Electric vehicle charging load prediction method based on travel characteristics

Abstract

The invention provides an electric vehicle charging load prediction method based on travel characteristics, which adopts a Monte Carlo method to simulate travel and charging behaviors of electric vehicles, and comprises the following steps: based on a travel chain model, extracting the initial travel time, the initial travel place, the initial charge state and the travel purpose of each time period of each electric vehicle; according to the travel purpose, extracting the functional area which each electric vehicle needs to reach in each time period; according to the functional area needing to be reached, path planning is carried out on the driving path of each electric vehicle; calculating the driving energy consumption on the driving path, and updating the charge state of the electric vehicle; the charging probability of the electric vehicle is judged according to the charge state; and finally, according to the charging probability of each electric vehicle in each area in different time periods, obtaining the charging load space-time distribution of the electric vehicles. By comprehensively considering various travel characteristics, the reliability of the prediction method is enhanced, and the problem of insufficient prediction precision is solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of load prediction, and particularly relates to an electric vehicle charging load prediction method based on travel characteristics. BACKGROUND

[0002] With the rapid increase in the number of electric vehicles, there will be a situation of concentrated charging in peak hours, which leads to the peak-peak phenomenon of distribution network load. Large-scale unordered access of electric vehicles to the power grid will cause problems such as increased load peak-valley difference, over-limit operation constraint, and decreased power quality, thereby affecting the stability and economy of the power system. Therefore, predicting the electric vehicle charging load is of great significance for evaluating the impact of unordered charging on the distribution network, formulating distribution network power planning, and optimizing the orderly charging and discharging strategy.

[0003] At present, the Monte Carlo method has been widely used in the research of electric vehicle charging load prediction, but the path planning model, travel energy consumption model, travel characteristic model and charging decision model used in existing researches still have certain simplification and idealization tendency, and the model accuracy needs to be improved. SUMMARY

[0004] In view of the deficiencies in the prior art, the application provides an electric vehicle charging load prediction method based on travel characteristics, which enhances the reliability of the prediction method by comprehensively considering various travel characteristics, and solves the problem of insufficient prediction accuracy.

[0005] The application achieves the above technical purpose through the following technical means.

[0006] The electric vehicle charging load prediction method based on travel characteristics adopts the Monte Carlo method to simulate the travel and charging behavior of electric vehicles, and obtains the time and space distribution of electric vehicle charging load, wherein:

[0007] Based on the trip chain model, the starting travel time, starting travel location, starting state of charge and travel purpose of each electric vehicle in each time period are extracted, and the functional area that each electric vehicle needs to reach in each time period is extracted according to the travel purpose;

[0008] The travel path of each electric vehicle is planned according to the functional area that needs to be reached;

[0009] The travel energy consumption on the travel path is calculated to update the state of charge of the electric vehicle;

[0010] The charging probability of the electric vehicle is determined according to the state of charge;

[0011] The time and space distribution of electric vehicle charging load is obtained according to the charging probability of each electric vehicle in each region in different time periods.

[0012] Further, the method for path planning is:

[0013] Calculate the road impedance for each road in the transportation network based on the following time-flow model:

[0014]

[0015]

[0016] Where S i,t 、R i,t 、N i,t 、T i,t are the saturation, section impedance, node impedance, and road impedance of the i-th road in period t, respectively. and is the impedance influencing parameter, α is the signal period, β is the vehicle arrival rate on the road section, and λ is the green-signal ratio;

[0017] The Floyd algorithm is used to plan the path with the optimal travel time, where the iterative process is expressed as:

[0018]

[0019] Where, It represents the shortest travel time from network node a to network node b within the t period found after the kth iteration of the algorithm, and is the road impedance T of each road in the found path. i,t the sum of is the shortest travel time found after the k-1th iteration, It represents the shortest travel time from network node a to network node b with node c as the intermediate node.

[0020] Furthermore, the driving energy consumption is calculated based on the following driving energy consumption model:

[0021] E=E1+E2

[0022] E1=∑w i e i

[0023]

[0024] P on =P F (1-ξ·A)

[0025]

[0026] Where E is the total energy consumption of electric vehicles, E1 is the total mileage energy consumption of electric vehicles, and e i is the energy consumption per unit mileage on the i-th road, w i is the length of the i-th road, li is the grade of the ith road, v i ′ is the driving speed of the electric vehicle on the ith road, E2 is the air conditioning energy consumption during the driving of the electric vehicle, ρ ac is the air conditioning power, T ac is the air conditioning on duration, rand is a random number in [0, 1], F out is the outdoor environment temperature, F 22 is the environmental temperature threshold, a4, b4, a5, b5 are coefficients in the air conditioning power formula, P on is the air conditioning opening probability, P F is the probability of the electric vehicle user opening the air conditioner only considering the influence of the environmental temperature, ξ is the anxiety adjustment coefficient, A is the perception degree of the electric vehicle user's range anxiety, SOC is the state of charge of the electric vehicle, SOC L and SOC H are the lower limit and upper limit of the state of charge anxiety perception threshold respectively, δ is the perception sensitivity.

[0027] Further, the trip chain model records:

[0028] the correspondence between each trip purpose and each functional area;

[0029] the trip purpose transfer probability and the starting trip location probability distribution;

[0030] the probability distribution of the starting trip time, parking duration and starting state of charge of the electric vehicle.

[0031] Further, the trip purpose transfer probability is calculated as follows:

[0032]

[0033] wherein, P h→h′,t represents the probability of the electric vehicle transferring from trip purpose h to trip purpose h' at time t, m h→h′,t is the number of electric vehicles transferring from trip purpose h to trip purpose h ′ at time t, and H is the total number of trip purposes.

