facilitate data processing;

[0075] 4) The clustering algorithm provided by the present invention combines breadth-first search (BFS) and Davies-Bouldin Index (DBI) for community detection and evaluation. Its characteristics are adaptive selection of the number of clusters, expansion of the community through breadth-first search and multiple iterations of redistribution of nodes to ensure that each node is reasonably assigned to the appropriate community. The steps include: first, randomly selecting seed nodes from unassigned nodes, and forming communities through BFS expansion; then, iteratively redistributing nodes to ensure that the number of nodes in each community is within a predetermined range, and updating the community center; finally, using DBI to evaluate the effects of different numbers of clusters, and selecting the optimal number of clusters for final clustering. The advantage of this algorithm is that it can adaptively select the optimal number of clusters based on the data, ensuring the rationality and stability of the clustering results, while taking into account the demand optimization in actual application scenarios;

[0076] 5) The present invention first uses Monte Carlo sampling to generate electric vehicle travel and battery parameters, and establishes a road network-distribution network coupling model to calculate the vehicle's energy consumption and remaining power; then, the corresponding charging operation is triggered according to the charging demand; finally, the community clustering algorithm is used to simplify the spatiotemporal distribution of charging demand, thereby effectively reducing data complexity and improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0078] Figure 1 is a flow chart of the present invention;

[0079] Figure 2 This is a schematic diagram of the number of different types of vehicles at each traffic node sampled by Monte Carlo;

[0080] Figure 3 A schematic diagram of charging demand for each node in each time period during the day;

[0081] Figure 4 This is a schematic diagram of the total charging demand of each distribution network node in 24 hours;

[0082] Figure 5 Traffic network diagram for calibrating road network weights;

[0083] Figure 6 This is a schematic diagram of the optimal clustering result obtained by iteration with reference to DBI parameters;

[0084] Figure 7 Schematic diagram of the total charging demand of each cluster center for 24 hours under the optimal DBI parameters;

[0085] Figure 8 A schematic diagram of the charging demand of each cluster center at each time period during the day;

[0086] Fig. 9 A schematic diagram showing the performance comparison between the algorithm proposed in the present invention and other traditional algorithms;

[0087] Fig.10 Schematic diagram of clustering quality and running time of different algorithms under different cluster numbers;

[0088] Fig.11 It is the basic structure diagram of the road network model;

[0089] Fig.12 This is the structure diagram of the IEEE33-node distribution network. DETAILED DESCRIPTION

[0090] like Figure 1 As shown, a method for generating electric vehicle charging demand based on community clustering and road network simulation includes the following steps:

[0091] Step 1. Use the Monte Carlo sampling method to generate 1,000 electric vehicle (EV) sample information in the road network, of which private cars account for 25%; taxis account for 45%; and public buses account for 30%. The different types of vehicles will affect the subsequent charging behavior of users.

[0092] Private cars: Private cars are idle most of the time, while commercial vehicles are in operation. The travel destinations of private cars are more fixed than those of commercial vehicles, and the charging method is mainly slow charging at the destination, which has a strong regularity.

[0093] Taxi: The travel time and starting and ending points of this type of vehicle are relatively random, the driving route is not fixed and the number of trips is frequent. The charging method is to go to the charging station for fast charging, and given the long driving distance, it may be charged multiple times within a day.

[0094] Public vehicles: This type of vehicle is mainly functional vehicles such as logistics vehicles and sanitation vehicles. Its starting and ending points are not fixed, the number of trips is large, the charging time and charging location are not fixed, which is similar to taxis. According to the charging needs, you can choose to slow charge at the destination or choose to fast charge at the charging station during the trip. For buses, given their fixed driving routes, dedicated charging facilities and the same vehicle model, their charging characteristics are less affected by traffic factors, so buses are not taken into consideration.

[0095] The specific steps are as follows:

[0096] Step 1.1 Randomly assign the starting and ending points for each electric vehicle user.