[0034] Further, the probability density functions of the starting trip time, parking duration and starting state of charge are fitted based on the following particle swarm algorithm:

[0035] In the particle swarm algorithm, the updating formula of the particle position and velocity is:

[0036]

[0037] wherein, and are the velocity and position of the i-th particle at the k+1-th iteration, and are the velocity and position of the i-th particle at the k-th iteration, is the acceleration of the i-th particle at the k-th iteration, is the individual optimal position of the i-th particle at the k-th iteration, p PSO_g is the optimal position of the population, w PSO is the particle’s inertia weight, c PSO1 and c PSO2 is the learning factor, r PSO1 and r PSO2 is a random number distributed in [0,1];

[0038] The acceleration of the particle is calculated as follows:

[0039]

[0040] Where, represents the net force on the i-th particle at the k-th iteration, represents the inertial mass of the i-th particle at the k-th iteration;

[0041] For inertial mass The definition is as follows:

[0042]

[0043] Where, is the unnormalized inertial mass of the i-th particle at the k-th iteration, The i-th iteration is the k-th iteration ′ The unnormalized inertial mass of particles, N PSO is the total number of particles in the population, is the fitness value of the i-th particle at the k-th iteration, and are the minimum fitness value and the maximum fitness value in the kth iteration respectively;

[0044] For the combined force The definition is as follows:

[0045]

[0046] Where rand OSO_i′ represents the random number [0,1] taken for the i′th particle, Indicates particle i at the kth iteration ′ The force on particle i; is the gravitational constant at the kth iteration, The i-th iteration is the k-th iteration ′The inertial mass of the particles, For particle i and particle i ′ The Euclidean distance between PSO is a constant, The i-th iteration is the k-th iteration ′ The position of each particle; is the initial gravitational constant, exp(·) represents the exponential function, ψ PSo is the attenuation coefficient, k is the number of iterations of the algorithm, K PsO is the maximum number of iterations of the particle swarm algorithm;

[0047] In each iteration, the following simulated annealing algorithm is used to update the optimal position of the particle swarm:

[0048]

[0049] Where, and They represent the optimal position p of the population after the kth iteration and the k-1th iteration respectively. PSO_g , is the optimal position of all particles at the kth iteration, rank T is a random number in [0,1], is the probability of accepting the difference solution at the kth iteration, is the fitness value of the optimal position among all particles at the kth iteration, and are the annealing temperatures at the kth iteration and the k-1th iteration, μ T is the temperature drop coefficient, represents the fitness value f of the optimal position of the population after the k-1th iteration PSO_g .

[0050] Furthermore, Gaussian mixture distribution and / or logistic distribution are used to fit the probability density function, where:

[0051] The Gaussian mixture distribution is:

[0052]

[0053] Where a G_g 、b G_g 、c G_g are the peak value, peak position, and half-width information of the g-th Gaussian distribution respectively;

[0054] The logical distribution is:

[0055]

[0056] Where a logic is the upper asymptote parameter, blogic is the shape parameter, c logic is the scale parameter, d logic is the lower asymptote parameter.

[0057] Furthermore, when the SOC is lower than the set rigid charging threshold, the charging probability of the electric vehicle is 1; otherwise, the charging probability is obtained using the following charging decision model:

[0058] Based on fuzzy logic reasoning, the fuzzy result of charging probability is derived according to the fuzzy inputs of average charging cost, state of charge and parking time.

[0059] A Mamdani-type fuzzy controller is used to convert the fuzzy results into charging probability values ​​through the minimum reasoning method, maximum aggregation method and center of gravity defuzzification method.

[0060] Furthermore, in the fuzzy logic reasoning:

[0061] Define four fuzzy sets of "lower", "medium", "higher" and "high" to describe the average charging cost;

[0062] Define four fuzzy sets: "lower", "medium", "higher", and "high" to describe the battery's state of charge;

[0063] Define three fuzzy sets: "short", "medium", and "long" to describe the parking duration;

[0064] Define five fuzzy sets, “low”, “lower”, “medium”, “higher” and “high”, to describe the charging probability;

[0065] The membership relations of the above fuzzy sets all adopt the double Gaussian membership function.

[0066] Furthermore, the Monte Carlo method is used to extract the charging behavior of electric vehicles based on the charging probability to obtain the charging load of electric vehicles; the charging load of each electric vehicle at each location in each time period is simulated separately to obtain the total charging load of electric vehicles in each time period in a region.

[0067] The beneficial effects of the present invention are:

[0068] (1) The present invention provides a method for predicting the charging load of electric vehicles based on travel characteristics, in which a time-flow model based on actual traffic flow and road attributes is used to more accurately reflect the driving time of vehicles under different road conditions at different times, and the shortest driving path is obtained by improving the Floyd algorithm, which is more in line with the actual travel habits of users.

[0069] (2) The electric vehicle driving energy consumption model adopted in the present invention includes mileage energy consumption and air conditioning energy consumption, and the vehicle air conditioning start-up probability model not only considers the influence of ambient temperature, but also takes into account the charge state and user's range anxiety.

[0070] (3) The present invention adopts an improved particle swarm optimization algorithm based on gravitational search and simulated annealing to accurately fit the probability density function of travel characteristics, which has a better fitting effect and can more accurately reflect the travel characteristics of electric vehicle users.

[0071] (4) The present invention comprehensively considers factors such as the state of charge of electric vehicles, average charging cost and parking time. The user charging decision model constructed based on the fuzzy logic reasoning system can more accurately simulate the user's flexible charging behavior and is more in line with actual conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 It is a structural block diagram of the prediction method of the present invention;

[0073] Figure 2 This is a diagram showing the traffic network topology and functional area division results of this embodiment;

[0074] Figure 3 This is a topological diagram of the coupling relationship between the transportation network and the distribution network in this embodiment;

[0075] Figure 4 This is a schematic diagram of the travel chain structure of this embodiment;

[0076] Figure 5 is the travel state transition probability matrix for different time periods on weekdays in this embodiment;

[0077] Figure 6 is the travel state transition probability matrix for different time periods on weekends in this embodiment;

[0078] Figure 7 This is a flow chart of the improved particle swarm optimization algorithm based on gravitational search and simulated annealing of the present invention;

[0079] Figure 8 This is the fitting result of the starting travel time of this embodiment;

[0080] Figure 9 is the initial state of charge fitting result of this embodiment;

[0081] Figure 10 This is the parking duration fitting result for different travel purposes in this embodiment;

[0082] Figure 11 is the membership relationship between the average charging cost of this embodiment and its fuzzy set;

[0083] Figure 12is the membership relationship between the state of charge and its fuzzy set in this embodiment;

[0084] Figure 13 is the membership relationship between the parking duration and its fuzzy set in this embodiment;

[0085] Figure 14 is the membership relationship between the charging probability of this embodiment and its fuzzy set;

[0086] Figure 15 is the fuzzy surface of the elastic charging probability of the electric vehicle in this embodiment;

[0087] Figure 16 This is the simulation result of the time distribution of the electric vehicle charging load in this embodiment;

[0088] Figure 17 This is the simulation result of the spatiotemporal distribution of electric vehicle charging load on weekends in this embodiment;

[0089] Figure 18 This is the simulation result of the spatiotemporal distribution of electric vehicle charging load on weekdays in this embodiment;

[0090] Figure 19 The simulation results of the spatiotemporal distribution of charging loads at the distribution network nodes on weekdays in this embodiment are as follows;

[0091] Figure 20 This is the simulation result of the spatiotemporal distribution of charging load at the distribution network nodes on weekends in this embodiment. DETAILED DESCRIPTION

[0092] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0093] 1. Plan

[0094] Reference Figure 1 As shown:

[0095] 1. Establish a transportation network-distribution network coupling framework

[0096] (1) Graph theory is used to extract information such as road length, actual traffic flow, road grade, maximum capacity, and free flow speed from the traffic network, and a road network model based on multi-source information fusion is constructed, as shown in the following formula:

[0097]

[0098] Where G tis the topology information of the traffic road network in the t period, N is a set of road network nodes, N is the total number of road network nodes, B is a set of roads (connecting between nodes), B represents the total number of roads, W is a set of lengths of roads, w i is the length of the i-th road, L is a set of grades of roads, l i is the grade of the i-th road, l i , 2, 3 respectively represent the city trunk road, the city secondary trunk road and the city branch road, V is a set of free flow speeds of roads, v i is the highest driving speed when the traffic flow density of the i-th road tends to zero, C is a set of maximum traffic capacities of roads, c i is the maximum traffic flow of the i-th road, T is a set of divided time periods, that is, a day is divided into T time periods; Q t is a set of actual traffic flows of roads in the t period, q i,t is the actual traffic flow of the i-th road in the t period.

[0099] (2) According to the city area attribute and the functional positioning, the functional area of the research area is divided into 8 types, including residential area, office area, industrial area, hospital, station, school, scenic area and commercial area, and the traffic road network topology structure and the functional area division result of the embodiment are shown in the attached Figure 2

[0100] (3) Refer to the graph theory method to model the distribution network, as shown in the following formula:

[0101]

[0102] In the formula, D is the topology information of the distribution network, N D is a set of distribution network nodes, N D is the total number of distribution network nodes, B D is a set of distribution network branches (connecting between nodes), B D is the total number of distribution network branches.

[0103] (4) Based on the above (1) (2) (3), the traffic road network-distribution network coupling framework is constructed, wherein the traffic road network and the distribution network coupling relationship topology diagram of the embodiment is shown in the attached Figure 3 , and the geographical coupling relationship between the distribution network nodes and the functional areas is given in the following table 1:

[0104] Table 1: Comparison table of distribution network nodes and functional area numbers

[0105]

[0106]

[0107] ​2. Build a time-traffic model and plan routes based on optimal time.

[0108] (1) The time-flow model includes the road segment impedance caused by factors such as traffic volume and road attributes, as well as the node impedance caused by traffic signal control, as follows:

[0109]

[0110] T i,t =R i,t +N i,t

[0111] Where S i,t 、R i,t 、N i,t 、T i,t are the saturation, section impedance, node impedance, and road impedance (average travel time) of the i-th road in period t, and is the impedance influencing parameter, which takes values ​​of 0.15 and 4 respectively in this embodiment, α is the signal period, β is the vehicle arrival rate of the road section, and λ is the green-signal ratio. In this embodiment, α, β, and λ take values ​​of 30, 0.8, and 0.7 respectively.

[0112] (2) Combined with the actual traffic network data, the road impedance T of each road in the road network is calculated based on the above time-flow model. i,t ; Then, based on this, the Floyd algorithm is used to plan the path with the optimal travel time.

[0113] The iterative process in Floyd's algorithm is expressed as:

[0114]

[0115] Where, It represents the shortest travel time from network node a to network node b within the t period found after the kth iteration of the algorithm, that is, the road impedance T of each road in the found path. i,t The same applies is the shortest travel time found after the k-1th iteration, It represents the shortest travel time from network node a to network node b with node c as the intermediate node.