[0097] Step 1.2 Generate the battery capacity parameter C of each vehicle that follows the gamma distribution according to the vehicle model bat , gamma distribution related data reference

[0098] Table 1, the formula is as follows:

[0099]

[0100] Where: are the standard battery capacity parameters for three types of EVs; α t and β t is the gamma distribution parameter. According to common electric vehicle data, the private car parameters are set as α1=10.08,β1=0.8;for taxi α2=4.5,β2=6.3;public car is α3=8.7, β3=3.2.

[0101] Step 1.3 Generate the average vehicle speed, travel return time and vehicle initial power SOC0 (state of charge, SOC0 is the battery level of the electric vehicle when it sets off) that obey the normal distribution. Please refer to Table 1 for the normal distribution related data.

[0102]

[0103] Table 1 Monte Carlo Sampling Composition Information Table

[0104] Step2. Establish a road network - distribution network coupling model, which includes a road network model and a distribution network model; the distribution network model is the IEEE33 - node standard distribution network model, and its structure is as Fig.12 shown; the road network model directly uses the distribution network nodes to represent the road network nodes, and the basic structure of the road network model is as Fig.11 shown. On this basis structure, add a road weight model and a set of time periods to generate a road network topology mathematical model.

[0105] The generation steps of the road network topology mathematical model are as follows:

[0106]

[0107] In the formula: G is the road network topology mathematical model; V represents the set of all nodes in the road network; E represents the set of all roads in the road network; W is the set of road weights; K represents the set of divided time periods, that is, the whole day is divided into m time periods;

[0108] The road weight represents the cost of road travel, which is composed of road impedance and node impedance. The electric vehicles studied in this invention mainly target urban internal roads. In the urban road network, traffic lights are generally set for control. Therefore, this invention will introduce the concepts of road impedance and node impedance when calculating the road weight. Therefore, the road weight model is composed of a road impedance model and a node impedance model.

[0109] This invention uses the saturation S index to describe different urban traffic conditions: unobstructed (0 < S ≤ 0.6), slow - moving (0.6 < S ≤ 0.8), congested (0.8 < S ≤ 1.0), and severely congested (1.0 < S ≤ 2.0). Different saturations can be obtained according to different road intersections and road capacities, and then different road impedances and node impedances can be obtained.

[0110] Step2.1 Road Impedance Model

[0111]

[0112] In the formula: saturation Q is the road traffic flow, C is the road capacity; t0 is the zero - flow travel time; α, β are impedance influence factors. Among them, both the road traffic flow and the road capacity follow a normal distribution, represents the road impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the road impedance when the saturation S satisfies 1.0 ≤ S ≤ 2.0. The relevant data is shown in Table 2 below.

[0113] Step2.2 Node Impedance Model

[0114]

[0115] Where: c is the traffic signal cycle; γ is the green ratio (i.e., the proportion of green light time in the whole signal cycle); q is the arrival rate of road vehicles per minute, represents the node impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the node impedance when the saturation S > 0.6. The relevant data is shown in Table 2 below.

[0116] Combining Equation (3) and Equation (4) can obtain the complete road impedance model.

[0117] Road traffic flow (Q) <![CDATA[Normal distribution N(1800, 150 2 )]]> Road capacity (C) <![CDATA[Normal distribution N(2100, 200 2 )]]> <![CDATA[Zero-flow travel time (t0)]]> <![CDATA[t0 = 10 (minutes)]]> Impedance influence factors (α, β) α＝1.3、β＝1.2 Traffic light signal cycle (c) c = 30 (seconds) Green letter ratio (γ) 0.7 Road vehicle arrival rate per minute (q) 2(vehicles / minute)

[0118] Table 2 Road Information Table Step2.3 Combining Step2.1 and Step2.2 can obtain the road weight model

[0119]

[0120] Where: represents the road impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the road impedance when the saturation S satisfies 1.0 ≤ S ≤ 2.0, represents the node impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the calculation result of the node impedance model when the saturation S > 0.6. See Figure 5

[0121] Step3. According to the origin-destination locations in the electric vehicle sample information obtained in Step1.1, use the Dijkstra algorithm to determine the shortest travel path for each EV to the destination in the road network graph generated in Step2 (here, the road weight can represent the road length, and the higher the weight, the longer the road), and split it into corresponding roads according to the nodes passed through. Calculate the energy consumption according to the electric vehicle energy consumption model, and finally obtain the remaining battery charge SOC (State of Charge) of each electric vehicle sample;

[0122] Step3.1 Electric Vehicle Energy Consumption Model

[0123]

[0124] Where: is the road weight at the road v where the vehicle is located at time k generated in Step2.3 ij at car is the average vehicle speed.