[0116] (3) To verify the effectiveness of the proposed method for selecting the optimal time path using the time-traffic model as the road resistance function, the following simulation scenario was set: a vehicle departs from node 13 at 9:00 a.m. on a weekday and travels to node 17. The proposed method was compared with the traditional Floyd shortest path algorithm (Method A) and the method for selecting the optimal time path using the BPR model as the road resistance function (Method B). The simulation results are shown in Table 2 below.

[0117] Table 2: Path planning comparison

[0118]

[0119] The comparison results show that the path selected by Method A is slightly better than that of the present invention in terms of mileage, but its total driving time and corresponding mileage energy consumption are significantly higher than those of the present invention. The path selected by Method B is consistent with that of the present invention, but the driving time calculated by it is shorter than that of the present invention. This difference is mainly due to the fact that the BPR model used by Method B does not fully consider the traffic delay factors in the multi-intersection environment of urban roads during the modeling process. This model is more suitable for highway sections, and does not take into account delays such as frequent intersection signal control and vehicle queues in complex urban road networks, resulting in deviations in the results.

[0120] In summary, the present invention can more accurately reflect the actual traffic conditions of urban roads while ensuring the rationality of the path, and has significant advantages in improving path selection accuracy and driving efficiency.

[0121] 3. Build an electric vehicle driving energy consumption model that takes into account mileage energy consumption and air conditioning energy consumption

[0122] (1) Considering the driving speed of electric vehicles, the mileage energy consumption model is constructed as follows:

[0123] E1=∑w i e i

[0124]

[0125] Where E1 is the total mileage energy consumption of the electric vehicle, and its expression means the sum of the energy consumption of each road in the driving path, e i is the energy consumption per unit mileage on the i-th road, v i ′ is the speed of the electric vehicle on the i-th road (the average speed is taken by default); a1, b1, c1, a2, b2, c2, a3, b3, and c3 are coefficients, which can be obtained through experiments. In this embodiment, the values ​​are -0.179, 0.004, 5.492, 0.21, -0.001, 1.531, 0.208, -0.002, and 1.553, respectively.

[0126] (2) Considering the ambient temperature, the state of charge of the electric vehicle and the user's range anxiety, a probability model for turning on the vehicle air conditioner is constructed.

[0127] 1) Introducing the Weber-Fechner law to construct an electric vehicle user range anxiety perception model:

[0128]

[0129] Where A is the range anxiety perception of electric vehicle users, SOC is the state of charge (SOC) of the electric vehicle, and SOC L and SOC H are the lower limit and upper limit of the charge state anxiety perception threshold, respectively. In this embodiment, they are set to 0.1 and 0.6 respectively. δ is the perception sensitivity, which is 1.28 in this embodiment.

[0130] 2) The anxiety adjustment coefficient ξ is introduced to describe the influence of the user's range anxiety perception on the probability of starting the car air conditioner, and the probability of the user turning on the car air conditioner P is obtained by comprehensively considering the ambient temperature, battery state of charge and user's range anxiety. on :

[0131] P on =P F (1-ξ·A)

[0132] Where, P F is the probability of electric vehicle users turning on the air conditioner when only considering the influence of ambient temperature. F The data can be obtained through big data statistics.

[0133] 3) Construct an air conditioning energy consumption model that takes ambient temperature into account:

[0134]

[0135] Where E2 is the air conditioning energy consumption during the driving process of the electric vehicle, ρ ac is the air conditioning power, T ac is the air conditioner on time, rand is a random number in [0,1], F out is the ambient temperature outside the vehicle, F 22 is the ambient temperature threshold, which is set to 22°C in this embodiment. a4, b4, a5, and b5 are coefficients in the air conditioning power formula, which can be obtained through experiments. In this embodiment, the values ​​are 33.47, 1.324, 36.69, and 1.084 respectively.

[0136] (3) In summary, the electric vehicle driving energy consumption model taking into account mileage energy consumption and air conditioning energy consumption is:

[0137] E=E1+E2

[0138] Where E is the total energy consumption of electric vehicles.

[0139] 4. Construct a multi-dimensional coupled travel chain model driven by travel purpose

[0140] (1) A multi-dimensional coupled travel chain model driven by travel purpose is constructed to simulate the travel process of electric vehicle users in one day. The model constructs a three-layer representation system including travel purpose layer, time chain layer and space chain layer. Among them, the time chain includes time characteristics such as starting travel time, driving time, arrival time, parking time and departure time, and the space chain includes space characteristics such as starting travel location, travel destination and driving mileage. The schematic diagram of the travel chain structure is shown in the attached figure. Figure 4 shown.

[0141] The users' travel purposes are divided into the following eight categories: family activities (FA), work (W), shopping (S), dining (D), leisure and entertainment (E), educational activities (EA), medical services (MC), and transportation (PU). The travel purposes are then mapped to urban functional areas, as shown in Table 3:

[0142] Table 3: Comparison of travel purposes and accessible functional areas

[0143]

[0144] (2) Considering two different scenarios, weekdays and weekends, the probability distribution of travel destination transfer and the probability distribution of starting travel location are obtained by statistical means respectively; where:

[0145] 1) Electric vehicle users transfer from travel purpose h to travel purpose h during period t ′ The transition probability is:

[0146]

[0147] Where, P h→h′,t represents the probability that an electric vehicle user transfers from travel destination h to travel destination h′ during period t, m h→h′,t The electric vehicle user transfers from travel purpose h to travel purpose h in period t ′ The data is obtained through sampling statistics, H is the total number of established travel purposes, and in this embodiment, a total of 8 travel purposes are established, so H=8.