[0125] Step 3.2 When the vehicle reaches a distribution network node, the remaining power (SOC) of the vehicle is calculated based on the power consumption:

[0126]

[0127] Where: SOC0 represents the initial state of charge, SOC1 and SOC2 represent the current power level and the expected power level to reach the next node respectively; C bat represents the battery capacity; X1 represents the cumulative energy consumption of all roads that the vehicle has traveled so far, and X2 represents the energy consumption expected to be generated when the vehicle reaches the next distribution network node.

[0128] Step 4: Determine whether the user has charging demand based on the estimated power level SOC2 of each electric vehicle obtained in Step 3.2, and judge the charging demand model. When the charging demand is triggered, perform corresponding slow charging at the destination or fast charging at the charging station. After charging, superimpose the charging load of each user in a specific time period to the corresponding road network node in the time period to obtain the spatiotemporal distribution of charging demand of 1,000 electric vehicle users in the entire area.

[0129] Step 4.1 Every time the user arrives at a distribution network node, a judgment is made. When the SOC2 of a private car is less than 0 or the SOC2 of a public car or taxi is less than 0.25, the user needs to charge.

[0130] Step 4.2 Compare the estimated power level to determine whether charging is needed. If the remaining power SOC2 of a private car is less than 0, slow charging is required, while public cars and taxis with SOC2 < 0.25 require fast charging. The charging time calculation formula is as follows:

[0131] Fast charging:

[0132]

[0133] Slow charging:

[0134]

[0135] Where: Take the slow charging power P slow =7KW, fast charging power P fast =60KW, C bat Indicates the battery capacity. SOC1 and SOC2 respectively represent the current power level and the expected power level to reach the next node.

[0136] Step 4.3 Obtain the time period from the start to the end of charging for each user in the distribution network based on the user's charging time. The calculation formula is as follows:

[0137] 1) Private car charging startsstart and end time t end :

[0138]

[0139] t start =t floos

[0140] t end =t floor +T slow (9)

[0141] Where: V car is the average speed of the vehicle; L ture is the accumulated mileage of the current node vehicle, t floor The cumulative time the vehicle travels at the current node.

[0142] 2) Charging start and end time for public buses and taxis:

[0143]

[0144] t start =t floor

[0145] t end =t floor +T fast (10)

[0146] Step 4.4 Obtain the charging load of each user (i.e., each electric vehicle) in the distribution network according to the user's charging time, i.e., the user's charging load in v i The charging demand at the node at time k, the charging load calculation formula is as follows:

[0147] Private car charging load:

[0148] P i,k =T slow *P slow (11)

[0149] Charging load for public vehicles and taxis:

[0150] P i,k =T fast *P fast (12)

[0151] Where k is the time corresponding to the start time of charging t start and end time t end And the travel time of the vehicle is determined.

[0152] Step 4.5: Accumulate the charging demand of each electric vehicle in the same period to the corresponding node (i.e. the node where the charging is located) to obtain the overall 24-hour EV charging demand spatiotemporal distribution diagram of the system, see Figure 4 ;

[0153] Step 5. The spatiotemporal distribution diagram of EV charging demand for 24 hours obtained in Step 4.5; clustering operation is performed on the charging demand of each node using an adaptive community detection clustering algorithm based on breadth-first search (BFS) and cluster evaluation, and the charging demand of all nodes in an area is accumulated to the cluster center node. The DBI (Davies-Bouldin) indicator is iteratively calculated to determine the number of clusters q to reduce the error caused by manual selection;

[0154] Step 5.1. Randomly select a node from the unassigned nodes as a seed node, and use breadth-first search (BFS) to expand the community starting from the seed node until the predetermined maximum number of nodes is reached.