[0148] The probability of transfer from travel destination P h→h′,t The constructed travel destination transfer probability matrix can be expressed as:

[0149]

[0150] Where h or h ′ The travel purposes represented by values ​​from 1 to 8 are: 1 family activities (FA), 2 work (W), 3 shopping (S), 4 medical services (MC), 5 pick-up and drop-off (PU), 6 leisure and entertainment (E), 7 dining (D), and 8 educational activities (EA).

[0151] In this embodiment, a day is divided into 6 periods on average, and the travel state transition probability matrix of urban vehicles under two scenarios, weekdays and weekends, is obtained by statistical calculation, as shown in FIG. Figure 5 and Figure 6 shown.

[0152] 2) The probability distribution of electric vehicle users’ starting travel locations in each functional area under the two scenarios of weekdays and weekends obtained from statistics in this example is shown in Table 4 below.

[0153] Table 4: Probability distribution of starting travel locations

[0154]

[0155] (3) Based on the sampled data, the probability density of the starting travel time, parking time, and starting state of charge of electric vehicle users is fitted.

[0156] 1) This embodiment adopts the following improved method based on the particle swarm algorithm (named "Improved Particle Swarm Optimization Algorithm Based on Gravitational Search and Simulated Annealing") to fit the probability density function of the starting travel time, parking time, and starting state of charge. In this embodiment, the mass, gravitational constant, and net force of each particle are calculated to determine the acceleration, which is used to update the particle's velocity and position. At the same time, the simulated annealing algorithm is used to update the optimal position of the particle swarm population and accept the difference solution with a certain probability, thereby enhancing the algorithm's sudden jump capability. Among them:

[0157] In the particle swarm optimization algorithm, the update formula for particle position and velocity is:

[0158]

[0159] Where, and are the velocity and position of the i-th particle at the k+1-th iteration, and are the velocity and position of the i-th particle at the k-th iteration, is the acceleration of the i-th particle at the k-th iteration, is the individual optimal position of the i-th particle at the k-th iteration, p PSO_g is the optimal position of the population, w PsOis the particle inertia weight, which is 0.65 in this embodiment, c PSO1 and c PSO2 is the learning factor, which is set to 0.95 in this embodiment, r PSO1 and r PSO2 is a random number distributed in [0,1].

[0160] 1.1) For the acceleration of a particle, the calculation formula is:

[0161]

[0162] Where, represents the net force on the i-th particle at the k-th iteration, represents the inertial mass of the i-th particle at the k-th iteration;

[0163] 1.1.1) For inertial mass In this embodiment, the fitness value of the particle is defined as follows:

[0164]

[0165] Where, is the unnormalized inertial mass of the i-th particle at the k-th iteration. Similarly, The i-th iteration is the k-th iteration ′ The unnormalized inertial mass of particles, N PSO is the total number of particles in the population, is the fitness value of the i-th particle at the k-th iteration, and are the minimum fitness value and the maximum fitness value in the kth iteration respectively.

[0166] 1.1.2) For the resultant force In this embodiment, it is defined as:

[0167]

[0168] Where rand PSO_i′ represents the random number [0,1] taken for the i′th particle, Indicates particle i at the kth iteration ′ The force acting on particle i in space;

[0169]

[0170] Where, is the gravitational constant at the kth iteration, The i-th iteration is the k-th iteration ′ The inertial mass of the particles, For particle i and particle i′ The Euclidean distance between PSO is a constant, and in this embodiment, it is set to 1×10 -6 , The i-th iteration is the k-th iteration ′ The position of each particle;

[0171]

[0172] Where, is the initial gravitational constant, which is 100 in this embodiment, exp(·) represents the exponential function, ψ PSO is the attenuation coefficient, which is 20 in this embodiment, and k is the number of iterations of the algorithm. PSO is the maximum number of iterations of the particle swarm optimization algorithm.

[0173] 1.2) The particle swarm algorithm has the disadvantage of being too premature. Therefore, in each iteration, the present invention uses a simulated annealing algorithm to update the optimal position of the particle swarm population to enhance the algorithm's sudden jump capability:

[0174] 1.2.1) Determine the annealing temperature:

[0175]

[0176] Where, and are the annealing temperatures at the kth iteration and the k-1th iteration, μ T is the temperature reduction coefficient, which is 0.95 in this embodiment. represents the fitness value f of the optimal position of the population after the k-1th iteration PSO_g ;

[0177] 1.2.2) Calculate the probability of accepting the difference solution based on the annealing temperature:

[0178]

[0179] Where, is the probability of accepting the difference solution at the kth iteration, is the fitness value of the optimal position among all particles at the kth iteration;

[0180] 1.2.3) Update the optimal position of the population according to the Metropolis criterion:

[0181]

[0182] Where, and They represent the optimal position p of the population after the kth iteration and the k-1th iteration respectively. PSO_g , is the optimal position of all particles at the kth iteration, rand T A random number in [0,1].

[0183] In summary, refer to Figure 7 As shown in Figure 2, the processing flow of the improved particle swarm optimization algorithm based on gravitational search and simulated annealing is as follows:

[0184] Step 1: Use chaotic mapping to initialize the position of the population and assign values ​​to each parameter;

[0185] Step 2: Calculate the fitness value of each particle and save the individual optimal position of the particle and the optimal position of the population;

[0186] Step 3: Calculate the maximum fitness value and minimum fitness value of the particle;

[0187] Step 4: Calculate the mass, gravitational constant, net force, and acceleration of each particle.

[0188] Step 5: Update the particle velocity and check if the velocity exceeds the constraint limit. If so, correct the particle velocity.