[0155] Step 5.2, reassign nodes in multiple iterations to ensure that each node is assigned to the most appropriate community, and update the community cluster center node by calculating the sum of the distances from the node in the community to all nodes (road weight is used here instead of node distance), and in this process, the charging demand corresponding to the nodes contained in the community is included in the cluster center.

[0156] Step 5.3. Use Davies-Bouldin indicator I during this period DBI Automatically determine the number of clusters q to reduce the error caused by manual selection. This indicator is defined as:

[0157]

[0158] Where: σ is the number of clusters, as shown in Table 3, when σ = 8, I DBI Minimum, C i and C j are the average distances from i and j to the corresponding cluster centers (road weights are used here to represent distances), D i,j is the distance from the center of cluster i to the center of cluster j; the smaller the Davies-Bouldin index is, the better the clustering performance is.

[0159] Figure 6 is the community distribution corresponding to the optimal clustering result, at this time I DBI =0.89502. The asterisk in the figure represents the node where the cluster center is located. Each cluster center corresponds to a cluster range (i.e., community). The charging demands of all points in the community are accumulated and added to the cluster center node. The nodes contained in each community and the corresponding cluster centers are shown in Table 3:

[0160]

[0161]

[0162] Table 3 Cluster centers and pulled nodes list Figure 8 The charging demand of each cluster center at different time periods is shown in Table 4:

[0163]

[0164] Table 4 Distribution of charging demand in cluster centers over 24 hours

[0165] Figure 7 and Figure 8 The total charging demand histogram after clustering and the spatiotemporal distribution of charging demand at the cluster center are compared. Figure 3 Figure 4 It can be seen that the complexity of the data has been significantly reduced.

[0166] Fig.11 The relevant parameters of the IEEE 33-node standard distribution network are shown in Table 5 and Table 6:

[0167]

[0168]

[0169] Table 5 IEEE33-node standard distribution network related parameters

[0170] <![CDATA[Reference voltage U B > 10KV <![CDATA[Reference power S B > 10MVA

[0171] Table 6 IEEE33 voltage level and reference power

[0172] There are also conventional loads in the distribution network that do not change over time, as shown in Table 7:

[0173] Node Number Active power(MW) Reactive power (MVA) 1 0 0 2 100 60 3 90 40 4 120 80 5 60 30 6 60 20 7 200 100 8 200 100 9 60 20 10 60 20 11 45 30 12 60 35 13 60 35 14 120 80 15 60 10 16 60 20 17 60 20 18 90 40 19 90 40 20 90 40 21 90 40 22 90 40 23 90 50 24 420 200 25 420 200 26 60 25 27 60 25 28 60 20 29 120 70 30 200 600 31 150 70 32 210 100 33 60 40

[0174] Table 7 Conventional load

[0175] Take the reference voltage as 10KV, the benchmark as 10MVA, and superimpose the conventional load and the load generated by the electric vehicle charging demand to calculate the voltage offset of the distribution network node, such as Fig. 9 As shown in the figure, due to the large number of EVs connected for charging, the voltage of the distribution network nodes has dropped and shifted. Among them, the voltage curves have the largest drop in the period of 17:00-18:00, and the drop at traffic node 18 is higher than that at other traffic nodes, with a drop of 6.6%.

[0176] The clustering algorithm provided by the present invention combines breadth-first search (BFS) and Davies-Bouldin Index (DBI) for community detection and evaluation. It is characterized by adaptively selecting the number of clusters, expanding the community through breadth-first search and iteratively redistributing nodes multiple times to ensure that each node is reasonably assigned to a suitable community so that the DBI index is minimized. The steps include: first, randomly selecting seed nodes from unassigned nodes, and forming communities through BFS expansion; then, iteratively redistributing nodes to ensure that the number of nodes in each community is within a predetermined range, and updating the community center; finally, using the DBI index to evaluate the effects of different numbers of clusters, and selecting the optimal number of clusters for final clustering. The advantage of this algorithm is that it can adaptively select the optimal number of clusters according to the data, ensure the rationality and stability of the clustering results, and take into account the demand optimization in the actual application scenario. It can significantly reduce the complexity of charging demand-related data, provide a reference for the site selection and capacity determination of subsequent electric vehicle charging stations, and reduce the difficulty of calculation.