[0189] Step 6: Update the position of the particles and correct the position of the particles that exceed the constraint limit;

[0190] Step 7: Calculate the fitness value of the particle, save the individual optimal position of each particle and the optimal position of all particles;

[0191] Step 8: Use simulated annealing algorithm to update the optimal position of the population;

[0192] Step 9: When the algorithm reaches the maximum number of iterations, it outputs the current optimal result, that is, the parameters in the fitted probability density function.

[0193] 2) Based on the sampled electric vehicle user travel data, in this embodiment, the parking duration for the two travel purposes of "shopping" and "dining" is fitted with a probability density function using a logistic distribution, and the rest are fitted using a Gaussian mixture distribution.

[0194] in:

[0195] The general expression for the probability density function of a Gaussian mixture distribution is:

[0196]

[0197] Where, f g (x|θ g ) is the probability density function of the g-th Gaussian distribution in the Gaussian mixture distribution, x represents the variable of the probability density function, which is applied in the present invention, that is, the time period t of the day, θ gRepresents the parameters of the g-th Gaussian distribution, including the mean μ g and standard deviation σ g ;P G (x|Θ) is the probability density function of the Gaussian mixture distribution, Θ is the parameter of the Gaussian mixture distribution, including the weight ω of each Gaussian distribution g and parameter θ g , N G is the number of individual Gaussian distributions in the Gaussian mixture distribution, ω g is the weight of the g-th Gaussian distribution, and satisfies

[0198] In order to facilitate the observation of the characteristics of the fitting results, the above Gaussian mixture distribution is simplified to:

[0199]

[0200] Where a G_g 、b G_g 、c G_g are the peak value, peak position, and half-width information of the g-th Gaussian distribution respectively.

[0201] The probability density function of the four-parameter form of the logistic distribution is:

[0202]

[0203] Where a oogic is the upper asymptote parameter, b logic is the shape parameter, c logic is the scale parameter, d logic is the lower asymptote parameter.

[0204] 3) Introducing the coefficient of determination R square To judge the quality of the final fitting results, the expression of the determination coefficient is:

[0205]

[0206] Where N S is the sample size, E SSR is the original data mean Fitting the model to the data The sum of the squares of the differences, E SST is the original data y i The average value of the original data The sum of the squares of the differences; That is, the probability density function (P G (x) or P logic (x)), where the subscript i represents the data of the i-th sample.

[0207] 4) This embodiment uses the above method to fit the Gaussian mixture distribution function of the starting travel time on weekdays and weekends. The fitting parameter values ​​are shown in Table 5. The fitting results are shown in Table 5. Figure 8 shown.

[0208] In order to compare the fitting effect of the present invention horizontally, the automatic fitting method of the MATLAB platform was used as a control. Figure 8 The green curve is obtained by automatic fitting using MATLAB, and the orange curve is obtained by the fitting method of the present invention. square is 0.97984, and the error square sum is 2057237.6821. square The sum of squared errors is 0.9957 and 436549.1628 respectively; in the fitting of the “weekend start travel time”, the R square The sum of squared errors were 0.98737 and 105913.6265 respectively. square The sum of square errors is 0.9958 and 35201.0023 respectively; the fitting performance data of the present invention are better than the MATLAB automatic fitting.

[0209] Table 5: Probability density fitting parameters of Gaussian mixture distribution of start travel time

[0210]

[0211] 5) The initial state of charge fitting results are as follows Figure 9 The fitting parameter values ​​are shown in Table 6.

[0212] Table 6: Initial state of charge Gaussian mixture distribution probability density fitting parameters

[0213]

[0214] 6) In this embodiment, the parking duration for activities such as family activities, work, leisure and entertainment, transportation, educational activities, and medical services follows a Gaussian mixture distribution, while the parking duration for shopping and dining follows a logistic distribution. The parking duration fitting results for different travel purposes are shown in the figure below. Figure 10 The specific fitting parameter values ​​are shown in Table 7.

[0215] Table 7: Parking duration probability density fitting parameters

[0216]

[0217]

[0218] R squareThe closer the value of is to 1, the better the fitting effect is. Using the fitting method of the present invention, the probability density function fitting results of the starting travel time, starting state of charge, and parking time are obtained. square The values ​​all exceeded 0.99.

[0219] 5. Constructing an electric vehicle user charging decision model based on fuzzy logic reasoning

[0220] (1) The input variables of the charging decision model include average charging cost, state of charge of electric vehicles, and parking time; the output of the model is the charging probability of electric vehicle users.

[0221] 1) Define four fuzzy sets of “lower”, “medium”, “higher” and “high” to characterize the perceptual attitude of electric vehicle users towards the average charging cost; use a double Gaussian membership function to reflect the membership relationship between the above fuzzy sets and the average charging cost. The membership function curve of the average charging cost in this embodiment is shown in the attached figure. Figure 11 As shown;

[0222] In this embodiment, the average charging cost is: total charging cost / charged amount, that is, the cost per unit charging amount (yuan / kWh).

[0223] 2) Define four fuzzy sets of “lower”, “medium”, “higher” and “high” to describe the state of charge of the battery; use a double Gaussian membership function to reflect the membership relationship between the above fuzzy sets and the state of charge. The membership function curve of the state of charge in this embodiment is shown in the attached figure. Figure 12 As shown;

[0224] 3) Define three fuzzy sets of "short", "medium" and "long" to describe parking duration; use a double Gaussian membership function to reflect the membership relationship between the above fuzzy sets and parking duration; however, the membership function of parking duration is greatly affected by the charging power. Therefore, membership functions are set separately for fast charging and slow charging. The membership function of parking duration in this embodiment is shown in the attached figure. Figure 13 As shown;

[0225] 4) Define five fuzzy sets, namely “low”, “lower”, “medium”, “higher” and “high”, to characterize the possibility of users charging behavior (charging probability); use double Gaussian membership function to reflect the membership relationship between the above fuzzy sets and charging probability. The membership function of charging probability is shown in the attached figure. Figure 14 shown.

[0226] (2) The fuzzy logic reasoning rules of the user charging decision model are shown in Table 8 below. The logical relationship between the input variables is "and":

[0227] Table 8: Fuzzy logic reasoning rules table

[0228]

[0229]

[0230] (3) Using the Mamdani type fuzzy controller, the fuzzy result is converted into a specific user charging probability through the minimum inference method, the maximum aggregation method and the center of gravity defuzzification method. In this embodiment, the fuzzy surface of the electric vehicle user elastic charging probability under the two scenarios of short-term parking and long-term parking is shown in the attached figure. Figure 15 shown.

[0231] 6. Use the Monte Carlo method to simulate users’ travel and charging behaviors and simulate the spatiotemporal distribution of electric vehicle charging load in the region. The specific steps are as follows:

[0232] Step 1: Enter vehicle (including the home and work addresses of each electric vehicle user), road network and distribution network data, set the total number of Monte Carlo cycles and the total number of electric vehicles in the area;

[0233] Step 2: extract the starting travel time, starting travel location and starting state of charge of the electric vehicle;

[0234] Step 3: Extract the purpose of travel based on the transition probability matrix; Based on the time-flow model and the actual traffic flow of each road, calculate the travel time of the vehicle on each road section, and use the Floyd algorithm to plan the route with the shortest time, thereby obtaining the travel time of each electric vehicle on the planned route;

[0235] Step 4: Calculate the total energy consumption of each electric vehicle based on its driving data (driving path, driving speed, driving duration) and update the state of charge.

[0236] Step 5: Determine the user's charging demand based on the state of charge (SOC). When the SOC is less than 0.2, it is a rigid charging demand, while if it is less than 0.2, it is a flexible charging demand. For rigid charging demand, the charging probability of the electric vehicle user is 1. For flexible charging demand, the user's charging probability is obtained using the user charging decision model based on the state of charge, average charging cost, and parking duration.

[0237] Step 6: Based on the charging probability, extract the charging behavior of the electric vehicle (whether to charge), and obtain the charging load of the corresponding electric vehicle;

[0238] Step 7: Repeat steps 2 to 6 to simulate the charging load of each electric vehicle at each time period and each location during the day. Sum the charging load of each electric vehicle at each time period to obtain the total charging load of electric vehicles at each time period in a region (or road network node).

[0239] The above steps 2 to 7 constitute a Monte Carlo cycle. The above Monte Carlo cycle process is repeated until the set total number of cycles is reached, and the charging load of each functional area and the total regional charging load are output.

[0240] This example simulates the spatiotemporal distribution of electric vehicle charging load in the study area. Figure 16 The simulation results of the time distribution of electric vehicle charging load in the entire area are as follows: Figure 17 and Figure 18 The temporal and spatial distribution of charging load for each functional area, Figure 19 and Figure 20 is the spatiotemporal distribution of charging load at each node in the distribution network.

[0241] The present invention is not limited to the above-mentioned embodiments. Any obvious improvement, replacement or modification that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A method for predicting electric vehicle charging load based on travel characteristics, characterized by: The Monte Carlo method is used to simulate the travel and charging behavior of electric vehicles to obtain the spatiotemporal distribution of electric vehicle charging load, where: Based on the travel chain model, the starting travel time, starting travel location, starting state of charge, and travel purpose of each electric vehicle in each time period are extracted, and the functional areas that each electric vehicle needs to reach in each time period are extracted according to the travel purpose; Plan the driving paths of each electric vehicle according to the functional areas that need to be reached; Calculating the driving energy consumption on the driving route to update the state of charge of the electric vehicle; Determine the charging probability of electric vehicles based on the state of charge; According to the charging probability of each electric vehicle in each area at different time periods, the spatiotemporal distribution of electric vehicle charging load is obtained.

2. The electric vehicle charging load prediction method based on travel characteristics according to claim 1 is characterized in that: The path planning method is: Calculate the road impedance for each road in the transportation network based on the following time-flow model: T i,t =R i,t +N i,t Where S i,t 、R i,t 、N i,t 、T i,t are the saturation, section impedance, node impedance, and road impedance of the i-th road in period t, respectively. and is the impedance influencing parameter, α is the signal period, β is the vehicle arrival rate on the road section, and λ is the green-signal ratio; The Floyd algorithm is used to plan the path with the optimal travel time, where the iterative process is expressed as: Where, It represents the shortest travel time from network node a to network node b within the t period found after the kth iteration of the algorithm, and is the road impedance T of each road in the found path. i,t the sum of is the shortest travel time found after the k-1th iteration, It represents the shortest travel time from network node a to network node b with node c as the intermediate node.

3. The electric vehicle charging load prediction method based on travel characteristics according to claim 1 is characterized in that: Driving energy consumption is calculated based on the following driving energy consumption model: E=E1+E2 E1=∑w i have been i P on =P G (1-ξ·A) Where E is the total energy consumption of electric vehicles, E1 is the total mileage energy consumption of electric vehicles, and e i is the energy consumption per unit mileage on the i-th road, w i is the length of the i-th road, l i is the grade of the i-th road, v i ′ is the speed of the electric vehicle on the i-th road, E2 is the air conditioning energy consumption of the electric vehicle during driving, ρ ac is the air conditioning power, T ac is the air conditioner on time, rand is a random number in [0,1], F out is the ambient temperature outside the vehicle, F 22 is the ambient temperature threshold, a4, b4, a5, b5 are the coefficients in the air conditioning power formula, P on is the probability of air conditioning turning on, P F is the probability that electric vehicle users turn on the air conditioner when only considering the influence of ambient temperature, ξ is the anxiety adjustment coefficient, A is the range anxiety perception of electric vehicle users, SOC is the state of charge of the electric vehicle, SOC L and SOC H are the lower and upper limits of the anxiety perception threshold of the state of charge, and δ is the perception sensitivity.