[0177] Fig.10 The DBI Comparison graph in the figure shows the clustering quality of different algorithms under different cluster numbers. The solid line in the figure is the clustering algorithm used in this patent. The DBI value corresponding to the lowest point of the curve is the smallest, indicating that the algorithm has the best clustering effect under this cluster number. The trend of the curve change can also reflect the stability of the algorithm under different cluster numbers.

[0178] The Runtime Comparison graph shows the running time of different algorithms under different cluster numbers. The solid line in the graph is the clustering algorithm used in this patent. The lowest point of the curve corresponds to the shortest running time, indicating that the algorithm is most efficient under this number of clusters. The trend of the curve can also reflect the computational complexity of the algorithm under different numbers of clusters.

[0179] Comprehensive analysis shows that the algorithm of the present invention has the lowest DBI index and the best comprehensive clustering performance when performing clustering operations, but the operating rate is low.

Claims

1. A method for generating electric vehicle charging demand based on community clustering and road network simulation, characterized in that: The following steps are involved: Step 1: Use the Monte Carlo sampling method to generate several electric vehicle EV sample information in the road network. The electric vehicle samples are divided into private car samples, taxi samples, and public car samples according to vehicle type; Step 2: Establish a road network-distribution network coupling model, which includes a road network model and a distribution network model; directly use the distribution network nodes to represent the road network nodes in the road network model, add the road weight model and time period set to the basic structure of the road network model, and generate a road network topology mathematical model; Step 3: Based on the starting and ending points of the electric vehicle sample information obtained in Step 1, the Dijkstra algorithm is used to determine the shortest travel path for each EV to the destination in the road network diagram generated in Step 2, and the path is divided into several corresponding sections according to the nodes passed through. At the end of each section of the road, the energy consumption is calculated according to the electric vehicle energy consumption model, and finally the power level SOC of each electric vehicle sample is obtained; Step 4: According to the estimated power level SOC2 of each electric vehicle arriving at the next node obtained in Step 3, determine whether the user needs to charge. If charging is required, fast charging or slow charging is performed according to the vehicle type of the user, so that the charging load of each user in a specific time period is superimposed on the corresponding road network node in the time period to obtain the spatiotemporal distribution of the 24-hour charging demand of several electric vehicle users in the entire road network; Step 5: The 24-hour EV charging demand spatiotemporal distribution diagram obtained from Step 4; the charging demand of each node is clustered using an adaptive community detection clustering algorithm based on breadth-first search BFS and clustering evaluation, and the charging demand of all nodes within a specific range is accumulated to the cluster center node. The DBI indicator is iteratively calculated to determine the cluster number σ to reduce the error caused by manual selection. Finally, the simplified 24-hour EV charging demand spatiotemporal distribution diagram is obtained after clustering.

2. The method according to claim 1, characterized in that Step 1 includes the following steps: Step 1.1: Randomly assign the starting and ending points of each electric vehicle user; Step 1.2: Generate battery parameters C for each vehicle that follow the gamma distribution according to the vehicle model bat , battery parameters C bat Specifically: Where: is the battery capacity parameter of three types of EV; α t and β t is the gamma distribution parameter; Γ(α t ) is the normalizing constant of the gamma distribution; Step 1.3: Generate the vehicle average speed, travel return time and vehicle initial power level SOC0 that follow the normal distribution.

3. The method according to claim 1, characterized in that In Step 2, the steps for generating the mathematical model of the road network topology are: Where: G is the overall model of the traffic network; V represents the set of all nodes in graph G; E represents the set of all road sections in graph G; W is the set of road weights, i.e., road resistance; K represents the set of divided time periods, i.e., the whole day is divided into 24 time periods; and the road weight W represents the cost of road travel.

4. The method according to claim 1, characterized in that: In Step 2, when obtaining the road weight model, the following steps are taken: Step 2.1: Road impedance model: Where: Saturation Q is the road traffic flow, C is the road capacity; t0 is the zero-flow travel time; α and β are impedance influence factors, where both the road traffic flow and the road capacity follow a normal distribution, represents the road impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the road impedance when the saturation S satisfies 1.0 ≤ S ≤ 2.0; Step 2.2: Node impedance model: Where: c is the traffic signal cycle; γ is the green ratio; q is the arrival rate of road vehicles per minute, represents the node impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the node impedance when the saturation S satisfies S > 0.6; Step 2.3: Combining Step 2.1 and Step 2.2, we can get the road weight model: In the formula: represents the road impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the road impedance when the saturation S satisfies 1.0 ≤ S ≤ 2.0, represents the node impedance when the saturation S satisfies 0 < S ≤ 1.0, represents the node impedance when the saturation S satisfies S > 0.

6.

5. The method according to any one of claims 1 to 4, characterized in that: In Step 3, the following steps are specifically included: Step 3.1 Electric vehicle energy consumption model Where: is the road v where the vehicle is located in the k-th period generated in Step 2.3 ij The road weight at V car is the average vehicle speed; Step 3.2 When the vehicle reaches a distribution network node, the remaining power SOC of the vehicle is calculated based on the power consumption: Where: SOC0 represents the initial power level, SOC1 and SOC2 represent the current power level and the expected power level to reach the next node respectively; C bat represents the battery capacity; X1 represents the cumulative energy consumption of all roads that the vehicle has traveled so far, and X2 represents the energy consumption expected to be generated when the vehicle reaches the next distribution network node.

6. The method according to claim 5, characterized in that In Step 4, the following steps are specifically included: Step 4.1 The user makes a judgment at each distribution network node. When the SOC2 of a private car is less than 0 or the SOC2 of a public car or taxi is less than 0.25, the user needs to charge. Step 4.2 Compare the estimated power level to determine whether charging is needed; if the remaining power SOC2 of a private car is less than 0, slow charging is required, while public cars and taxis SOC2 < 0.25 require fast charging; the charging time calculation formula is as follows: Fast charging: Slow charging: Where: P slow is the slow charging power, P fast is the fast charging power, C bat Indicates the battery capacity, SOC1 and SOC2 respectively represent the current power level and the expected power level to reach the next node; Step 4.3 Obtain the time period from the start to the end of charging for each user in the distribution network based on the user's charging time. The calculation formula is as follows: 1) Private car charging starts start and end time t end : t start =t floor t end =t floor +T slow Where: V car is the average speed of the vehicle; L ture is the accumulated mileage of the current node vehicle, t floor The accumulated time of vehicle travel at the current node of the vehicle; 2) Charging start and end time for public buses and taxis: t start =t floor t end =t floor +T fast Where: V car is the average speed of the vehicle; L ture is the accumulated mileage of the current node vehicle, t floor The accumulated time of vehicle travel at the current node of the vehicle; Step 4.4 Obtain the charging load of each user in the distribution network according to the user's charging time, that is, the user's charging load at v i The charging demand at the node at time k, the charging load calculation formula is as follows: Private car charging load: P i,k =T slow *P slow Charging load for public vehicles and taxis: P i,k =T fast *P fast Where k is the time corresponding to the start time of charging t start and end time t end and the travel time of the vehicle is determined; Step 4.5 accumulates the charging demand of each electric vehicle in the same period to the corresponding node to obtain the overall 24-hour EV charging demand spatiotemporal distribution diagram of the system.

7. The method according to claim 5, characterized in that In Step 5, the following steps are specifically included: Step 5.1, randomly select a node from the unassigned nodes as a seed node, and use breadth-first search BFS to expand the community from the seed node until the predetermined maximum number of nodes is reached; Step 5.2, iterate and reallocate nodes multiple times to ensure that each node is assigned to the most appropriate community. Update the community cluster center node by calculating the sum of the distances from the node in the community to all nodes, and in this process, include the charging demand corresponding to the nodes included in the community into the cluster center; Step 5.3, use DBI indicator I during DBI Automatically determine the number of clusters q to reduce the error caused by manual selection; this indicator is defined as: Where: σ is the number of clusters, as shown in Figure 7, when σ = 8, I DBI Minimum, C i and C j are the average distances from i and j to the corresponding cluster center, D i,j is the distance from the center of cluster i to the center of cluster j; the smaller the Davies-Bouldin index is, the better the clustering performance is.

Images