4. The electric vehicle charging load prediction method based on travel characteristics according to claim 1 is characterized in that: The travel chain model records: The correspondence between each travel purpose and each functional area; The probability distribution of travel destination transfer and starting travel location; Probability distribution of electric vehicle starting travel time, parking time, and initial state of charge.

5. The electric vehicle charging load prediction method based on travel characteristics according to claim 4 is characterized in that: The travel destination transfer probability is calculated as follows: Where, P h→h′,t Indicates that the electric vehicle transfers from travel destination h to travel destination h during period t ′ The probability of m h→h′,t The electric vehicle transfers from travel destination h to travel destination h during period t ′ , H is the total number of travel purposes.

6. The electric vehicle charging load prediction method based on travel characteristics according to claim 4 is characterized in that: The probability density functions of the starting travel time, parking time, and starting state of charge are fitted based on the following particle swarm algorithm: In the particle swarm optimization algorithm, the update formula for particle position and velocity is: Where, and are the velocity and position of the i-th particle at the k+1-th iteration, and are the velocity and position of the i-th particle at the k-th iteration, is the acceleration of the i-th particle at the k-th iteration, is the individual optimal position of the i-th particle at the k-th iteration, p PSO_g is the optimal position of the population, w PSO is the particle’s inertia weight, c PSO1 and c PSO2 is the learning factor, r PSO1 and r PSO2 is a random number distributed in [0,1]; The acceleration of the particle is calculated as follows: Where, represents the net force on the i-th particle at the k-th iteration, represents the inertial mass of the i-th particle at the k-th iteration; For inertial mass The definition is as follows: Where, is the unnormalized inertial mass of the i-th particle at the k-th iteration, The i-th iteration is the k-th iteration ′ The unnormalized inertial mass of particles, N PSO is the total number of particles in the population, is the fitness value of the i-th particle at the k-th iteration, and are the minimum fitness value and the maximum fitness value in the kth iteration respectively; For the combined force The definition is as follows: Where rand PSO_i′ Indicates that for the i ′ The random number [0,1] taken by each particle, Indicates particle i at the kth iteration ′ The force on particle i; is the gravitational constant at the kth iteration, The i-th iteration is the k-th iteration ′ The inertial mass of the particles, For particle i and particle i ′ The Euclidean distance between PSO is a constant, The i-th iteration is the k-th iteration ′ The position of each particle; is the initial gravitational constant, exp(·) represents the exponential function, ψ PSo is the attenuation coefficient, k is the number of iterations of the algorithm, K PsO is the maximum number of iterations of the particle swarm algorithm; In each iteration, the following simulated annealing algorithm is used to update the optimal position of the particle swarm: Where, and They represent the optimal position p of the population after the kth iteration and the k-1th iteration respectively. PSO_g , is the optimal position of all particles at the kth iteration, rand T is a random number in [0,1], is the probability of accepting the difference solution at the kth iteration, is the fitness value of the optimal position among all particles at the kth iteration, and are the annealing temperatures at the kth iteration and the k-1th iteration, μ T is the temperature drop coefficient, represents the fitness value f of the optimal position of the population after the k-1th iteration PSO_g .

7. The electric vehicle charging load prediction method based on travel characteristics according to claim 6 is characterized in that: The probability density function is fitted using Gaussian mixture distribution and / or logistic distribution, where: The Gaussian mixture distribution is: Where a G_g 、b G_g 、c G_g are the peak value, peak position, and half-width information of the g-th Gaussian distribution respectively; The logical distribution is: Where a logic is the upper asymptote parameter, b logic is the shape parameter, c logic is the scale parameter, d logic is the lower asymptote parameter.

8. The electric vehicle charging load prediction method based on travel characteristics according to claim 1 is characterized in that: When the SOC is lower than the set rigid charging threshold, the charging probability of the electric vehicle is 1; otherwise, the charging probability is obtained using the following charging decision model: Based on fuzzy logic reasoning, the fuzzy result of charging probability is derived according to the fuzzy inputs of average charging cost, state of charge and parking time. A Mamdani-type fuzzy controller is used to convert the fuzzy results into charging probability values ​​through the minimum reasoning method, maximum aggregation method and center of gravity defuzzification method.

9. The method for predicting electric vehicle charging load based on travel characteristics according to claim 8, characterized in that: In the fuzzy logic reasoning: Define four fuzzy sets: "lower", "medium", "higher", and "high" to describe the average charging cost; Define four fuzzy sets: "low", "medium", "high", and "high" to describe the battery's state of charge; Define three fuzzy sets: "short", "medium", and "long" to describe the parking duration; Define five fuzzy sets, "low", "lower", "medium", "higher", and "high", to describe the charging probability; The membership relations of the above fuzzy sets all adopt the double Gaussian membership function.

10. The electric vehicle charging load prediction method based on travel characteristics according to claim 1 is characterized in that: The Monte Carlo method is used to extract the charging behavior of electric vehicles based on the charging probability to obtain the charging load of electric vehicles; the charging load of each electric vehicle at each location in each time period is simulated separately to obtain the total charging load of electric vehicles in each time period in a region.

Images